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