git.oblomov.eu Git - linux-2.6/blob - ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/slab.h>
  20 #include <linux/mm.h>
  21 #include <linux/hugetlb.h>
  22 #include <linux/shm.h>
  23 #include <linux/init.h>
  24 #include <linux/file.h>
  25 #include <linux/mman.h>
  26 #include <linux/shmem_fs.h>
  27 #include <linux/security.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/seq_file.h>
  32
  33 #include <asm/uaccess.h>
  34
  35 #include "util.h"
  36
  37 #define shm_flags       shm_perm.mode
  38
  39 static struct file_operations shm_file_operations;
  40 static struct vm_operations_struct shm_vm_ops;
  41
  42 static struct ipc_ids shm_ids;
  43
  44 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  45 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  46 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  47 #define shm_buildid(id, seq) \
  48         ipc_buildid(&shm_ids, id, seq)
  49
  50 static int newseg (key_t key, int shmflg, size_t size);
  51 static void shm_open (struct vm_area_struct *shmd);
  52 static void shm_close (struct vm_area_struct *shmd);
  53 #ifdef CONFIG_PROC_FS
  54 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  55 #endif
  56
  57 size_t  shm_ctlmax = SHMMAX;
  58 size_t  shm_ctlall = SHMALL;
  59 int     shm_ctlmni = SHMMNI;
  60
  61 static int shm_tot; /* total number of shared memory pages */
  62
  63 void __init shm_init (void)
  64 {
  65         ipc_init_ids(&shm_ids, 1);
  66         ipc_init_proc_interface("sysvipc/shm",
  67                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
  68                                 &shm_ids,
  69                                 sysvipc_shm_proc_show);
  70 }
  71
  72 static inline int shm_checkid(struct shmid_kernel *s, int id)
  73 {
  74         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  75                 return -EIDRM;
  76         return 0;
  77 }
  78
  79 static inline struct shmid_kernel *shm_rmid(int id)
  80 {
  81         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  82 }
  83
  84 static inline int shm_addid(struct shmid_kernel *shp)
  85 {
  86         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  87 }
  88
  89
  90
  91 static inline void shm_inc (int id) {
  92         struct shmid_kernel *shp;
  93
  94         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         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 119         shm_rmid (shp->id);
 120         shm_unlock(shp);
 121         if (!is_file_hugepages(shp->shm_file))
 122                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 123         else
 124                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 125                                                 shp->mlock_user);
 126         fput (shp->shm_file);
 127         security_shm_free(shp);
 128         ipc_rcu_putref(shp);
 129 }
 130
 131 /*
 132  * remove the attach descriptor shmd.
 133  * free memory for segment if it is marked destroyed.
 134  * The descriptor has already been removed from the current->mm->mmap list
 135  * and will later be kfree()d.
 136  */
 137 static void shm_close (struct vm_area_struct *shmd)
 138 {
 139         struct file * file = shmd->vm_file;
 140         int id = file->f_dentry->d_inode->i_ino;
 141         struct shmid_kernel *shp;
 142
 143         down (&shm_ids.sem);
 144         /* remove from the list of attaches of the shm segment */
 145         if(!(shp = shm_lock(id)))
 146                 BUG();
 147         shp->shm_lprid = current->tgid;
 148         shp->shm_dtim = get_seconds();
 149         shp->shm_nattch--;
 150         if(shp->shm_nattch == 0 &&
 151            shp->shm_flags & SHM_DEST)
 152                 shm_destroy (shp);
 153         else
 154                 shm_unlock(shp);
 155         up (&shm_ids.sem);
 156 }
 157
 158 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 159 {
 160         file_accessed(file);
 161         vma->vm_ops = &shm_vm_ops;
 162         shm_inc(file->f_dentry->d_inode->i_ino);
 163         return 0;
 164 }
 165
 166 static struct file_operations shm_file_operations = {
 167         .mmap   = shm_mmap
 168 };
 169
 170 static struct vm_operations_struct shm_vm_ops = {
 171         .open   = shm_open,     /* callback for a new vm-area open */
 172         .close  = shm_close,    /* callback for when the vm-area is released */
 173         .nopage = shmem_nopage,
 174 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 175         .set_policy = shmem_set_policy,
 176         .get_policy = shmem_get_policy,
 177 #endif
 178 };
 179
 180 static int newseg (key_t key, int shmflg, size_t size)
 181 {
 182         int error;
 183         struct shmid_kernel *shp;
 184         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 185         struct file * file;
 186         char name[13];
 187         int id;
 188
 189         if (size < SHMMIN || size > shm_ctlmax)
 190                 return -EINVAL;
 191
 192         if (shm_tot + numpages >= shm_ctlall)
 193                 return -ENOSPC;
 194
 195         shp = ipc_rcu_alloc(sizeof(*shp));
 196         if (!shp)
 197                 return -ENOMEM;
 198
 199         shp->shm_perm.key = key;
 200         shp->shm_flags = (shmflg & S_IRWXUGO);
 201         shp->mlock_user = NULL;
 202
 203         shp->shm_perm.security = NULL;
 204         error = security_shm_alloc(shp);
 205         if (error) {
 206                 ipc_rcu_putref(shp);
 207                 return error;
 208         }
 209
 210         if (shmflg & SHM_HUGETLB) {
 211                 /* hugetlb_zero_setup takes care of mlock user accounting */
 212                 file = hugetlb_zero_setup(size);
 213                 shp->mlock_user = current->user;
 214         } else {
 215                 sprintf (name, "SYSV%08x", key);
 216                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 217         }
 218         error = PTR_ERR(file);
 219         if (IS_ERR(file))
 220                 goto no_file;
 221
 222         error = -ENOSPC;
 223         id = shm_addid(shp);
 224         if(id == -1)
 225                 goto no_id;
 226
 227         shp->shm_cprid = current->tgid;
 228         shp->shm_lprid = 0;
 229         shp->shm_atim = shp->shm_dtim = 0;
 230         shp->shm_ctim = get_seconds();
 231         shp->shm_segsz = size;
 232         shp->shm_nattch = 0;
 233         shp->id = shm_buildid(id,shp->shm_perm.seq);
 234         shp->shm_file = file;
 235         file->f_dentry->d_inode->i_ino = shp->id;
 236
 237         /* Hugetlb ops would have already been assigned. */
 238         if (!(shmflg & SHM_HUGETLB))
 239                 file->f_op = &shm_file_operations;
 240
 241         shm_tot += numpages;
 242         shm_unlock(shp);
 243         return shp->id;
 244
 245 no_id:
 246         fput(file);
 247 no_file:
 248         security_shm_free(shp);
 249         ipc_rcu_putref(shp);
 250         return error;
 251 }
 252
 253 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 254 {
 255         struct shmid_kernel *shp;
 256         int err, id = 0;
 257
 258         down(&shm_ids.sem);
 259         if (key == IPC_PRIVATE) {
 260                 err = newseg(key, shmflg, size);
 261         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 262                 if (!(shmflg & IPC_CREAT))
 263                         err = -ENOENT;
 264                 else
 265                         err = newseg(key, shmflg, size);
 266         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 267                 err = -EEXIST;
 268         } else {
 269                 shp = shm_lock(id);
 270                 if(shp==NULL)
 271                         BUG();
 272                 if (shp->shm_segsz < size)
 273                         err = -EINVAL;
 274                 else if (ipcperms(&shp->shm_perm, shmflg))
 275                         err = -EACCES;
 276                 else {
 277                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 278                         err = security_shm_associate(shp, shmflg);
 279                         if (!err)
 280                                 err = shmid;
 281                 }
 282                 shm_unlock(shp);
 283         }
 284         up(&shm_ids.sem);
 285
 286         return err;
 287 }
 288
 289 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 290 {
 291         switch(version) {
 292         case IPC_64:
 293                 return copy_to_user(buf, in, sizeof(*in));
 294         case IPC_OLD:
 295             {
 296                 struct shmid_ds out;
 297
 298                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 299                 out.shm_segsz   = in->shm_segsz;
 300                 out.shm_atime   = in->shm_atime;
 301                 out.shm_dtime   = in->shm_dtime;
 302                 out.shm_ctime   = in->shm_ctime;
 303                 out.shm_cpid    = in->shm_cpid;
 304                 out.shm_lpid    = in->shm_lpid;
 305                 out.shm_nattch  = in->shm_nattch;
 306
 307                 return copy_to_user(buf, &out, sizeof(out));
 308             }
 309         default:
 310                 return -EINVAL;
 311         }
 312 }
 313
 314 struct shm_setbuf {
 315         uid_t   uid;
 316         gid_t   gid;
 317         mode_t  mode;
 318 };
 319
 320 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 321 {
 322         switch(version) {
 323         case IPC_64:
 324             {
 325                 struct shmid64_ds tbuf;
 326
 327                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 328                         return -EFAULT;
 329
 330                 out->uid        = tbuf.shm_perm.uid;
 331                 out->gid        = tbuf.shm_perm.gid;
 332                 out->mode       = tbuf.shm_flags;
 333
 334                 return 0;
 335             }
 336         case IPC_OLD:
 337             {
 338                 struct shmid_ds tbuf_old;
 339
 340                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 341                         return -EFAULT;
 342
 343                 out->uid        = tbuf_old.shm_perm.uid;
 344                 out->gid        = tbuf_old.shm_perm.gid;
 345                 out->mode       = tbuf_old.shm_flags;
 346
 347                 return 0;
 348             }
 349         default:
 350                 return -EINVAL;
 351         }
 352 }
 353
 354 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 355 {
 356         switch(version) {
 357         case IPC_64:
 358                 return copy_to_user(buf, in, sizeof(*in));
 359         case IPC_OLD:
 360             {
 361                 struct shminfo out;
 362
 363                 if(in->shmmax > INT_MAX)
 364                         out.shmmax = INT_MAX;
 365                 else
 366                         out.shmmax = (int)in->shmmax;
 367
 368                 out.shmmin      = in->shmmin;
 369                 out.shmmni      = in->shmmni;
 370                 out.shmseg      = in->shmseg;
 371                 out.shmall      = in->shmall;
 372
 373                 return copy_to_user(buf, &out, sizeof(out));
 374             }
 375         default:
 376                 return -EINVAL;
 377         }
 378 }
 379
 380 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 381 {
 382         int i;
 383
 384         *rss = 0;
 385         *swp = 0;
 386
 387         for (i = 0; i <= shm_ids.max_id; i++) {
 388                 struct shmid_kernel *shp;
 389                 struct inode *inode;
 390
 391                 shp = shm_get(i);
 392                 if(!shp)
 393                         continue;
 394
 395                 inode = shp->shm_file->f_dentry->d_inode;
 396
 397                 if (is_file_hugepages(shp->shm_file)) {
 398                         struct address_space *mapping = inode->i_mapping;
 399                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 400                 } else {
 401                         struct shmem_inode_info *info = SHMEM_I(inode);
 402                         spin_lock(&info->lock);
 403                         *rss += inode->i_mapping->nrpages;
 404                         *swp += info->swapped;
 405                         spin_unlock(&info->lock);
 406                 }
 407         }
 408 }
 409
 410 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 411 {
 412         struct shm_setbuf setbuf;
 413         struct shmid_kernel *shp;
 414         int err, version;
 415
 416         if (cmd < 0 || shmid < 0) {
 417                 err = -EINVAL;
 418                 goto out;
 419         }
 420
 421         version = ipc_parse_version(&cmd);
 422
 423         switch (cmd) { /* replace with proc interface ? */
 424         case IPC_INFO:
 425         {
 426                 struct shminfo64 shminfo;
 427
 428                 err = security_shm_shmctl(NULL, cmd);
 429                 if (err)
 430                         return err;
 431
 432                 memset(&shminfo,0,sizeof(shminfo));
 433                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 434                 shminfo.shmmax = shm_ctlmax;
 435                 shminfo.shmall = shm_ctlall;
 436
 437                 shminfo.shmmin = SHMMIN;
 438                 if(copy_shminfo_to_user (buf, &shminfo, version))
 439                         return -EFAULT;
 440                 /* reading a integer is always atomic */
 441                 err= shm_ids.max_id;
 442                 if(err<0)
 443                         err = 0;
 444                 goto out;
 445         }
 446         case SHM_INFO:
 447         {
 448                 struct shm_info shm_info;
 449
 450                 err = security_shm_shmctl(NULL, cmd);
 451                 if (err)
 452                         return err;
 453
 454                 memset(&shm_info,0,sizeof(shm_info));
 455                 down(&shm_ids.sem);
 456                 shm_info.used_ids = shm_ids.in_use;
 457                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 458                 shm_info.shm_tot = shm_tot;
 459                 shm_info.swap_attempts = 0;
 460                 shm_info.swap_successes = 0;
 461                 err = shm_ids.max_id;
 462                 up(&shm_ids.sem);
 463                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 464                         err = -EFAULT;
 465                         goto out;
 466                 }
 467
 468                 err = err < 0 ? 0 : err;
 469                 goto out;
 470         }
 471         case SHM_STAT:
 472         case IPC_STAT:
 473         {
 474                 struct shmid64_ds tbuf;
 475                 int result;
 476                 memset(&tbuf, 0, sizeof(tbuf));
 477                 shp = shm_lock(shmid);
 478                 if(shp==NULL) {
 479                         err = -EINVAL;
 480                         goto out;
 481                 } else if(cmd==SHM_STAT) {
 482                         err = -EINVAL;
 483                         if (shmid > shm_ids.max_id)
 484                                 goto out_unlock;
 485                         result = shm_buildid(shmid, shp->shm_perm.seq);
 486                 } else {
 487                         err = shm_checkid(shp,shmid);
 488                         if(err)
 489                                 goto out_unlock;
 490                         result = 0;
 491                 }
 492                 err=-EACCES;
 493                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 494                         goto out_unlock;
 495                 err = security_shm_shmctl(shp, cmd);
 496                 if (err)
 497                         goto out_unlock;
 498                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 499                 tbuf.shm_segsz  = shp->shm_segsz;
 500                 tbuf.shm_atime  = shp->shm_atim;
 501                 tbuf.shm_dtime  = shp->shm_dtim;
 502                 tbuf.shm_ctime  = shp->shm_ctim;
 503                 tbuf.shm_cpid   = shp->shm_cprid;
 504                 tbuf.shm_lpid   = shp->shm_lprid;
 505                 if (!is_file_hugepages(shp->shm_file))
 506                         tbuf.shm_nattch = shp->shm_nattch;
 507                 else
 508                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 509                 shm_unlock(shp);
 510                 if(copy_shmid_to_user (buf, &tbuf, version))
 511                         err = -EFAULT;
 512                 else
 513                         err = result;
 514                 goto out;
 515         }
 516         case SHM_LOCK:
 517         case SHM_UNLOCK:
 518         {
 519                 shp = shm_lock(shmid);
 520                 if(shp==NULL) {
 521                         err = -EINVAL;
 522                         goto out;
 523                 }
 524                 err = shm_checkid(shp,shmid);
 525                 if(err)
 526                         goto out_unlock;
 527
 528                 if (!capable(CAP_IPC_LOCK)) {
 529                         err = -EPERM;
 530                         if (current->euid != shp->shm_perm.uid &&
 531                             current->euid != shp->shm_perm.cuid)
 532                                 goto out_unlock;
 533                         if (cmd == SHM_LOCK &&
 534                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 535                                 goto out_unlock;
 536                 }
 537
 538                 err = security_shm_shmctl(shp, cmd);
 539                 if (err)
 540                         goto out_unlock;
 541
 542                 if(cmd==SHM_LOCK) {
 543                         struct user_struct * user = current->user;
 544                         if (!is_file_hugepages(shp->shm_file)) {
 545                                 err = shmem_lock(shp->shm_file, 1, user);
 546                                 if (!err) {
 547                                         shp->shm_flags |= SHM_LOCKED;
 548                                         shp->mlock_user = user;
 549                                 }
 550                         }
 551                 } else if (!is_file_hugepages(shp->shm_file)) {
 552                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 553                         shp->shm_flags &= ~SHM_LOCKED;
 554                         shp->mlock_user = NULL;
 555                 }
 556                 shm_unlock(shp);
 557                 goto out;
 558         }
 559         case IPC_RMID:
 560         {
 561                 /*
 562                  *      We cannot simply remove the file. The SVID states
 563                  *      that the block remains until the last person
 564                  *      detaches from it, then is deleted. A shmat() on
 565                  *      an RMID segment is legal in older Linux and if
 566                  *      we change it apps break...
 567                  *
 568                  *      Instead we set a destroyed flag, and then blow
 569                  *      the name away when the usage hits zero.
 570                  */
 571                 down(&shm_ids.sem);
 572                 shp = shm_lock(shmid);
 573                 err = -EINVAL;
 574                 if (shp == NULL)
 575                         goto out_up;
 576                 err = shm_checkid(shp, shmid);
 577                 if(err)
 578                         goto out_unlock_up;
 579
 580                 if (current->euid != shp->shm_perm.uid &&
 581                     current->euid != shp->shm_perm.cuid &&
 582                     !capable(CAP_SYS_ADMIN)) {
 583                         err=-EPERM;
 584                         goto out_unlock_up;
 585                 }
 586
 587                 err = security_shm_shmctl(shp, cmd);
 588                 if (err)
 589                         goto out_unlock_up;
 590
 591                 if (shp->shm_nattch){
 592                         shp->shm_flags |= SHM_DEST;
 593                         /* Do not find it any more */
 594                         shp->shm_perm.key = IPC_PRIVATE;
 595                         shm_unlock(shp);
 596                 } else
 597                         shm_destroy (shp);
 598                 up(&shm_ids.sem);
 599                 goto out;
 600         }
 601
 602         case IPC_SET:
 603         {
 604                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 605                         err = -EFAULT;
 606                         goto out;
 607                 }
 608                 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
 609                         return err;
 610                 down(&shm_ids.sem);
 611                 shp = shm_lock(shmid);
 612                 err=-EINVAL;
 613                 if(shp==NULL)
 614                         goto out_up;
 615                 err = shm_checkid(shp,shmid);
 616                 if(err)
 617                         goto out_unlock_up;
 618                 err=-EPERM;
 619                 if (current->euid != shp->shm_perm.uid &&
 620                     current->euid != shp->shm_perm.cuid &&
 621                     !capable(CAP_SYS_ADMIN)) {
 622                         goto out_unlock_up;
 623                 }
 624
 625                 err = security_shm_shmctl(shp, cmd);
 626                 if (err)
 627                         goto out_unlock_up;
 628
 629                 shp->shm_perm.uid = setbuf.uid;
 630                 shp->shm_perm.gid = setbuf.gid;
 631                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 632                         | (setbuf.mode & S_IRWXUGO);
 633                 shp->shm_ctim = get_seconds();
 634                 break;
 635         }
 636
 637         default:
 638                 err = -EINVAL;
 639                 goto out;
 640         }
 641
 642         err = 0;
 643 out_unlock_up:
 644         shm_unlock(shp);
 645 out_up:
 646         up(&shm_ids.sem);
 647         goto out;
 648 out_unlock:
 649         shm_unlock(shp);
 650 out:
 651         return err;
 652 }
 653
 654 /*
 655  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 656  *
 657  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 658  * "raddr" thing points to kernel space, and there has to be a wrapper around
 659  * this.
 660  */
 661 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 662 {
 663         struct shmid_kernel *shp;
 664         unsigned long addr;
 665         unsigned long size;
 666         struct file * file;
 667         int    err;
 668         unsigned long flags;
 669         unsigned long prot;
 670         unsigned long o_flags;
 671         int acc_mode;
 672         void *user_addr;
 673
 674         if (shmid < 0) {
 675                 err = -EINVAL;
 676                 goto out;
 677         } else if ((addr = (ulong)shmaddr)) {
 678                 if (addr & (SHMLBA-1)) {
 679                         if (shmflg & SHM_RND)
 680                                 addr &= ~(SHMLBA-1);       /* round down */
 681                         else
 682 #ifndef __ARCH_FORCE_SHMLBA
 683                                 if (addr & ~PAGE_MASK)
 684 #endif
 685                                         return -EINVAL;
 686                 }
 687                 flags = MAP_SHARED | MAP_FIXED;
 688         } else {
 689                 if ((shmflg & SHM_REMAP))
 690                         return -EINVAL;
 691
 692                 flags = MAP_SHARED;
 693         }
 694
 695         if (shmflg & SHM_RDONLY) {
 696                 prot = PROT_READ;
 697                 o_flags = O_RDONLY;
 698                 acc_mode = S_IRUGO;
 699         } else {
 700                 prot = PROT_READ | PROT_WRITE;
 701                 o_flags = O_RDWR;
 702                 acc_mode = S_IRUGO | S_IWUGO;
 703         }
 704         if (shmflg & SHM_EXEC) {
 705                 prot |= PROT_EXEC;
 706                 acc_mode |= S_IXUGO;
 707         }
 708
 709         /*
 710          * We cannot rely on the fs check since SYSV IPC does have an
 711          * additional creator id...
 712          */
 713         shp = shm_lock(shmid);
 714         if(shp == NULL) {
 715                 err = -EINVAL;
 716                 goto out;
 717         }
 718         err = shm_checkid(shp,shmid);
 719         if (err) {
 720                 shm_unlock(shp);
 721                 goto out;
 722         }
 723         if (ipcperms(&shp->shm_perm, acc_mode)) {
 724                 shm_unlock(shp);
 725                 err = -EACCES;
 726                 goto out;
 727         }
 728
 729         err = security_shm_shmat(shp, shmaddr, shmflg);
 730         if (err) {
 731                 shm_unlock(shp);
 732                 return err;
 733         }
 734
 735         file = shp->shm_file;
 736         size = i_size_read(file->f_dentry->d_inode);
 737         shp->shm_nattch++;
 738         shm_unlock(shp);
 739
 740         down_write(&current->mm->mmap_sem);
 741         if (addr && !(shmflg & SHM_REMAP)) {
 742                 user_addr = ERR_PTR(-EINVAL);
 743                 if (find_vma_intersection(current->mm, addr, addr + size))
 744                         goto invalid;
 745                 /*
 746                  * If shm segment goes below stack, make sure there is some
 747                  * space left for the stack to grow (at least 4 pages).
 748                  */
 749                 if (addr < current->mm->start_stack &&
 750                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 751                         goto invalid;
 752         }
 753
 754         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 755
 756 invalid:
 757         up_write(&current->mm->mmap_sem);
 758
 759         down (&shm_ids.sem);
 760         if(!(shp = shm_lock(shmid)))
 761                 BUG();
 762         shp->shm_nattch--;
 763         if(shp->shm_nattch == 0 &&
 764            shp->shm_flags & SHM_DEST)
 765                 shm_destroy (shp);
 766         else
 767                 shm_unlock(shp);
 768         up (&shm_ids.sem);
 769
 770         *raddr = (unsigned long) user_addr;
 771         err = 0;
 772         if (IS_ERR(user_addr))
 773                 err = PTR_ERR(user_addr);
 774 out:
 775         return err;
 776 }
 777
 778 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 779 {
 780         unsigned long ret;
 781         long err;
 782
 783         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 784         if (err)
 785                 return err;
 786         force_successful_syscall_return();
 787         return (long)ret;
 788 }
 789
 790 /*
 791  * detach and kill segment if marked destroyed.
 792  * The work is done in shm_close.
 793  */
 794 asmlinkage long sys_shmdt(char __user *shmaddr)
 795 {
 796         struct mm_struct *mm = current->mm;
 797         struct vm_area_struct *vma, *next;
 798         unsigned long addr = (unsigned long)shmaddr;
 799         loff_t size = 0;
 800         int retval = -EINVAL;
 801
 802         down_write(&mm->mmap_sem);
 803
 804         /*
 805          * This function tries to be smart and unmap shm segments that
 806          * were modified by partial mlock or munmap calls:
 807          * - It first determines the size of the shm segment that should be
 808          *   unmapped: It searches for a vma that is backed by shm and that
 809          *   started at address shmaddr. It records it's size and then unmaps
 810          *   it.
 811          * - Then it unmaps all shm vmas that started at shmaddr and that
 812          *   are within the initially determined size.
 813          * Errors from do_munmap are ignored: the function only fails if
 814          * it's called with invalid parameters or if it's called to unmap
 815          * a part of a vma. Both calls in this function are for full vmas,
 816          * the parameters are directly copied from the vma itself and always
 817          * valid - therefore do_munmap cannot fail. (famous last words?)
 818          */
 819         /*
 820          * If it had been mremap()'d, the starting address would not
 821          * match the usual checks anyway. So assume all vma's are
 822          * above the starting address given.
 823          */
 824         vma = find_vma(mm, addr);
 825
 826         while (vma) {
 827                 next = vma->vm_next;
 828
 829                 /*
 830                  * Check if the starting address would match, i.e. it's
 831                  * a fragment created by mprotect() and/or munmap(), or it
 832                  * otherwise it starts at this address with no hassles.
 833                  */
 834                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 835                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 836
 837
 838                         size = vma->vm_file->f_dentry->d_inode->i_size;
 839                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 840                         /*
 841                          * We discovered the size of the shm segment, so
 842                          * break out of here and fall through to the next
 843                          * loop that uses the size information to stop
 844                          * searching for matching vma's.
 845                          */
 846                         retval = 0;
 847                         vma = next;
 848                         break;
 849                 }
 850                 vma = next;
 851         }
 852
 853         /*
 854          * We need look no further than the maximum address a fragment
 855          * could possibly have landed at. Also cast things to loff_t to
 856          * prevent overflows and make comparisions vs. equal-width types.
 857          */
 858         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 859                 next = vma->vm_next;
 860
 861                 /* finding a matching vma now does not alter retval */
 862                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 863                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 864
 865                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 866                 vma = next;
 867         }
 868
 869         up_write(&mm->mmap_sem);
 870         return retval;
 871 }
 872
 873 #ifdef CONFIG_PROC_FS
 874 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 875 {
 876         struct shmid_kernel *shp = it;
 877         char *format;
 878
 879 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 880 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 881
 882         if (sizeof(size_t) <= sizeof(int))
 883                 format = SMALL_STRING;
 884         else
 885                 format = BIG_STRING;
 886         return seq_printf(s, format,
 887                           shp->shm_perm.key,
 888                           shp->id,
 889                           shp->shm_flags,
 890                           shp->shm_segsz,
 891                           shp->shm_cprid,
 892                           shp->shm_lprid,
 893                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 894                           shp->shm_perm.uid,
 895                           shp->shm_perm.gid,
 896                           shp->shm_perm.cuid,
 897                           shp->shm_perm.cgid,
 898                           shp->shm_atim,
 899                           shp->shm_dtim,
 900                           shp->shm_ctim);
 901 }
 902 #endif