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