ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/slab.h>
  20 #include <linux/mm.h>
  21 #include <linux/hugetlb.h>
  22 #include <linux/shm.h>
  23 #include <linux/init.h>
  24 #include <linux/file.h>
  25 #include <linux/mman.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/shmem_fs.h>
  28 #include <linux/security.h>
  29 #include <linux/vs_base.h>
  30
  31 #include <asm/uaccess.h>
  32
  33 #include "util.h"
  34
  35 #define shm_flags       shm_perm.mode
  36
  37 static struct file_operations shm_file_operations;
  38 static struct vm_operations_struct shm_vm_ops;
  39
  40 static struct ipc_ids shm_ids;
  41
  42 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  43 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  44 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  45 #define shm_buildid(id, seq) \
  46         ipc_buildid(&shm_ids, id, seq)
  47
  48 static int newseg (key_t key, int shmflg, size_t size);
  49 static void shm_open (struct vm_area_struct *shmd);
  50 static void shm_close (struct vm_area_struct *shmd);
  51 #ifdef CONFIG_PROC_FS
  52 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  53 #endif
  54
  55 size_t  shm_ctlmax = SHMMAX;
  56 size_t  shm_ctlall = SHMALL;
  57 int     shm_ctlmni = SHMMNI;
  58
  59 static int shm_tot; /* total number of shared memory pages */
  60
  61 void __init shm_init (void)
  62 {
  63         ipc_init_ids(&shm_ids, 1);
  64 #ifdef CONFIG_PROC_FS
  65         create_proc_read_entry("sysvipc/shm", 0, NULL, sysvipc_shm_read_proc, NULL);
  66 #endif
  67 }
  68
  69 static inline int shm_checkid(struct shmid_kernel *s, int id)
  70 {
  71         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  72                 return -EIDRM;
  73         return 0;
  74 }
  75
  76 static inline struct shmid_kernel *shm_rmid(int id)
  77 {
  78         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  79 }
  80
  81 static inline int shm_addid(struct shmid_kernel *shp)
  82 {
  83         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  84 }
  85
  86
  87
  88 static inline void shm_inc (int id) {
  89         struct shmid_kernel *shp;
  90
  91         if(!(shp = shm_lock(id)))
  92                 BUG();
  93         shp->shm_atim = get_seconds();
  94         shp->shm_lprid = current->tgid;
  95         shp->shm_nattch++;
  96         shm_unlock(shp);
  97 }
  98
  99 /* This is called by fork, once for every shm attach. */
 100 static void shm_open (struct vm_area_struct *shmd)
 101 {
 102         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 103 }
 104
 105 /*
 106  * shm_destroy - free the struct shmid_kernel
 107  *
 108  * @shp: struct to free
 109  *
 110  * It has to be called with shp and shm_ids.sem locked,
 111  * but returns with shp unlocked and freed.
 112  */
 113 static void shm_destroy (struct shmid_kernel *shp)
 114 {
 115         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 116         shm_rmid (shp->id);
 117         shm_unlock(shp);
 118         if (!is_file_hugepages(shp->shm_file))
 119                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 120         else
 121                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 122                                                 shp->mlock_user);
 123         fput (shp->shm_file);
 124         security_shm_free(shp);
 125         ipc_rcu_putref(shp);
 126 }
 127
 128 /*
 129  * remove the attach descriptor shmd.
 130  * free memory for segment if it is marked destroyed.
 131  * The descriptor has already been removed from the current->mm->mmap list
 132  * and will later be kfree()d.
 133  */
 134 static void shm_close (struct vm_area_struct *shmd)
 135 {
 136         struct file * file = shmd->vm_file;
 137         int id = file->f_dentry->d_inode->i_ino;
 138         struct shmid_kernel *shp;
 139
 140         down (&shm_ids.sem);
 141         /* remove from the list of attaches of the shm segment */
 142         if(!(shp = shm_lock(id)))
 143                 BUG();
 144         shp->shm_lprid = current->tgid;
 145         shp->shm_dtim = get_seconds();
 146         shp->shm_nattch--;
 147         if(shp->shm_nattch == 0 &&
 148            shp->shm_flags & SHM_DEST)
 149                 shm_destroy (shp);
 150         else
 151                 shm_unlock(shp);
 152         up (&shm_ids.sem);
 153 }
 154
 155 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 156 {
 157         file_accessed(file);
 158         vma->vm_ops = &shm_vm_ops;
 159         shm_inc(file->f_dentry->d_inode->i_ino);
 160         return 0;
 161 }
 162
 163 static struct file_operations shm_file_operations = {
 164         .mmap   = shm_mmap
 165 };
 166
 167 static struct vm_operations_struct shm_vm_ops = {
 168         .open   = shm_open,     /* callback for a new vm-area open */
 169         .close  = shm_close,    /* callback for when the vm-area is released */
 170         .nopage = shmem_nopage,
 171 #ifdef CONFIG_NUMA
 172         .set_policy = shmem_set_policy,
 173         .get_policy = shmem_get_policy,
 174 #endif
 175 };
 176
 177 static int newseg (key_t key, int shmflg, size_t size)
 178 {
 179         int error;
 180         struct shmid_kernel *shp;
 181         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 182         struct file * file;
 183         char name[13];
 184         int id;
 185
 186         if (size < SHMMIN || size > shm_ctlmax)
 187                 return -EINVAL;
 188
 189         if (shm_tot + numpages >= shm_ctlall)
 190                 return -ENOSPC;
 191
 192         shp = ipc_rcu_alloc(sizeof(*shp));
 193         if (!shp)
 194                 return -ENOMEM;
 195
 196         shp->shm_perm.key = key;
 197         shp->shm_perm.xid = current->xid;
 198         shp->shm_flags = (shmflg & S_IRWXUGO);
 199         shp->mlock_user = NULL;
 200
 201         shp->shm_perm.security = NULL;
 202         error = security_shm_alloc(shp);
 203         if (error) {
 204                 ipc_rcu_putref(shp);
 205                 return error;
 206         }
 207
 208         if (shmflg & SHM_HUGETLB) {
 209                 /* hugetlb_zero_setup takes care of mlock user accounting */
 210                 file = hugetlb_zero_setup(size);
 211                 shp->mlock_user = current->user;
 212         } else {
 213                 sprintf (name, "SYSV%08x", key);
 214                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 215         }
 216         error = PTR_ERR(file);
 217         if (IS_ERR(file))
 218                 goto no_file;
 219
 220         error = -ENOSPC;
 221         id = shm_addid(shp);
 222         if(id == -1)
 223                 goto no_id;
 224
 225         shp->shm_cprid = current->tgid;
 226         shp->shm_lprid = 0;
 227         shp->shm_atim = shp->shm_dtim = 0;
 228         shp->shm_ctim = get_seconds();
 229         shp->shm_segsz = size;
 230         shp->shm_nattch = 0;
 231         shp->id = shm_buildid(id,shp->shm_perm.seq);
 232         shp->shm_file = file;
 233         file->f_dentry->d_inode->i_ino = shp->id;
 234         if (shmflg & SHM_HUGETLB)
 235                 set_file_hugepages(file);
 236         else
 237                 file->f_op = &shm_file_operations;
 238         shm_tot += numpages;
 239         shm_unlock(shp);
 240         return shp->id;
 241
 242 no_id:
 243         fput(file);
 244 no_file:
 245         security_shm_free(shp);
 246         ipc_rcu_putref(shp);
 247         return error;
 248 }
 249
 250 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 251 {
 252         struct shmid_kernel *shp;
 253         int err, id = 0;
 254
 255         down(&shm_ids.sem);
 256         if (key == IPC_PRIVATE) {
 257                 err = newseg(key, shmflg, size);
 258         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 259                 if (!(shmflg & IPC_CREAT))
 260                         err = -ENOENT;
 261                 else
 262                         err = newseg(key, shmflg, size);
 263         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 264                 err = -EEXIST;
 265         } else {
 266                 shp = shm_lock(id);
 267                 if(shp==NULL)
 268                         BUG();
 269                 if (shp->shm_segsz < size)
 270                         err = -EINVAL;
 271                 else if (ipcperms(&shp->shm_perm, shmflg))
 272                         err = -EACCES;
 273                 else {
 274                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 275                         err = security_shm_associate(shp, shmflg);
 276                         if (!err)
 277                                 err = shmid;
 278                 }
 279                 shm_unlock(shp);
 280         }
 281         up(&shm_ids.sem);
 282
 283         return err;
 284 }
 285
 286 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 287 {
 288         switch(version) {
 289         case IPC_64:
 290                 return copy_to_user(buf, in, sizeof(*in));
 291         case IPC_OLD:
 292             {
 293                 struct shmid_ds out;
 294
 295                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 296                 out.shm_segsz   = in->shm_segsz;
 297                 out.shm_atime   = in->shm_atime;
 298                 out.shm_dtime   = in->shm_dtime;
 299                 out.shm_ctime   = in->shm_ctime;
 300                 out.shm_cpid    = in->shm_cpid;
 301                 out.shm_lpid    = in->shm_lpid;
 302                 out.shm_nattch  = in->shm_nattch;
 303
 304                 return copy_to_user(buf, &out, sizeof(out));
 305             }
 306         default:
 307                 return -EINVAL;
 308         }
 309 }
 310
 311 struct shm_setbuf {
 312         uid_t   uid;
 313         gid_t   gid;
 314         mode_t  mode;
 315 };
 316
 317 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 318 {
 319         switch(version) {
 320         case IPC_64:
 321             {
 322                 struct shmid64_ds tbuf;
 323
 324                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 325                         return -EFAULT;
 326
 327                 out->uid        = tbuf.shm_perm.uid;
 328                 out->gid        = tbuf.shm_perm.gid;
 329                 out->mode       = tbuf.shm_flags;
 330
 331                 return 0;
 332             }
 333         case IPC_OLD:
 334             {
 335                 struct shmid_ds tbuf_old;
 336
 337                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 338                         return -EFAULT;
 339
 340                 out->uid        = tbuf_old.shm_perm.uid;
 341                 out->gid        = tbuf_old.shm_perm.gid;
 342                 out->mode       = tbuf_old.shm_flags;
 343
 344                 return 0;
 345             }
 346         default:
 347                 return -EINVAL;
 348         }
 349 }
 350
 351 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 352 {
 353         switch(version) {
 354         case IPC_64:
 355                 return copy_to_user(buf, in, sizeof(*in));
 356         case IPC_OLD:
 357             {
 358                 struct shminfo out;
 359
 360                 if(in->shmmax > INT_MAX)
 361                         out.shmmax = INT_MAX;
 362                 else
 363                         out.shmmax = (int)in->shmmax;
 364
 365                 out.shmmin      = in->shmmin;
 366                 out.shmmni      = in->shmmni;
 367                 out.shmseg      = in->shmseg;
 368                 out.shmall      = in->shmall;
 369
 370                 return copy_to_user(buf, &out, sizeof(out));
 371             }
 372         default:
 373                 return -EINVAL;
 374         }
 375 }
 376
 377 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 378 {
 379         int i;
 380
 381         *rss = 0;
 382         *swp = 0;
 383
 384         for (i = 0; i <= shm_ids.max_id; i++) {
 385                 struct shmid_kernel *shp;
 386                 struct inode *inode;
 387
 388                 shp = shm_get(i);
 389                 if(!shp)
 390                         continue;
 391
 392                 inode = shp->shm_file->f_dentry->d_inode;
 393
 394                 if (is_file_hugepages(shp->shm_file)) {
 395                         struct address_space *mapping = inode->i_mapping;
 396                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 397                 } else {
 398                         struct shmem_inode_info *info = SHMEM_I(inode);
 399                         spin_lock(&info->lock);
 400                         *rss += inode->i_mapping->nrpages;
 401                         *swp += info->swapped;
 402                         spin_unlock(&info->lock);
 403                 }
 404         }
 405 }
 406
 407 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 408 {
 409         struct shm_setbuf setbuf;
 410         struct shmid_kernel *shp;
 411         int err, version;
 412
 413         if (cmd < 0 || shmid < 0) {
 414                 err = -EINVAL;
 415                 goto out;
 416         }
 417
 418         version = ipc_parse_version(&cmd);
 419
 420         switch (cmd) { /* replace with proc interface ? */
 421         case IPC_INFO:
 422         {
 423                 struct shminfo64 shminfo;
 424
 425                 err = security_shm_shmctl(NULL, cmd);
 426                 if (err)
 427                         return err;
 428
 429                 memset(&shminfo,0,sizeof(shminfo));
 430                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 431                 shminfo.shmmax = shm_ctlmax;
 432                 shminfo.shmall = shm_ctlall;
 433
 434                 shminfo.shmmin = SHMMIN;
 435                 if(copy_shminfo_to_user (buf, &shminfo, version))
 436                         return -EFAULT;
 437                 /* reading a integer is always atomic */
 438                 err= shm_ids.max_id;
 439                 if(err<0)
 440                         err = 0;
 441                 goto out;
 442         }
 443         case SHM_INFO:
 444         {
 445                 struct shm_info shm_info;
 446
 447                 err = security_shm_shmctl(NULL, cmd);
 448                 if (err)
 449                         return err;
 450
 451                 memset(&shm_info,0,sizeof(shm_info));
 452                 down(&shm_ids.sem);
 453                 shm_info.used_ids = shm_ids.in_use;
 454                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 455                 shm_info.shm_tot = shm_tot;
 456                 shm_info.swap_attempts = 0;
 457                 shm_info.swap_successes = 0;
 458                 err = shm_ids.max_id;
 459                 up(&shm_ids.sem);
 460                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 461                         err = -EFAULT;
 462                         goto out;
 463                 }
 464
 465                 err = err < 0 ? 0 : err;
 466                 goto out;
 467         }
 468         case SHM_STAT:
 469         case IPC_STAT:
 470         {
 471                 struct shmid64_ds tbuf;
 472                 int result;
 473                 memset(&tbuf, 0, sizeof(tbuf));
 474                 shp = shm_lock(shmid);
 475                 if(shp==NULL) {
 476                         err = -EINVAL;
 477                         goto out;
 478                 } else if(cmd==SHM_STAT) {
 479                         err = -EINVAL;
 480                         if (shmid > shm_ids.max_id)
 481                                 goto out_unlock;
 482                         result = shm_buildid(shmid, shp->shm_perm.seq);
 483                 } else {
 484                         err = shm_checkid(shp,shmid);
 485                         if(err)
 486                                 goto out_unlock;
 487                         result = 0;
 488                 }
 489                 err=-EACCES;
 490                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 491                         goto out_unlock;
 492                 err = security_shm_shmctl(shp, cmd);
 493                 if (err)
 494                         goto out_unlock;
 495                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 496                 tbuf.shm_segsz  = shp->shm_segsz;
 497                 tbuf.shm_atime  = shp->shm_atim;
 498                 tbuf.shm_dtime  = shp->shm_dtim;
 499                 tbuf.shm_ctime  = shp->shm_ctim;
 500                 tbuf.shm_cpid   = shp->shm_cprid;
 501                 tbuf.shm_lpid   = shp->shm_lprid;
 502                 if (!is_file_hugepages(shp->shm_file))
 503                         tbuf.shm_nattch = shp->shm_nattch;
 504                 else
 505                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 506                 shm_unlock(shp);
 507                 if(copy_shmid_to_user (buf, &tbuf, version))
 508                         err = -EFAULT;
 509                 else
 510                         err = result;
 511                 goto out;
 512         }
 513         case SHM_LOCK:
 514         case SHM_UNLOCK:
 515         {
 516                 /* Allow superuser to lock segment in memory */
 517                 if (!can_do_mlock() && cmd == SHM_LOCK) {
 518                         err = -EPERM;
 519                         goto out;
 520                 }
 521                 shp = shm_lock(shmid);
 522                 if(shp==NULL) {
 523                         err = -EINVAL;
 524                         goto out;
 525                 }
 526                 err = shm_checkid(shp,shmid);
 527                 if(err)
 528                         goto out_unlock;
 529
 530                 err = security_shm_shmctl(shp, cmd);
 531                 if (err)
 532                         goto out_unlock;
 533
 534                 if(cmd==SHM_LOCK) {
 535                         struct user_struct * user = current->user;
 536                         if (!is_file_hugepages(shp->shm_file)) {
 537                                 err = shmem_lock(shp->shm_file, 1, user);
 538                                 if (!err) {
 539                                         shp->shm_flags |= SHM_LOCKED;
 540                                         shp->mlock_user = user;
 541                                 }
 542                         }
 543                 } else if (!is_file_hugepages(shp->shm_file)) {
 544                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 545                         shp->shm_flags &= ~SHM_LOCKED;
 546                         shp->mlock_user = NULL;
 547                 }
 548                 shm_unlock(shp);
 549                 goto out;
 550         }
 551         case IPC_RMID:
 552         {
 553                 /*
 554                  *      We cannot simply remove the file. The SVID states
 555                  *      that the block remains until the last person
 556                  *      detaches from it, then is deleted. A shmat() on
 557                  *      an RMID segment is legal in older Linux and if
 558                  *      we change it apps break...
 559                  *
 560                  *      Instead we set a destroyed flag, and then blow
 561                  *      the name away when the usage hits zero.
 562                  */
 563                 down(&shm_ids.sem);
 564                 shp = shm_lock(shmid);
 565                 err = -EINVAL;
 566                 if (shp == NULL)
 567                         goto out_up;
 568                 err = shm_checkid(shp, shmid);
 569                 if(err)
 570                         goto out_unlock_up;
 571
 572                 if (current->euid != shp->shm_perm.uid &&
 573                     current->euid != shp->shm_perm.cuid &&
 574                     !capable(CAP_SYS_ADMIN)) {
 575                         err=-EPERM;
 576                         goto out_unlock_up;
 577                 }
 578
 579                 err = security_shm_shmctl(shp, cmd);
 580                 if (err)
 581                         goto out_unlock_up;
 582
 583                 if (shp->shm_nattch){
 584                         shp->shm_flags |= SHM_DEST;
 585                         /* Do not find it any more */
 586                         shp->shm_perm.key = IPC_PRIVATE;
 587                         shm_unlock(shp);
 588                 } else
 589                         shm_destroy (shp);
 590                 up(&shm_ids.sem);
 591                 goto out;
 592         }
 593
 594         case IPC_SET:
 595         {
 596                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 597                         err = -EFAULT;
 598                         goto out;
 599                 }
 600                 down(&shm_ids.sem);
 601                 shp = shm_lock(shmid);
 602                 err=-EINVAL;
 603                 if(shp==NULL)
 604                         goto out_up;
 605                 err = shm_checkid(shp,shmid);
 606                 if(err)
 607                         goto out_unlock_up;
 608                 err=-EPERM;
 609                 if (current->euid != shp->shm_perm.uid &&
 610                     current->euid != shp->shm_perm.cuid &&
 611                     !capable(CAP_SYS_ADMIN)) {
 612                         goto out_unlock_up;
 613                 }
 614
 615                 err = security_shm_shmctl(shp, cmd);
 616                 if (err)
 617                         goto out_unlock_up;
 618
 619                 shp->shm_perm.uid = setbuf.uid;
 620                 shp->shm_perm.gid = setbuf.gid;
 621                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 622                         | (setbuf.mode & S_IRWXUGO);
 623                 shp->shm_ctim = get_seconds();
 624                 break;
 625         }
 626
 627         default:
 628                 err = -EINVAL;
 629                 goto out;
 630         }
 631
 632         err = 0;
 633 out_unlock_up:
 634         shm_unlock(shp);
 635 out_up:
 636         up(&shm_ids.sem);
 637         goto out;
 638 out_unlock:
 639         shm_unlock(shp);
 640 out:
 641         return err;
 642 }
 643
 644 /*
 645  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 646  *
 647  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 648  * "raddr" thing points to kernel space, and there has to be a wrapper around
 649  * this.
 650  */
 651 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 652 {
 653         struct shmid_kernel *shp;
 654         unsigned long addr;
 655         unsigned long size;
 656         struct file * file;
 657         int    err;
 658         unsigned long flags;
 659         unsigned long prot;
 660         unsigned long o_flags;
 661         int acc_mode;
 662         void *user_addr;
 663
 664         if (shmid < 0) {
 665                 err = -EINVAL;
 666                 goto out;
 667         } else if ((addr = (ulong)shmaddr)) {
 668                 if (addr & (SHMLBA-1)) {
 669                         if (shmflg & SHM_RND)
 670                                 addr &= ~(SHMLBA-1);       /* round down */
 671                         else
 672 #ifndef __ARCH_FORCE_SHMLBA
 673                                 if (addr & ~PAGE_MASK)
 674 #endif
 675                                         return -EINVAL;
 676                 }
 677                 flags = MAP_SHARED | MAP_FIXED;
 678         } else {
 679                 if ((shmflg & SHM_REMAP))
 680                         return -EINVAL;
 681
 682                 flags = MAP_SHARED;
 683         }
 684
 685         if (shmflg & SHM_RDONLY) {
 686                 prot = PROT_READ;
 687                 o_flags = O_RDONLY;
 688                 acc_mode = S_IRUGO;
 689         } else {
 690                 prot = PROT_READ | PROT_WRITE;
 691                 o_flags = O_RDWR;
 692                 acc_mode = S_IRUGO | S_IWUGO;
 693         }
 694         if (shmflg & SHM_EXEC) {
 695                 prot |= PROT_EXEC;
 696                 acc_mode |= S_IXUGO;
 697         }
 698
 699         /*
 700          * We cannot rely on the fs check since SYSV IPC does have an
 701          * additional creator id...
 702          */
 703         shp = shm_lock(shmid);
 704         if(shp == NULL) {
 705                 err = -EINVAL;
 706                 goto out;
 707         }
 708         err = shm_checkid(shp,shmid);
 709         if (err) {
 710                 shm_unlock(shp);
 711                 goto out;
 712         }
 713         if (ipcperms(&shp->shm_perm, acc_mode)) {
 714                 shm_unlock(shp);
 715                 err = -EACCES;
 716                 goto out;
 717         }
 718
 719         err = security_shm_shmat(shp, shmaddr, shmflg);
 720         if (err) {
 721                 shm_unlock(shp);
 722                 return err;
 723         }
 724
 725         file = shp->shm_file;
 726         size = i_size_read(file->f_dentry->d_inode);
 727         shp->shm_nattch++;
 728         shm_unlock(shp);
 729
 730         down_write(&current->mm->mmap_sem);
 731         if (addr && !(shmflg & SHM_REMAP)) {
 732                 user_addr = ERR_PTR(-EINVAL);
 733                 if (find_vma_intersection(current->mm, addr, addr + size))
 734                         goto invalid;
 735                 /*
 736                  * If shm segment goes below stack, make sure there is some
 737                  * space left for the stack to grow (at least 4 pages).
 738                  */
 739                 if (addr < current->mm->start_stack &&
 740                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 741                         goto invalid;
 742         }
 743
 744         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 745
 746 invalid:
 747         up_write(&current->mm->mmap_sem);
 748
 749         down (&shm_ids.sem);
 750         if(!(shp = shm_lock(shmid)))
 751                 BUG();
 752         shp->shm_nattch--;
 753         if(shp->shm_nattch == 0 &&
 754            shp->shm_flags & SHM_DEST)
 755                 shm_destroy (shp);
 756         else
 757                 shm_unlock(shp);
 758         up (&shm_ids.sem);
 759
 760         *raddr = (unsigned long) user_addr;
 761         err = 0;
 762         if (IS_ERR(user_addr))
 763                 err = PTR_ERR(user_addr);
 764 out:
 765         return err;
 766 }
 767
 768 /*
 769  * detach and kill segment if marked destroyed.
 770  * The work is done in shm_close.
 771  */
 772 asmlinkage long sys_shmdt(char __user *shmaddr)
 773 {
 774         struct mm_struct *mm = current->mm;
 775         struct vm_area_struct *vma, *next;
 776         unsigned long addr = (unsigned long)shmaddr;
 777         loff_t size = 0;
 778         int retval = -EINVAL;
 779
 780         down_write(&mm->mmap_sem);
 781
 782         /*
 783          * This function tries to be smart and unmap shm segments that
 784          * were modified by partial mlock or munmap calls:
 785          * - It first determines the size of the shm segment that should be
 786          *   unmapped: It searches for a vma that is backed by shm and that
 787          *   started at address shmaddr. It records it's size and then unmaps
 788          *   it.
 789          * - Then it unmaps all shm vmas that started at shmaddr and that
 790          *   are within the initially determined size.
 791          * Errors from do_munmap are ignored: the function only fails if
 792          * it's called with invalid parameters or if it's called to unmap
 793          * a part of a vma. Both calls in this function are for full vmas,
 794          * the parameters are directly copied from the vma itself and always
 795          * valid - therefore do_munmap cannot fail. (famous last words?)
 796          */
 797         /*
 798          * If it had been mremap()'d, the starting address would not
 799          * match the usual checks anyway. So assume all vma's are
 800          * above the starting address given.
 801          */
 802         vma = find_vma(mm, addr);
 803
 804         while (vma) {
 805                 next = vma->vm_next;
 806
 807                 /*
 808                  * Check if the starting address would match, i.e. it's
 809                  * a fragment created by mprotect() and/or munmap(), or it
 810                  * otherwise it starts at this address with no hassles.
 811                  */
 812                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 813                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 814
 815
 816                         size = vma->vm_file->f_dentry->d_inode->i_size;
 817                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 818                         /*
 819                          * We discovered the size of the shm segment, so
 820                          * break out of here and fall through to the next
 821                          * loop that uses the size information to stop
 822                          * searching for matching vma's.
 823                          */
 824                         retval = 0;
 825                         vma = next;
 826                         break;
 827                 }
 828                 vma = next;
 829         }
 830
 831         /*
 832          * We need look no further than the maximum address a fragment
 833          * could possibly have landed at. Also cast things to loff_t to
 834          * prevent overflows and make comparisions vs. equal-width types.
 835          */
 836         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 837                 next = vma->vm_next;
 838
 839                 /* finding a matching vma now does not alter retval */
 840                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 841                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 842
 843                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 844                 vma = next;
 845         }
 846
 847         up_write(&mm->mmap_sem);
 848         return retval;
 849 }
 850
 851 #ifdef CONFIG_PROC_FS
 852 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 853 {
 854         off_t pos = 0;
 855         off_t begin = 0;
 856         int i, len = 0;
 857
 858         down(&shm_ids.sem);
 859         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 860
 861         for(i = 0; i <= shm_ids.max_id; i++) {
 862                 struct shmid_kernel* shp;
 863
 864                 shp = shm_lock(i);
 865                 if (shp) {
 866 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 867 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 868                         char *format;
 869
 870                         if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
 871                                 shm_unlock(shp);
 872                                 continue;
 873                         }
 874                         if (sizeof(size_t) <= sizeof(int))
 875                                 format = SMALL_STRING;
 876                         else
 877                                 format = BIG_STRING;
 878                         len += sprintf(buffer + len, format,
 879                                 shp->shm_perm.key,
 880                                 shm_buildid(i, shp->shm_perm.seq),
 881                                 shp->shm_flags,
 882                                 shp->shm_segsz,
 883                                 shp->shm_cprid,
 884                                 shp->shm_lprid,
 885                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 886                                 shp->shm_perm.uid,
 887                                 shp->shm_perm.gid,
 888                                 shp->shm_perm.cuid,
 889                                 shp->shm_perm.cgid,
 890                                 shp->shm_atim,
 891                                 shp->shm_dtim,
 892                                 shp->shm_ctim);
 893                         shm_unlock(shp);
 894
 895                         pos += len;
 896                         if(pos < offset) {
 897                                 len = 0;
 898                                 begin = pos;
 899                         }
 900                         if(pos > offset + length)
 901                                 goto done;
 902                 }
 903         }
 904         *eof = 1;
 905 done:
 906         up(&shm_ids.sem);
 907         *start = buffer + (offset - begin);
 908         len -= (offset - begin);
 909         if(len > length)
 910                 len = length;
 911         if(len < 0)
 912                 len = 0;
 913         return len;
 914 }
 915 #endif