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