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