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/syscalls.h>
  30 #include <linux/audit.h>
  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_flags = (shmflg & S_IRWXUGO);
 198         shp->mlock_user = NULL;
 199
 200         shp->shm_perm.security = NULL;
 201         error = security_shm_alloc(shp);
 202         if (error) {
 203                 ipc_rcu_putref(shp);
 204                 return error;
 205         }
 206
 207         if (shmflg & SHM_HUGETLB) {
 208                 /* hugetlb_zero_setup takes care of mlock user accounting */
 209                 file = hugetlb_zero_setup(size);
 210                 shp->mlock_user = current->user;
 211         } else {
 212                 sprintf (name, "SYSV%08x", key);
 213                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 214         }
 215         error = PTR_ERR(file);
 216         if (IS_ERR(file))
 217                 goto no_file;
 218
 219         error = -ENOSPC;
 220         id = shm_addid(shp);
 221         if(id == -1)
 222                 goto no_id;
 223
 224         shp->shm_cprid = current->tgid;
 225         shp->shm_lprid = 0;
 226         shp->shm_atim = shp->shm_dtim = 0;
 227         shp->shm_ctim = get_seconds();
 228         shp->shm_segsz = size;
 229         shp->shm_nattch = 0;
 230         shp->id = shm_buildid(id,shp->shm_perm.seq);
 231         shp->shm_file = file;
 232         file->f_dentry->d_inode->i_ino = shp->id;
 233         if (shmflg & SHM_HUGETLB)
 234                 set_file_hugepages(file);
 235         else
 236                 file->f_op = &shm_file_operations;
 237         shm_tot += numpages;
 238         shm_unlock(shp);
 239         return shp->id;
 240
 241 no_id:
 242         fput(file);
 243 no_file:
 244         security_shm_free(shp);
 245         ipc_rcu_putref(shp);
 246         return error;
 247 }
 248
 249 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 250 {
 251         struct shmid_kernel *shp;
 252         int err, id = 0;
 253
 254         down(&shm_ids.sem);
 255         if (key == IPC_PRIVATE) {
 256                 err = newseg(key, shmflg, size);
 257         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 258                 if (!(shmflg & IPC_CREAT))
 259                         err = -ENOENT;
 260                 else
 261                         err = newseg(key, shmflg, size);
 262         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 263                 err = -EEXIST;
 264         } else {
 265                 shp = shm_lock(id);
 266                 if(shp==NULL)
 267                         BUG();
 268                 if (shp->shm_segsz < size)
 269                         err = -EINVAL;
 270                 else if (ipcperms(&shp->shm_perm, shmflg))
 271                         err = -EACCES;
 272                 else {
 273                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 274                         err = security_shm_associate(shp, shmflg);
 275                         if (!err)
 276                                 err = shmid;
 277                 }
 278                 shm_unlock(shp);
 279         }
 280         up(&shm_ids.sem);
 281
 282         return err;
 283 }
 284
 285 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 286 {
 287         switch(version) {
 288         case IPC_64:
 289                 return copy_to_user(buf, in, sizeof(*in));
 290         case IPC_OLD:
 291             {
 292                 struct shmid_ds out;
 293
 294                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 295                 out.shm_segsz   = in->shm_segsz;
 296                 out.shm_atime   = in->shm_atime;
 297                 out.shm_dtime   = in->shm_dtime;
 298                 out.shm_ctime   = in->shm_ctime;
 299                 out.shm_cpid    = in->shm_cpid;
 300                 out.shm_lpid    = in->shm_lpid;
 301                 out.shm_nattch  = in->shm_nattch;
 302
 303                 return copy_to_user(buf, &out, sizeof(out));
 304             }
 305         default:
 306                 return -EINVAL;
 307         }
 308 }
 309
 310 struct shm_setbuf {
 311         uid_t   uid;
 312         gid_t   gid;
 313         mode_t  mode;
 314 };
 315
 316 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 317 {
 318         switch(version) {
 319         case IPC_64:
 320             {
 321                 struct shmid64_ds tbuf;
 322
 323                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 324                         return -EFAULT;
 325
 326                 out->uid        = tbuf.shm_perm.uid;
 327                 out->gid        = tbuf.shm_perm.gid;
 328                 out->mode       = tbuf.shm_flags;
 329
 330                 return 0;
 331             }
 332         case IPC_OLD:
 333             {
 334                 struct shmid_ds tbuf_old;
 335
 336                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 337                         return -EFAULT;
 338
 339                 out->uid        = tbuf_old.shm_perm.uid;
 340                 out->gid        = tbuf_old.shm_perm.gid;
 341                 out->mode       = tbuf_old.shm_flags;
 342
 343                 return 0;
 344             }
 345         default:
 346                 return -EINVAL;
 347         }
 348 }
 349
 350 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 351 {
 352         switch(version) {
 353         case IPC_64:
 354                 return copy_to_user(buf, in, sizeof(*in));
 355         case IPC_OLD:
 356             {
 357                 struct shminfo out;
 358
 359                 if(in->shmmax > INT_MAX)
 360                         out.shmmax = INT_MAX;
 361                 else
 362                         out.shmmax = (int)in->shmmax;
 363
 364                 out.shmmin      = in->shmmin;
 365                 out.shmmni      = in->shmmni;
 366                 out.shmseg      = in->shmseg;
 367                 out.shmall      = in->shmall;
 368
 369                 return copy_to_user(buf, &out, sizeof(out));
 370             }
 371         default:
 372                 return -EINVAL;
 373         }
 374 }
 375
 376 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 377 {
 378         int i;
 379
 380         *rss = 0;
 381         *swp = 0;
 382
 383         for (i = 0; i <= shm_ids.max_id; i++) {
 384                 struct shmid_kernel *shp;
 385                 struct inode *inode;
 386
 387                 shp = shm_get(i);
 388                 if(!shp)
 389                         continue;
 390
 391                 inode = shp->shm_file->f_dentry->d_inode;
 392
 393                 if (is_file_hugepages(shp->shm_file)) {
 394                         struct address_space *mapping = inode->i_mapping;
 395                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 396                 } else {
 397                         struct shmem_inode_info *info = SHMEM_I(inode);
 398                         spin_lock(&info->lock);
 399                         *rss += inode->i_mapping->nrpages;
 400                         *swp += info->swapped;
 401                         spin_unlock(&info->lock);
 402                 }
 403         }
 404 }
 405
 406 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 407 {
 408         struct shm_setbuf setbuf;
 409         struct shmid_kernel *shp;
 410         int err, version;
 411
 412         if (cmd < 0 || shmid < 0) {
 413                 err = -EINVAL;
 414                 goto out;
 415         }
 416
 417         version = ipc_parse_version(&cmd);
 418
 419         switch (cmd) { /* replace with proc interface ? */
 420         case IPC_INFO:
 421         {
 422                 struct shminfo64 shminfo;
 423
 424                 err = security_shm_shmctl(NULL, cmd);
 425                 if (err)
 426                         return err;
 427
 428                 memset(&shminfo,0,sizeof(shminfo));
 429                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 430                 shminfo.shmmax = shm_ctlmax;
 431                 shminfo.shmall = shm_ctlall;
 432
 433                 shminfo.shmmin = SHMMIN;
 434                 if(copy_shminfo_to_user (buf, &shminfo, version))
 435                         return -EFAULT;
 436                 /* reading a integer is always atomic */
 437                 err= shm_ids.max_id;
 438                 if(err<0)
 439                         err = 0;
 440                 goto out;
 441         }
 442         case SHM_INFO:
 443         {
 444                 struct shm_info shm_info;
 445
 446                 err = security_shm_shmctl(NULL, cmd);
 447                 if (err)
 448                         return err;
 449
 450                 memset(&shm_info,0,sizeof(shm_info));
 451                 down(&shm_ids.sem);
 452                 shm_info.used_ids = shm_ids.in_use;
 453                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 454                 shm_info.shm_tot = shm_tot;
 455                 shm_info.swap_attempts = 0;
 456                 shm_info.swap_successes = 0;
 457                 err = shm_ids.max_id;
 458                 up(&shm_ids.sem);
 459                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 460                         err = -EFAULT;
 461                         goto out;
 462                 }
 463
 464                 err = err < 0 ? 0 : err;
 465                 goto out;
 466         }
 467         case SHM_STAT:
 468         case IPC_STAT:
 469         {
 470                 struct shmid64_ds tbuf;
 471                 int result;
 472                 memset(&tbuf, 0, sizeof(tbuf));
 473                 shp = shm_lock(shmid);
 474                 if(shp==NULL) {
 475                         err = -EINVAL;
 476                         goto out;
 477                 } else if(cmd==SHM_STAT) {
 478                         err = -EINVAL;
 479                         if (shmid > shm_ids.max_id)
 480                                 goto out_unlock;
 481                         result = shm_buildid(shmid, shp->shm_perm.seq);
 482                 } else {
 483                         err = shm_checkid(shp,shmid);
 484                         if(err)
 485                                 goto out_unlock;
 486                         result = 0;
 487                 }
 488                 err=-EACCES;
 489                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 490                         goto out_unlock;
 491                 err = security_shm_shmctl(shp, cmd);
 492                 if (err)
 493                         goto out_unlock;
 494                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 495                 tbuf.shm_segsz  = shp->shm_segsz;
 496                 tbuf.shm_atime  = shp->shm_atim;
 497                 tbuf.shm_dtime  = shp->shm_dtim;
 498                 tbuf.shm_ctime  = shp->shm_ctim;
 499                 tbuf.shm_cpid   = shp->shm_cprid;
 500                 tbuf.shm_lpid   = shp->shm_lprid;
 501                 if (!is_file_hugepages(shp->shm_file))
 502                         tbuf.shm_nattch = shp->shm_nattch;
 503                 else
 504                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 505                 shm_unlock(shp);
 506                 if(copy_shmid_to_user (buf, &tbuf, version))
 507                         err = -EFAULT;
 508                 else
 509                         err = result;
 510                 goto out;
 511         }
 512         case SHM_LOCK:
 513         case SHM_UNLOCK:
 514         {
 515                 shp = shm_lock(shmid);
 516                 if(shp==NULL) {
 517                         err = -EINVAL;
 518                         goto out;
 519                 }
 520                 err = shm_checkid(shp,shmid);
 521                 if(err)
 522                         goto out_unlock;
 523
 524                 if (!capable(CAP_IPC_LOCK)) {
 525                         err = -EPERM;
 526                         if (current->euid != shp->shm_perm.uid &&
 527                             current->euid != shp->shm_perm.cuid)
 528                                 goto out_unlock;
 529                         if (cmd == SHM_LOCK &&
 530                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 531                                 goto out_unlock;
 532                 }
 533
 534                 err = security_shm_shmctl(shp, cmd);
 535                 if (err)
 536                         goto out_unlock;
 537
 538                 if(cmd==SHM_LOCK) {
 539                         struct user_struct * user = current->user;
 540                         if (!is_file_hugepages(shp->shm_file)) {
 541                                 err = shmem_lock(shp->shm_file, 1, user);
 542                                 if (!err) {
 543                                         shp->shm_flags |= SHM_LOCKED;
 544                                         shp->mlock_user = user;
 545                                 }
 546                         }
 547                 } else if (!is_file_hugepages(shp->shm_file)) {
 548                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 549                         shp->shm_flags &= ~SHM_LOCKED;
 550                         shp->mlock_user = NULL;
 551                 }
 552                 shm_unlock(shp);
 553                 goto out;
 554         }
 555         case IPC_RMID:
 556         {
 557                 /*
 558                  *      We cannot simply remove the file. The SVID states
 559                  *      that the block remains until the last person
 560                  *      detaches from it, then is deleted. A shmat() on
 561                  *      an RMID segment is legal in older Linux and if
 562                  *      we change it apps break...
 563                  *
 564                  *      Instead we set a destroyed flag, and then blow
 565                  *      the name away when the usage hits zero.
 566                  */
 567                 down(&shm_ids.sem);
 568                 shp = shm_lock(shmid);
 569                 err = -EINVAL;
 570                 if (shp == NULL)
 571                         goto out_up;
 572                 err = shm_checkid(shp, shmid);
 573                 if(err)
 574                         goto out_unlock_up;
 575
 576                 if (current->euid != shp->shm_perm.uid &&
 577                     current->euid != shp->shm_perm.cuid &&
 578                     !capable(CAP_SYS_ADMIN)) {
 579                         err=-EPERM;
 580                         goto out_unlock_up;
 581                 }
 582
 583                 err = security_shm_shmctl(shp, cmd);
 584                 if (err)
 585                         goto out_unlock_up;
 586
 587                 if (shp->shm_nattch){
 588                         shp->shm_flags |= SHM_DEST;
 589                         /* Do not find it any more */
 590                         shp->shm_perm.key = IPC_PRIVATE;
 591                         shm_unlock(shp);
 592                 } else
 593                         shm_destroy (shp);
 594                 up(&shm_ids.sem);
 595                 goto out;
 596         }
 597
 598         case IPC_SET:
 599         {
 600                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 601                         err = -EFAULT;
 602                         goto out;
 603                 }
 604                 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
 605                         return err;
 606                 down(&shm_ids.sem);
 607                 shp = shm_lock(shmid);
 608                 err=-EINVAL;
 609                 if(shp==NULL)
 610                         goto out_up;
 611                 err = shm_checkid(shp,shmid);
 612                 if(err)
 613                         goto out_unlock_up;
 614                 err=-EPERM;
 615                 if (current->euid != shp->shm_perm.uid &&
 616                     current->euid != shp->shm_perm.cuid &&
 617                     !capable(CAP_SYS_ADMIN)) {
 618                         goto out_unlock_up;
 619                 }
 620
 621                 err = security_shm_shmctl(shp, cmd);
 622                 if (err)
 623                         goto out_unlock_up;
 624
 625                 shp->shm_perm.uid = setbuf.uid;
 626                 shp->shm_perm.gid = setbuf.gid;
 627                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 628                         | (setbuf.mode & S_IRWXUGO);
 629                 shp->shm_ctim = get_seconds();
 630                 break;
 631         }
 632
 633         default:
 634                 err = -EINVAL;
 635                 goto out;
 636         }
 637
 638         err = 0;
 639 out_unlock_up:
 640         shm_unlock(shp);
 641 out_up:
 642         up(&shm_ids.sem);
 643         goto out;
 644 out_unlock:
 645         shm_unlock(shp);
 646 out:
 647         return err;
 648 }
 649
 650 /*
 651  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 652  *
 653  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 654  * "raddr" thing points to kernel space, and there has to be a wrapper around
 655  * this.
 656  */
 657 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 658 {
 659         struct shmid_kernel *shp;
 660         unsigned long addr;
 661         unsigned long size;
 662         struct file * file;
 663         int    err;
 664         unsigned long flags;
 665         unsigned long prot;
 666         unsigned long o_flags;
 667         int acc_mode;
 668         void *user_addr;
 669
 670         if (shmid < 0) {
 671                 err = -EINVAL;
 672                 goto out;
 673         } else if ((addr = (ulong)shmaddr)) {
 674                 if (addr & (SHMLBA-1)) {
 675                         if (shmflg & SHM_RND)
 676                                 addr &= ~(SHMLBA-1);       /* round down */
 677                         else
 678 #ifndef __ARCH_FORCE_SHMLBA
 679                                 if (addr & ~PAGE_MASK)
 680 #endif
 681                                         return -EINVAL;
 682                 }
 683                 flags = MAP_SHARED | MAP_FIXED;
 684         } else {
 685                 if ((shmflg & SHM_REMAP))
 686                         return -EINVAL;
 687
 688                 flags = MAP_SHARED;
 689         }
 690
 691         if (shmflg & SHM_RDONLY) {
 692                 prot = PROT_READ;
 693                 o_flags = O_RDONLY;
 694                 acc_mode = S_IRUGO;
 695         } else {
 696                 prot = PROT_READ | PROT_WRITE;
 697                 o_flags = O_RDWR;
 698                 acc_mode = S_IRUGO | S_IWUGO;
 699         }
 700         if (shmflg & SHM_EXEC) {
 701                 prot |= PROT_EXEC;
 702                 acc_mode |= S_IXUGO;
 703         }
 704
 705         /*
 706          * We cannot rely on the fs check since SYSV IPC does have an
 707          * additional creator id...
 708          */
 709         shp = shm_lock(shmid);
 710         if(shp == NULL) {
 711                 err = -EINVAL;
 712                 goto out;
 713         }
 714         err = shm_checkid(shp,shmid);
 715         if (err) {
 716                 shm_unlock(shp);
 717                 goto out;
 718         }
 719         if (ipcperms(&shp->shm_perm, acc_mode)) {
 720                 shm_unlock(shp);
 721                 err = -EACCES;
 722                 goto out;
 723         }
 724
 725         err = security_shm_shmat(shp, shmaddr, shmflg);
 726         if (err) {
 727                 shm_unlock(shp);
 728                 return err;
 729         }
 730
 731         file = shp->shm_file;
 732         size = i_size_read(file->f_dentry->d_inode);
 733         shp->shm_nattch++;
 734         shm_unlock(shp);
 735
 736         down_write(&current->mm->mmap_sem);
 737         if (addr && !(shmflg & SHM_REMAP)) {
 738                 user_addr = ERR_PTR(-EINVAL);
 739                 if (find_vma_intersection(current->mm, addr, addr + size))
 740                         goto invalid;
 741                 /*
 742                  * If shm segment goes below stack, make sure there is some
 743                  * space left for the stack to grow (at least 4 pages).
 744                  */
 745                 if (addr < current->mm->start_stack &&
 746                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 747                         goto invalid;
 748         }
 749
 750         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 751
 752 invalid:
 753         up_write(&current->mm->mmap_sem);
 754
 755         down (&shm_ids.sem);
 756         if(!(shp = shm_lock(shmid)))
 757                 BUG();
 758         shp->shm_nattch--;
 759         if(shp->shm_nattch == 0 &&
 760            shp->shm_flags & SHM_DEST)
 761                 shm_destroy (shp);
 762         else
 763                 shm_unlock(shp);
 764         up (&shm_ids.sem);
 765
 766         *raddr = (unsigned long) user_addr;
 767         err = 0;
 768         if (IS_ERR(user_addr))
 769                 err = PTR_ERR(user_addr);
 770 out:
 771         return err;
 772 }
 773
 774 /*
 775  * detach and kill segment if marked destroyed.
 776  * The work is done in shm_close.
 777  */
 778 asmlinkage long sys_shmdt(char __user *shmaddr)
 779 {
 780         struct mm_struct *mm = current->mm;
 781         struct vm_area_struct *vma, *next;
 782         unsigned long addr = (unsigned long)shmaddr;
 783         loff_t size = 0;
 784         int retval = -EINVAL;
 785
 786         down_write(&mm->mmap_sem);
 787
 788         /*
 789          * This function tries to be smart and unmap shm segments that
 790          * were modified by partial mlock or munmap calls:
 791          * - It first determines the size of the shm segment that should be
 792          *   unmapped: It searches for a vma that is backed by shm and that
 793          *   started at address shmaddr. It records it's size and then unmaps
 794          *   it.
 795          * - Then it unmaps all shm vmas that started at shmaddr and that
 796          *   are within the initially determined size.
 797          * Errors from do_munmap are ignored: the function only fails if
 798          * it's called with invalid parameters or if it's called to unmap
 799          * a part of a vma. Both calls in this function are for full vmas,
 800          * the parameters are directly copied from the vma itself and always
 801          * valid - therefore do_munmap cannot fail. (famous last words?)
 802          */
 803         /*
 804          * If it had been mremap()'d, the starting address would not
 805          * match the usual checks anyway. So assume all vma's are
 806          * above the starting address given.
 807          */
 808         vma = find_vma(mm, addr);
 809
 810         while (vma) {
 811                 next = vma->vm_next;
 812
 813                 /*
 814                  * Check if the starting address would match, i.e. it's
 815                  * a fragment created by mprotect() and/or munmap(), or it
 816                  * otherwise it starts at this address with no hassles.
 817                  */
 818                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 819                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 820
 821
 822                         size = vma->vm_file->f_dentry->d_inode->i_size;
 823                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 824                         /*
 825                          * We discovered the size of the shm segment, so
 826                          * break out of here and fall through to the next
 827                          * loop that uses the size information to stop
 828                          * searching for matching vma's.
 829                          */
 830                         retval = 0;
 831                         vma = next;
 832                         break;
 833                 }
 834                 vma = next;
 835         }
 836
 837         /*
 838          * We need look no further than the maximum address a fragment
 839          * could possibly have landed at. Also cast things to loff_t to
 840          * prevent overflows and make comparisions vs. equal-width types.
 841          */
 842         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 843                 next = vma->vm_next;
 844
 845                 /* finding a matching vma now does not alter retval */
 846                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 847                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 848
 849                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 850                 vma = next;
 851         }
 852
 853         up_write(&mm->mmap_sem);
 854         return retval;
 855 }
 856
 857 #ifdef CONFIG_PROC_FS
 858 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 859 {
 860         off_t pos = 0;
 861         off_t begin = 0;
 862         int i, len = 0;
 863
 864         down(&shm_ids.sem);
 865         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 866
 867         for(i = 0; i <= shm_ids.max_id; i++) {
 868                 struct shmid_kernel* shp;
 869
 870                 shp = shm_lock(i);
 871                 if(shp!=NULL) {
 872 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 873 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 874                         char *format;
 875
 876                         if (sizeof(size_t) <= sizeof(int))
 877                                 format = SMALL_STRING;
 878                         else
 879                                 format = BIG_STRING;
 880                         len += sprintf(buffer + len, format,
 881                                 shp->shm_perm.key,
 882                                 shm_buildid(i, shp->shm_perm.seq),
 883                                 shp->shm_flags,
 884                                 shp->shm_segsz,
 885                                 shp->shm_cprid,
 886                                 shp->shm_lprid,
 887                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 888                                 shp->shm_perm.uid,
 889                                 shp->shm_perm.gid,
 890                                 shp->shm_perm.cuid,
 891                                 shp->shm_perm.cgid,
 892                                 shp->shm_atim,
 893                                 shp->shm_dtim,
 894                                 shp->shm_ctim);
 895                         shm_unlock(shp);
 896
 897                         pos += len;
 898                         if(pos < offset) {
 899                                 len = 0;
 900                                 begin = pos;
 901                         }
 902                         if(pos > offset + length)
 903                                 goto done;
 904                 }
 905         }
 906         *eof = 1;
 907 done:
 908         up(&shm_ids.sem);
 909         *start = buffer + (offset - begin);
 910         len -= (offset - begin);
 911         if(len > length)
 912                 len = length;
 913         if(len < 0)
 914                 len = 0;
 915         return len;
 916 }
 917 #endif