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