ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/slab.h>
  20 #include <linux/mm.h>
  21 #include <linux/hugetlb.h>
  22 #include <linux/shm.h>
  23 #include <linux/init.h>
  24 #include <linux/file.h>
  25 #include <linux/mman.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/shmem_fs.h>
  28 #include <linux/security.h>
  29 #include <asm/uaccess.h>
  30
  31 #include "util.h"
  32
  33 #define shm_flags       shm_perm.mode
  34
  35 static struct file_operations shm_file_operations;
  36 static struct vm_operations_struct shm_vm_ops;
  37
  38 static struct ipc_ids shm_ids;
  39
  40 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  41 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  42 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  43 #define shm_buildid(id, seq) \
  44         ipc_buildid(&shm_ids, id, seq)
  45
  46 static int newseg (key_t key, int shmflg, size_t size);
  47 static void shm_open (struct vm_area_struct *shmd);
  48 static void shm_close (struct vm_area_struct *shmd);
  49 #ifdef CONFIG_PROC_FS
  50 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  51 #endif
  52
  53 size_t  shm_ctlmax = SHMMAX;
  54 size_t  shm_ctlall = SHMALL;
  55 int     shm_ctlmni = SHMMNI;
  56
  57 static int shm_tot; /* total number of shared memory pages */
  58
  59 void __init shm_init (void)
  60 {
  61         ipc_init_ids(&shm_ids, 1);
  62 #ifdef CONFIG_PROC_FS
  63         create_proc_read_entry("sysvipc/shm", 0, 0, sysvipc_shm_read_proc, NULL);
  64 #endif
  65 }
  66
  67 static inline int shm_checkid(struct shmid_kernel *s, int id)
  68 {
  69         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  70                 return -EIDRM;
  71         return 0;
  72 }
  73
  74 static inline struct shmid_kernel *shm_rmid(int id)
  75 {
  76         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  77 }
  78
  79 static inline int shm_addid(struct shmid_kernel *shp)
  80 {
  81         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni+1);
  82 }
  83
  84
  85
  86 static inline void shm_inc (int id) {
  87         struct shmid_kernel *shp;
  88
  89         if(!(shp = shm_lock(id)))
  90                 BUG();
  91         shp->shm_atim = get_seconds();
  92         shp->shm_lprid = current->tgid;
  93         shp->shm_nattch++;
  94         shm_unlock(shp);
  95 }
  96
  97 /* This is called by fork, once for every shm attach. */
  98 static void shm_open (struct vm_area_struct *shmd)
  99 {
 100         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 101 }
 102
 103 /*
 104  * shm_destroy - free the struct shmid_kernel
 105  *
 106  * @shp: struct to free
 107  *
 108  * It has to be called with shp and shm_ids.sem locked,
 109  * but returns with shp unlocked and freed.
 110  */
 111 static void shm_destroy (struct shmid_kernel *shp)
 112 {
 113         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 114         shm_rmid (shp->id);
 115         shm_unlock(shp);
 116         if (!is_file_hugepages(shp->shm_file))
 117                 shmem_lock(shp->shm_file, 0);
 118         fput (shp->shm_file);
 119         security_shm_free(shp);
 120         ipc_rcu_free(shp, sizeof(struct shmid_kernel));
 121 }
 122
 123 /*
 124  * remove the attach descriptor shmd.
 125  * free memory for segment if it is marked destroyed.
 126  * The descriptor has already been removed from the current->mm->mmap list
 127  * and will later be kfree()d.
 128  */
 129 static void shm_close (struct vm_area_struct *shmd)
 130 {
 131         struct file * file = shmd->vm_file;
 132         int id = file->f_dentry->d_inode->i_ino;
 133         struct shmid_kernel *shp;
 134
 135         down (&shm_ids.sem);
 136         /* remove from the list of attaches of the shm segment */
 137         if(!(shp = shm_lock(id)))
 138                 BUG();
 139         shp->shm_lprid = current->tgid;
 140         shp->shm_dtim = get_seconds();
 141         shp->shm_nattch--;
 142         if(shp->shm_nattch == 0 &&
 143            shp->shm_flags & SHM_DEST)
 144                 shm_destroy (shp);
 145         else
 146                 shm_unlock(shp);
 147         up (&shm_ids.sem);
 148 }
 149
 150 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 151 {
 152         file_accessed(file);
 153         vma->vm_ops = &shm_vm_ops;
 154         shm_inc(file->f_dentry->d_inode->i_ino);
 155         return 0;
 156 }
 157
 158 static struct file_operations shm_file_operations = {
 159         .mmap   = shm_mmap
 160 };
 161
 162 static struct vm_operations_struct shm_vm_ops = {
 163         .open   = shm_open,     /* callback for a new vm-area open */
 164         .close  = shm_close,    /* callback for when the vm-area is released */
 165         .nopage = shmem_nopage,
 166 };
 167
 168 static int newseg (key_t key, int shmflg, size_t size)
 169 {
 170         int error;
 171         struct shmid_kernel *shp;
 172         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 173         struct file * file;
 174         char name[13];
 175         int id;
 176
 177         if (size < SHMMIN || size > shm_ctlmax)
 178                 return -EINVAL;
 179
 180         if (shm_tot + numpages >= shm_ctlall)
 181                 return -ENOSPC;
 182
 183         shp = ipc_rcu_alloc(sizeof(*shp));
 184         if (!shp)
 185                 return -ENOMEM;
 186
 187         shp->shm_perm.key = key;
 188         shp->shm_perm.xid = current->xid;
 189         shp->shm_flags = (shmflg & S_IRWXUGO);
 190
 191         shp->shm_perm.security = NULL;
 192         error = security_shm_alloc(shp);
 193         if (error) {
 194                 ipc_rcu_free(shp, sizeof(*shp));
 195                 return error;
 196         }
 197
 198         if (shmflg & SHM_HUGETLB)
 199                 file = hugetlb_zero_setup(size);
 200         else {
 201                 sprintf (name, "SYSV%08x", key);
 202                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 203         }
 204         error = PTR_ERR(file);
 205         if (IS_ERR(file))
 206                 goto no_file;
 207
 208         error = -ENOSPC;
 209         id = shm_addid(shp);
 210         if(id == -1)
 211                 goto no_id;
 212
 213         shp->shm_cprid = current->tgid;
 214         shp->shm_lprid = 0;
 215         shp->shm_atim = shp->shm_dtim = 0;
 216         shp->shm_ctim = get_seconds();
 217         shp->shm_segsz = size;
 218         shp->shm_nattch = 0;
 219         shp->id = shm_buildid(id,shp->shm_perm.seq);
 220         shp->shm_file = file;
 221         file->f_dentry->d_inode->i_ino = shp->id;
 222         if (shmflg & SHM_HUGETLB)
 223                 set_file_hugepages(file);
 224         else
 225                 file->f_op = &shm_file_operations;
 226         shm_tot += numpages;
 227         shm_unlock(shp);
 228         return shp->id;
 229
 230 no_id:
 231         fput(file);
 232 no_file:
 233         security_shm_free(shp);
 234         ipc_rcu_free(shp, sizeof(*shp));
 235         return error;
 236 }
 237
 238 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 239 {
 240         struct shmid_kernel *shp;
 241         int err, id = 0;
 242
 243         down(&shm_ids.sem);
 244         if (key == IPC_PRIVATE) {
 245                 err = newseg(key, shmflg, size);
 246         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 247                 if (!(shmflg & IPC_CREAT))
 248                         err = -ENOENT;
 249                 else
 250                         err = newseg(key, shmflg, size);
 251         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 252                 err = -EEXIST;
 253         } else {
 254                 shp = shm_lock(id);
 255                 if(shp==NULL)
 256                         BUG();
 257                 if (shp->shm_segsz < size)
 258                         err = -EINVAL;
 259                 else if (ipcperms(&shp->shm_perm, shmflg))
 260                         err = -EACCES;
 261                 else {
 262                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 263                         err = security_shm_associate(shp, shmflg);
 264                         if (!err)
 265                                 err = shmid;
 266                 }
 267                 shm_unlock(shp);
 268         }
 269         up(&shm_ids.sem);
 270
 271         return err;
 272 }
 273
 274 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 275 {
 276         switch(version) {
 277         case IPC_64:
 278                 return copy_to_user(buf, in, sizeof(*in));
 279         case IPC_OLD:
 280             {
 281                 struct shmid_ds out;
 282
 283                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 284                 out.shm_segsz   = in->shm_segsz;
 285                 out.shm_atime   = in->shm_atime;
 286                 out.shm_dtime   = in->shm_dtime;
 287                 out.shm_ctime   = in->shm_ctime;
 288                 out.shm_cpid    = in->shm_cpid;
 289                 out.shm_lpid    = in->shm_lpid;
 290                 out.shm_nattch  = in->shm_nattch;
 291
 292                 return copy_to_user(buf, &out, sizeof(out));
 293             }
 294         default:
 295                 return -EINVAL;
 296         }
 297 }
 298
 299 struct shm_setbuf {
 300         uid_t   uid;
 301         gid_t   gid;
 302         mode_t  mode;
 303 };
 304
 305 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 306 {
 307         switch(version) {
 308         case IPC_64:
 309             {
 310                 struct shmid64_ds tbuf;
 311
 312                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 313                         return -EFAULT;
 314
 315                 out->uid        = tbuf.shm_perm.uid;
 316                 out->gid        = tbuf.shm_perm.gid;
 317                 out->mode       = tbuf.shm_flags;
 318
 319                 return 0;
 320             }
 321         case IPC_OLD:
 322             {
 323                 struct shmid_ds tbuf_old;
 324
 325                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 326                         return -EFAULT;
 327
 328                 out->uid        = tbuf_old.shm_perm.uid;
 329                 out->gid        = tbuf_old.shm_perm.gid;
 330                 out->mode       = tbuf_old.shm_flags;
 331
 332                 return 0;
 333             }
 334         default:
 335                 return -EINVAL;
 336         }
 337 }
 338
 339 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 340 {
 341         switch(version) {
 342         case IPC_64:
 343                 return copy_to_user(buf, in, sizeof(*in));
 344         case IPC_OLD:
 345             {
 346                 struct shminfo out;
 347
 348                 if(in->shmmax > INT_MAX)
 349                         out.shmmax = INT_MAX;
 350                 else
 351                         out.shmmax = (int)in->shmmax;
 352
 353                 out.shmmin      = in->shmmin;
 354                 out.shmmni      = in->shmmni;
 355                 out.shmseg      = in->shmseg;
 356                 out.shmall      = in->shmall;
 357
 358                 return copy_to_user(buf, &out, sizeof(out));
 359             }
 360         default:
 361                 return -EINVAL;
 362         }
 363 }
 364
 365 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 366 {
 367         int i;
 368
 369         *rss = 0;
 370         *swp = 0;
 371
 372         for (i = 0; i <= shm_ids.max_id; i++) {
 373                 struct shmid_kernel *shp;
 374                 struct inode *inode;
 375
 376                 shp = shm_get(i);
 377                 if(!shp)
 378                         continue;
 379
 380                 inode = shp->shm_file->f_dentry->d_inode;
 381
 382                 if (is_file_hugepages(shp->shm_file)) {
 383                         struct address_space *mapping = inode->i_mapping;
 384                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 385                 } else {
 386                         struct shmem_inode_info *info = SHMEM_I(inode);
 387                         spin_lock(&info->lock);
 388                         *rss += inode->i_mapping->nrpages;
 389                         *swp += info->swapped;
 390                         spin_unlock(&info->lock);
 391                 }
 392         }
 393 }
 394
 395 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 396 {
 397         struct shm_setbuf setbuf;
 398         struct shmid_kernel *shp;
 399         int err, version;
 400
 401         if (cmd < 0 || shmid < 0) {
 402                 err = -EINVAL;
 403                 goto out;
 404         }
 405
 406         version = ipc_parse_version(&cmd);
 407
 408         switch (cmd) { /* replace with proc interface ? */
 409         case IPC_INFO:
 410         {
 411                 struct shminfo64 shminfo;
 412
 413                 err = security_shm_shmctl(NULL, cmd);
 414                 if (err)
 415                         return err;
 416
 417                 memset(&shminfo,0,sizeof(shminfo));
 418                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 419                 shminfo.shmmax = shm_ctlmax;
 420                 shminfo.shmall = shm_ctlall;
 421
 422                 shminfo.shmmin = SHMMIN;
 423                 if(copy_shminfo_to_user (buf, &shminfo, version))
 424                         return -EFAULT;
 425                 /* reading a integer is always atomic */
 426                 err= shm_ids.max_id;
 427                 if(err<0)
 428                         err = 0;
 429                 goto out;
 430         }
 431         case SHM_INFO:
 432         {
 433                 struct shm_info shm_info;
 434
 435                 err = security_shm_shmctl(NULL, cmd);
 436                 if (err)
 437                         return err;
 438
 439                 memset(&shm_info,0,sizeof(shm_info));
 440                 down(&shm_ids.sem);
 441                 shm_info.used_ids = shm_ids.in_use;
 442                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 443                 shm_info.shm_tot = shm_tot;
 444                 shm_info.swap_attempts = 0;
 445                 shm_info.swap_successes = 0;
 446                 err = shm_ids.max_id;
 447                 up(&shm_ids.sem);
 448                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 449                         err = -EFAULT;
 450                         goto out;
 451                 }
 452
 453                 err = err < 0 ? 0 : err;
 454                 goto out;
 455         }
 456         case SHM_STAT:
 457         case IPC_STAT:
 458         {
 459                 struct shmid64_ds tbuf;
 460                 int result;
 461                 memset(&tbuf, 0, sizeof(tbuf));
 462                 shp = shm_lock(shmid);
 463                 if(shp==NULL) {
 464                         err = -EINVAL;
 465                         goto out;
 466                 } else if(cmd==SHM_STAT) {
 467                         err = -EINVAL;
 468                         if (shmid > shm_ids.max_id)
 469                                 goto out_unlock;
 470                         result = shm_buildid(shmid, shp->shm_perm.seq);
 471                 } else {
 472                         err = shm_checkid(shp,shmid);
 473                         if(err)
 474                                 goto out_unlock;
 475                         result = 0;
 476                 }
 477                 err=-EACCES;
 478                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 479                         goto out_unlock;
 480                 err = security_shm_shmctl(shp, cmd);
 481                 if (err)
 482                         goto out_unlock;
 483                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 484                 tbuf.shm_segsz  = shp->shm_segsz;
 485                 tbuf.shm_atime  = shp->shm_atim;
 486                 tbuf.shm_dtime  = shp->shm_dtim;
 487                 tbuf.shm_ctime  = shp->shm_ctim;
 488                 tbuf.shm_cpid   = shp->shm_cprid;
 489                 tbuf.shm_lpid   = shp->shm_lprid;
 490                 if (!is_file_hugepages(shp->shm_file))
 491                         tbuf.shm_nattch = shp->shm_nattch;
 492                 else
 493                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 494                 shm_unlock(shp);
 495                 if(copy_shmid_to_user (buf, &tbuf, version))
 496                         err = -EFAULT;
 497                 else
 498                         err = result;
 499                 goto out;
 500         }
 501         case SHM_LOCK:
 502         case SHM_UNLOCK:
 503         {
 504 /* Allow superuser to lock segment in memory */
 505 /* Should the pages be faulted in here or leave it to user? */
 506 /* need to determine interaction with current->swappable */
 507                 if (!capable(CAP_IPC_LOCK)) {
 508                         err = -EPERM;
 509                         goto out;
 510                 }
 511
 512                 shp = shm_lock(shmid);
 513                 if(shp==NULL) {
 514                         err = -EINVAL;
 515                         goto out;
 516                 }
 517                 err = shm_checkid(shp,shmid);
 518                 if(err)
 519                         goto out_unlock;
 520
 521                 err = security_shm_shmctl(shp, cmd);
 522                 if (err)
 523                         goto out_unlock;
 524
 525                 if(cmd==SHM_LOCK) {
 526                         if (!is_file_hugepages(shp->shm_file))
 527                                 shmem_lock(shp->shm_file, 1);
 528                         shp->shm_flags |= SHM_LOCKED;
 529                 } else {
 530                         if (!is_file_hugepages(shp->shm_file))
 531                                 shmem_lock(shp->shm_file, 0);
 532                         shp->shm_flags &= ~SHM_LOCKED;
 533                 }
 534                 shm_unlock(shp);
 535                 goto out;
 536         }
 537         case IPC_RMID:
 538         {
 539                 /*
 540                  *      We cannot simply remove the file. The SVID states
 541                  *      that the block remains until the last person
 542                  *      detaches from it, then is deleted. A shmat() on
 543                  *      an RMID segment is legal in older Linux and if
 544                  *      we change it apps break...
 545                  *
 546                  *      Instead we set a destroyed flag, and then blow
 547                  *      the name away when the usage hits zero.
 548                  */
 549                 down(&shm_ids.sem);
 550                 shp = shm_lock(shmid);
 551                 err = -EINVAL;
 552                 if (shp == NULL)
 553                         goto out_up;
 554                 err = shm_checkid(shp, shmid);
 555                 if(err)
 556                         goto out_unlock_up;
 557
 558                 if (current->euid != shp->shm_perm.uid &&
 559                     current->euid != shp->shm_perm.cuid &&
 560                     !capable(CAP_SYS_ADMIN)) {
 561                         err=-EPERM;
 562                         goto out_unlock_up;
 563                 }
 564
 565                 err = security_shm_shmctl(shp, cmd);
 566                 if (err)
 567                         goto out_unlock_up;
 568
 569                 if (shp->shm_nattch){
 570                         shp->shm_flags |= SHM_DEST;
 571                         /* Do not find it any more */
 572                         shp->shm_perm.key = IPC_PRIVATE;
 573                         shm_unlock(shp);
 574                 } else
 575                         shm_destroy (shp);
 576                 up(&shm_ids.sem);
 577                 goto out;
 578         }
 579
 580         case IPC_SET:
 581         {
 582                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 583                         err = -EFAULT;
 584                         goto out;
 585                 }
 586                 down(&shm_ids.sem);
 587                 shp = shm_lock(shmid);
 588                 err=-EINVAL;
 589                 if(shp==NULL)
 590                         goto out_up;
 591                 err = shm_checkid(shp,shmid);
 592                 if(err)
 593                         goto out_unlock_up;
 594                 err=-EPERM;
 595                 if (current->euid != shp->shm_perm.uid &&
 596                     current->euid != shp->shm_perm.cuid &&
 597                     !capable(CAP_SYS_ADMIN)) {
 598                         goto out_unlock_up;
 599                 }
 600
 601                 err = security_shm_shmctl(shp, cmd);
 602                 if (err)
 603                         goto out_unlock_up;
 604
 605                 shp->shm_perm.uid = setbuf.uid;
 606                 shp->shm_perm.gid = setbuf.gid;
 607                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 608                         | (setbuf.mode & S_IRWXUGO);
 609                 shp->shm_ctim = get_seconds();
 610                 break;
 611         }
 612
 613         default:
 614                 err = -EINVAL;
 615                 goto out;
 616         }
 617
 618         err = 0;
 619 out_unlock_up:
 620         shm_unlock(shp);
 621 out_up:
 622         up(&shm_ids.sem);
 623         goto out;
 624 out_unlock:
 625         shm_unlock(shp);
 626 out:
 627         return err;
 628 }
 629
 630 /*
 631  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 632  *
 633  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 634  * "raddr" thing points to kernel space, and there has to be a wrapper around
 635  * this.
 636  */
 637 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 638 {
 639         struct shmid_kernel *shp;
 640         unsigned long addr;
 641         unsigned long size;
 642         struct file * file;
 643         int    err;
 644         unsigned long flags;
 645         unsigned long prot;
 646         unsigned long o_flags;
 647         int acc_mode;
 648         void *user_addr;
 649
 650         if (shmid < 0) {
 651                 err = -EINVAL;
 652                 goto out;
 653         } else if ((addr = (ulong)shmaddr)) {
 654                 if (addr & (SHMLBA-1)) {
 655                         if (shmflg & SHM_RND)
 656                                 addr &= ~(SHMLBA-1);       /* round down */
 657                         else
 658 #ifndef __ARCH_FORCE_SHMLBA
 659                                 if (addr & ~PAGE_MASK)
 660 #endif
 661                                         return -EINVAL;
 662                 }
 663                 flags = MAP_SHARED | MAP_FIXED;
 664         } else {
 665                 if ((shmflg & SHM_REMAP))
 666                         return -EINVAL;
 667
 668                 flags = MAP_SHARED;
 669         }
 670
 671         if (shmflg & SHM_RDONLY) {
 672                 prot = PROT_READ;
 673                 o_flags = O_RDONLY;
 674                 acc_mode = S_IRUGO;
 675         } else {
 676                 prot = PROT_READ | PROT_WRITE;
 677                 o_flags = O_RDWR;
 678                 acc_mode = S_IRUGO | S_IWUGO;
 679         }
 680
 681         /*
 682          * We cannot rely on the fs check since SYSV IPC does have an
 683          * additional creator id...
 684          */
 685         shp = shm_lock(shmid);
 686         if(shp == NULL) {
 687                 err = -EINVAL;
 688                 goto out;
 689         }
 690         err = shm_checkid(shp,shmid);
 691         if (err) {
 692                 shm_unlock(shp);
 693                 goto out;
 694         }
 695         if (ipcperms(&shp->shm_perm, acc_mode)) {
 696                 shm_unlock(shp);
 697                 err = -EACCES;
 698                 goto out;
 699         }
 700
 701         err = security_shm_shmat(shp, shmaddr, shmflg);
 702         if (err) {
 703                 shm_unlock(shp);
 704                 return err;
 705         }
 706
 707         file = shp->shm_file;
 708         size = i_size_read(file->f_dentry->d_inode);
 709         shp->shm_nattch++;
 710         shm_unlock(shp);
 711
 712         down_write(&current->mm->mmap_sem);
 713         if (addr && !(shmflg & SHM_REMAP)) {
 714                 user_addr = ERR_PTR(-EINVAL);
 715                 if (find_vma_intersection(current->mm, addr, addr + size))
 716                         goto invalid;
 717                 /*
 718                  * If shm segment goes below stack, make sure there is some
 719                  * space left for the stack to grow (at least 4 pages).
 720                  */
 721                 if (addr < current->mm->start_stack &&
 722                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 723                         goto invalid;
 724         }
 725
 726         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 727
 728 invalid:
 729         up_write(&current->mm->mmap_sem);
 730
 731         down (&shm_ids.sem);
 732         if(!(shp = shm_lock(shmid)))
 733                 BUG();
 734         shp->shm_nattch--;
 735         if(shp->shm_nattch == 0 &&
 736            shp->shm_flags & SHM_DEST)
 737                 shm_destroy (shp);
 738         else
 739                 shm_unlock(shp);
 740         up (&shm_ids.sem);
 741
 742         *raddr = (unsigned long) user_addr;
 743         err = 0;
 744         if (IS_ERR(user_addr))
 745                 err = PTR_ERR(user_addr);
 746 out:
 747         return err;
 748 }
 749
 750 /*
 751  * detach and kill segment if marked destroyed.
 752  * The work is done in shm_close.
 753  */
 754 asmlinkage long sys_shmdt(char __user *shmaddr)
 755 {
 756         struct mm_struct *mm = current->mm;
 757         struct vm_area_struct *vma, *next;
 758         unsigned long addr = (unsigned long)shmaddr;
 759         loff_t size = 0;
 760         int retval = -EINVAL;
 761
 762         down_write(&mm->mmap_sem);
 763
 764         /*
 765          * This function tries to be smart and unmap shm segments that
 766          * were modified by partial mlock or munmap calls:
 767          * - It first determines the size of the shm segment that should be
 768          *   unmapped: It searches for a vma that is backed by shm and that
 769          *   started at address shmaddr. It records it's size and then unmaps
 770          *   it.
 771          * - Then it unmaps all shm vmas that started at shmaddr and that
 772          *   are within the initially determined size.
 773          * Errors from do_munmap are ignored: the function only fails if
 774          * it's called with invalid parameters or if it's called to unmap
 775          * a part of a vma. Both calls in this function are for full vmas,
 776          * the parameters are directly copied from the vma itself and always
 777          * valid - therefore do_munmap cannot fail. (famous last words?)
 778          */
 779         /*
 780          * If it had been mremap()'d, the starting address would not
 781          * match the usual checks anyway. So assume all vma's are
 782          * above the starting address given.
 783          */
 784         vma = find_vma(mm, addr);
 785
 786         while (vma) {
 787                 next = vma->vm_next;
 788
 789                 /*
 790                  * Check if the starting address would match, i.e. it's
 791                  * a fragment created by mprotect() and/or munmap(), or it
 792                  * otherwise it starts at this address with no hassles.
 793                  */
 794                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 795                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 796
 797
 798                         size = vma->vm_file->f_dentry->d_inode->i_size;
 799                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 800                         /*
 801                          * We discovered the size of the shm segment, so
 802                          * break out of here and fall through to the next
 803                          * loop that uses the size information to stop
 804                          * searching for matching vma's.
 805                          */
 806                         retval = 0;
 807                         vma = next;
 808                         break;
 809                 }
 810                 vma = next;
 811         }
 812
 813         /*
 814          * We need look no further than the maximum address a fragment
 815          * could possibly have landed at. Also cast things to loff_t to
 816          * prevent overflows and make comparisions vs. equal-width types.
 817          */
 818         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 819                 next = vma->vm_next;
 820
 821                 /* finding a matching vma now does not alter retval */
 822                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 823                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 824
 825                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 826                 vma = next;
 827         }
 828
 829         up_write(&mm->mmap_sem);
 830         return retval;
 831 }
 832
 833 #ifdef CONFIG_PROC_FS
 834 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 835 {
 836         off_t pos = 0;
 837         off_t begin = 0;
 838         int i, len = 0;
 839
 840         down(&shm_ids.sem);
 841         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 842
 843         for(i = 0; i <= shm_ids.max_id; i++) {
 844                 struct shmid_kernel* shp;
 845
 846                 shp = shm_lock(i);
 847                 if (shp) {
 848 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 849 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 850                         char *format;
 851
 852                         if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
 853                                 shm_unlock(shp);
 854                                 continue;
 855                         }
 856                         if (sizeof(size_t) <= sizeof(int))
 857                                 format = SMALL_STRING;
 858                         else
 859                                 format = BIG_STRING;
 860                         len += sprintf(buffer + len, format,
 861                                 shp->shm_perm.key,
 862                                 shm_buildid(i, shp->shm_perm.seq),
 863                                 shp->shm_flags,
 864                                 shp->shm_segsz,
 865                                 shp->shm_cprid,
 866                                 shp->shm_lprid,
 867                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 868                                 shp->shm_perm.uid,
 869                                 shp->shm_perm.gid,
 870                                 shp->shm_perm.cuid,
 871                                 shp->shm_perm.cgid,
 872                                 shp->shm_atim,
 873                                 shp->shm_dtim,
 874                                 shp->shm_ctim);
 875                         shm_unlock(shp);
 876
 877                         pos += len;
 878                         if(pos < offset) {
 879                                 len = 0;
 880                                 begin = pos;
 881                         }
 882                         if(pos > offset + length)
 883                                 goto done;
 884                 }
 885         }
 886         *eof = 1;
 887 done:
 888         up(&shm_ids.sem);
 889         *start = buffer + (offset - begin);
 890         len -= (offset - begin);
 891         if(len > length)
 892                 len = length;
 893         if(len < 0)
 894                 len = 0;
 895         return len;
 896 }
 897 #endif