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  * support for audit of ipc object properties and permission changes
  17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  18  *
  19  * namespaces support
  20  * OpenVZ, SWsoft Inc.
  21  * Pavel Emelianov <xemul@openvz.org>
  22  */
  23
  24 #include <linux/slab.h>
  25 #include <linux/mm.h>
  26 #include <linux/hugetlb.h>
  27 #include <linux/shm.h>
  28 #include <linux/init.h>
  29 #include <linux/file.h>
  30 #include <linux/mman.h>
  31 #include <linux/shmem_fs.h>
  32 #include <linux/security.h>
  33 #include <linux/syscalls.h>
  34 #include <linux/audit.h>
  35 #include <linux/capability.h>
  36 #include <linux/ptrace.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/mutex.h>
  39 #include <linux/nsproxy.h>
  40
  41 #include <asm/uaccess.h>
  42
  43 #include "util.h"
  44
  45 static struct file_operations shm_file_operations;
  46 static struct vm_operations_struct shm_vm_ops;
  47
  48 static struct ipc_ids init_shm_ids;
  49
  50 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
  51
  52 #define shm_lock(ns, id)                \
  53         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
  54 #define shm_unlock(shp)                 \
  55         ipc_unlock(&(shp)->shm_perm)
  56 #define shm_get(ns, id)                 \
  57         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
  58 #define shm_buildid(ns, id, seq)        \
  59         ipc_buildid(&shm_ids(ns), id, seq)
  60
  61 static int newseg (struct ipc_namespace *ns, key_t key,
  62                 int shmflg, size_t size);
  63 static void shm_open (struct vm_area_struct *shmd);
  64 static void shm_close (struct vm_area_struct *shmd);
  65 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  66 #ifdef CONFIG_PROC_FS
  67 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  68 #endif
  69
  70 static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  71 {
  72         ns->ids[IPC_SHM_IDS] = ids;
  73         ns->shm_ctlmax = SHMMAX;
  74         ns->shm_ctlall = SHMALL;
  75         ns->shm_ctlmni = SHMMNI;
  76         ns->shm_tot = 0;
  77         ipc_init_ids(ids, 1);
  78 }
  79
  80 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  81 {
  82         if (shp->shm_nattch){
  83                 shp->shm_perm.mode |= SHM_DEST;
  84                 /* Do not find it any more */
  85                 shp->shm_perm.key = IPC_PRIVATE;
  86                 shm_unlock(shp);
  87         } else
  88                 shm_destroy(ns, shp);
  89 }
  90
  91 #ifdef CONFIG_IPC_NS
  92 int shm_init_ns(struct ipc_namespace *ns)
  93 {
  94         struct ipc_ids *ids;
  95
  96         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
  97         if (ids == NULL)
  98                 return -ENOMEM;
  99
 100         __shm_init_ns(ns, ids);
 101         return 0;
 102 }
 103
 104 void shm_exit_ns(struct ipc_namespace *ns)
 105 {
 106         int i;
 107         struct shmid_kernel *shp;
 108
 109         mutex_lock(&shm_ids(ns).mutex);
 110         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 111                 shp = shm_lock(ns, i);
 112                 if (shp == NULL)
 113                         continue;
 114
 115                 do_shm_rmid(ns, shp);
 116         }
 117         mutex_unlock(&shm_ids(ns).mutex);
 118
 119         kfree(ns->ids[IPC_SHM_IDS]);
 120         ns->ids[IPC_SHM_IDS] = NULL;
 121 }
 122 #endif
 123
 124 void __init shm_init (void)
 125 {
 126         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 127         ipc_init_proc_interface("sysvipc/shm",
 128                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 129                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 130 }
 131
 132 static inline int shm_checkid(struct ipc_namespace *ns,
 133                 struct shmid_kernel *s, int id)
 134 {
 135         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 136                 return -EIDRM;
 137         return 0;
 138 }
 139
 140 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
 141 {
 142         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
 143 }
 144
 145 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 146 {
 147         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 148 }
 149
 150
 151
 152 static inline void shm_inc(struct ipc_namespace *ns, int id)
 153 {
 154         struct shmid_kernel *shp;
 155
 156         shp = shm_lock(ns, id);
 157         BUG_ON(!shp);
 158         shp->shm_atim = get_seconds();
 159         shp->shm_lprid = current->tgid;
 160         shp->shm_nattch++;
 161         shm_unlock(shp);
 162 }
 163
 164 #define shm_file_ns(file) (*((struct ipc_namespace **)&(file)->private_data))
 165
 166 /* This is called by fork, once for every shm attach. */
 167 static void shm_open(struct vm_area_struct *shmd)
 168 {
 169         shm_inc(shm_file_ns(shmd->vm_file),
 170                         shmd->vm_file->f_dentry->d_inode->i_ino);
 171 }
 172
 173 /*
 174  * shm_destroy - free the struct shmid_kernel
 175  *
 176  * @shp: struct to free
 177  *
 178  * It has to be called with shp and shm_ids.mutex locked,
 179  * but returns with shp unlocked and freed.
 180  */
 181 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 182 {
 183         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 184         shm_rmid(ns, shp->id);
 185         shm_unlock(shp);
 186         if (!is_file_hugepages(shp->shm_file))
 187                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 188         else
 189                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 190                                                 shp->mlock_user);
 191         fput (shp->shm_file);
 192         security_shm_free(shp);
 193         ipc_rcu_putref(shp);
 194 }
 195
 196 /*
 197  * remove the attach descriptor shmd.
 198  * free memory for segment if it is marked destroyed.
 199  * The descriptor has already been removed from the current->mm->mmap list
 200  * and will later be kfree()d.
 201  */
 202 static void shm_close (struct vm_area_struct *shmd)
 203 {
 204         struct file * file = shmd->vm_file;
 205         int id = file->f_dentry->d_inode->i_ino;
 206         struct shmid_kernel *shp;
 207         struct ipc_namespace *ns;
 208
 209         ns = shm_file_ns(file);
 210
 211         mutex_lock(&shm_ids(ns).mutex);
 212         /* remove from the list of attaches of the shm segment */
 213         shp = shm_lock(ns, id);
 214         BUG_ON(!shp);
 215         shp->shm_lprid = current->tgid;
 216         shp->shm_dtim = get_seconds();
 217         shp->shm_nattch--;
 218         if(shp->shm_nattch == 0 &&
 219            shp->shm_perm.mode & SHM_DEST)
 220                 shm_destroy(ns, shp);
 221         else
 222                 shm_unlock(shp);
 223         mutex_unlock(&shm_ids(ns).mutex);
 224 }
 225
 226 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 227 {
 228         int ret;
 229
 230         ret = shmem_mmap(file, vma);
 231         if (ret == 0) {
 232                 vma->vm_ops = &shm_vm_ops;
 233                 if (!(vma->vm_flags & VM_WRITE))
 234                         vma->vm_flags &= ~VM_MAYWRITE;
 235                 shm_inc(shm_file_ns(file), file->f_dentry->d_inode->i_ino);
 236         }
 237
 238         return ret;
 239 }
 240
 241 static int shm_release(struct inode *ino, struct file *file)
 242 {
 243         struct ipc_namespace *ns;
 244
 245         ns = shm_file_ns(file);
 246         put_ipc_ns(ns);
 247         shm_file_ns(file) = NULL;
 248         return 0;
 249 }
 250
 251 static struct file_operations shm_file_operations = {
 252         .mmap           = shm_mmap,
 253         .release        = shm_release,
 254 #ifndef CONFIG_MMU
 255         .get_unmapped_area = shmem_get_unmapped_area,
 256 #endif
 257 };
 258
 259 static struct vm_operations_struct shm_vm_ops = {
 260         .open   = shm_open,     /* callback for a new vm-area open */
 261         .close  = shm_close,    /* callback for when the vm-area is released */
 262         .nopage = shmem_nopage,
 263 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 264         .set_policy = shmem_set_policy,
 265         .get_policy = shmem_get_policy,
 266 #endif
 267 };
 268
 269 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 270 {
 271         int error;
 272         struct shmid_kernel *shp;
 273         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 274         struct file * file;
 275         char name[13];
 276         int id;
 277
 278         if (size < SHMMIN || size > ns->shm_ctlmax)
 279                 return -EINVAL;
 280
 281         if (ns->shm_tot + numpages >= ns->shm_ctlall)
 282                 return -ENOSPC;
 283
 284         shp = ipc_rcu_alloc(sizeof(*shp));
 285         if (!shp)
 286                 return -ENOMEM;
 287
 288         shp->shm_perm.key = key;
 289         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 290         shp->mlock_user = NULL;
 291
 292         shp->shm_perm.security = NULL;
 293         error = security_shm_alloc(shp);
 294         if (error) {
 295                 ipc_rcu_putref(shp);
 296                 return error;
 297         }
 298
 299         if (shmflg & SHM_HUGETLB) {
 300                 /* hugetlb_zero_setup takes care of mlock user accounting */
 301                 file = hugetlb_zero_setup(size);
 302                 shp->mlock_user = current->user;
 303         } else {
 304                 int acctflag = VM_ACCOUNT;
 305                 /*
 306                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 307                  * if it's asked for.
 308                  */
 309                 if  ((shmflg & SHM_NORESERVE) &&
 310                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 311                         acctflag = 0;
 312                 sprintf (name, "SYSV%08x", key);
 313                 file = shmem_file_setup(name, size, acctflag);
 314         }
 315         error = PTR_ERR(file);
 316         if (IS_ERR(file))
 317                 goto no_file;
 318
 319         error = -ENOSPC;
 320         id = shm_addid(ns, shp);
 321         if(id == -1)
 322                 goto no_id;
 323
 324         shp->shm_cprid = current->tgid;
 325         shp->shm_lprid = 0;
 326         shp->shm_atim = shp->shm_dtim = 0;
 327         shp->shm_ctim = get_seconds();
 328         shp->shm_segsz = size;
 329         shp->shm_nattch = 0;
 330         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 331         shp->shm_file = file;
 332         file->f_dentry->d_inode->i_ino = shp->id;
 333
 334         shm_file_ns(file) = get_ipc_ns(ns);
 335
 336         /* Hugetlb ops would have already been assigned. */
 337         if (!(shmflg & SHM_HUGETLB))
 338                 file->f_op = &shm_file_operations;
 339
 340         ns->shm_tot += numpages;
 341         shm_unlock(shp);
 342         return shp->id;
 343
 344 no_id:
 345         fput(file);
 346 no_file:
 347         security_shm_free(shp);
 348         ipc_rcu_putref(shp);
 349         return error;
 350 }
 351
 352 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 353 {
 354         struct shmid_kernel *shp;
 355         int err, id = 0;
 356         struct ipc_namespace *ns;
 357
 358         ns = current->nsproxy->ipc_ns;
 359
 360         mutex_lock(&shm_ids(ns).mutex);
 361         if (key == IPC_PRIVATE) {
 362                 err = newseg(ns, key, shmflg, size);
 363         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 364                 if (!(shmflg & IPC_CREAT))
 365                         err = -ENOENT;
 366                 else
 367                         err = newseg(ns, key, shmflg, size);
 368         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 369                 err = -EEXIST;
 370         } else {
 371                 shp = shm_lock(ns, id);
 372                 BUG_ON(shp==NULL);
 373                 if (shp->shm_segsz < size)
 374                         err = -EINVAL;
 375                 else if (ipcperms(&shp->shm_perm, shmflg))
 376                         err = -EACCES;
 377                 else {
 378                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 379                         err = security_shm_associate(shp, shmflg);
 380                         if (!err)
 381                                 err = shmid;
 382                 }
 383                 shm_unlock(shp);
 384         }
 385         mutex_unlock(&shm_ids(ns).mutex);
 386
 387         return err;
 388 }
 389
 390 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 391 {
 392         switch(version) {
 393         case IPC_64:
 394                 return copy_to_user(buf, in, sizeof(*in));
 395         case IPC_OLD:
 396             {
 397                 struct shmid_ds out;
 398
 399                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 400                 out.shm_segsz   = in->shm_segsz;
 401                 out.shm_atime   = in->shm_atime;
 402                 out.shm_dtime   = in->shm_dtime;
 403                 out.shm_ctime   = in->shm_ctime;
 404                 out.shm_cpid    = in->shm_cpid;
 405                 out.shm_lpid    = in->shm_lpid;
 406                 out.shm_nattch  = in->shm_nattch;
 407
 408                 return copy_to_user(buf, &out, sizeof(out));
 409             }
 410         default:
 411                 return -EINVAL;
 412         }
 413 }
 414
 415 struct shm_setbuf {
 416         uid_t   uid;
 417         gid_t   gid;
 418         mode_t  mode;
 419 };
 420
 421 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 422 {
 423         switch(version) {
 424         case IPC_64:
 425             {
 426                 struct shmid64_ds tbuf;
 427
 428                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 429                         return -EFAULT;
 430
 431                 out->uid        = tbuf.shm_perm.uid;
 432                 out->gid        = tbuf.shm_perm.gid;
 433                 out->mode       = tbuf.shm_perm.mode;
 434
 435                 return 0;
 436             }
 437         case IPC_OLD:
 438             {
 439                 struct shmid_ds tbuf_old;
 440
 441                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 442                         return -EFAULT;
 443
 444                 out->uid        = tbuf_old.shm_perm.uid;
 445                 out->gid        = tbuf_old.shm_perm.gid;
 446                 out->mode       = tbuf_old.shm_perm.mode;
 447
 448                 return 0;
 449             }
 450         default:
 451                 return -EINVAL;
 452         }
 453 }
 454
 455 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 456 {
 457         switch(version) {
 458         case IPC_64:
 459                 return copy_to_user(buf, in, sizeof(*in));
 460         case IPC_OLD:
 461             {
 462                 struct shminfo out;
 463
 464                 if(in->shmmax > INT_MAX)
 465                         out.shmmax = INT_MAX;
 466                 else
 467                         out.shmmax = (int)in->shmmax;
 468
 469                 out.shmmin      = in->shmmin;
 470                 out.shmmni      = in->shmmni;
 471                 out.shmseg      = in->shmseg;
 472                 out.shmall      = in->shmall;
 473
 474                 return copy_to_user(buf, &out, sizeof(out));
 475             }
 476         default:
 477                 return -EINVAL;
 478         }
 479 }
 480
 481 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 482                 unsigned long *swp)
 483 {
 484         int i;
 485
 486         *rss = 0;
 487         *swp = 0;
 488
 489         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 490                 struct shmid_kernel *shp;
 491                 struct inode *inode;
 492
 493                 shp = shm_get(ns, i);
 494                 if(!shp)
 495                         continue;
 496
 497                 inode = shp->shm_file->f_dentry->d_inode;
 498
 499                 if (is_file_hugepages(shp->shm_file)) {
 500                         struct address_space *mapping = inode->i_mapping;
 501                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 502                 } else {
 503                         struct shmem_inode_info *info = SHMEM_I(inode);
 504                         spin_lock(&info->lock);
 505                         *rss += inode->i_mapping->nrpages;
 506                         *swp += info->swapped;
 507                         spin_unlock(&info->lock);
 508                 }
 509         }
 510 }
 511
 512 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 513 {
 514         struct shm_setbuf setbuf;
 515         struct shmid_kernel *shp;
 516         int err, version;
 517         struct ipc_namespace *ns;
 518
 519         if (cmd < 0 || shmid < 0) {
 520                 err = -EINVAL;
 521                 goto out;
 522         }
 523
 524         version = ipc_parse_version(&cmd);
 525         ns = current->nsproxy->ipc_ns;
 526
 527         switch (cmd) { /* replace with proc interface ? */
 528         case IPC_INFO:
 529         {
 530                 struct shminfo64 shminfo;
 531
 532                 err = security_shm_shmctl(NULL, cmd);
 533                 if (err)
 534                         return err;
 535
 536                 memset(&shminfo,0,sizeof(shminfo));
 537                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 538                 shminfo.shmmax = ns->shm_ctlmax;
 539                 shminfo.shmall = ns->shm_ctlall;
 540
 541                 shminfo.shmmin = SHMMIN;
 542                 if(copy_shminfo_to_user (buf, &shminfo, version))
 543                         return -EFAULT;
 544                 /* reading a integer is always atomic */
 545                 err= shm_ids(ns).max_id;
 546                 if(err<0)
 547                         err = 0;
 548                 goto out;
 549         }
 550         case SHM_INFO:
 551         {
 552                 struct shm_info shm_info;
 553
 554                 err = security_shm_shmctl(NULL, cmd);
 555                 if (err)
 556                         return err;
 557
 558                 memset(&shm_info,0,sizeof(shm_info));
 559                 mutex_lock(&shm_ids(ns).mutex);
 560                 shm_info.used_ids = shm_ids(ns).in_use;
 561                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 562                 shm_info.shm_tot = ns->shm_tot;
 563                 shm_info.swap_attempts = 0;
 564                 shm_info.swap_successes = 0;
 565                 err = shm_ids(ns).max_id;
 566                 mutex_unlock(&shm_ids(ns).mutex);
 567                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 568                         err = -EFAULT;
 569                         goto out;
 570                 }
 571
 572                 err = err < 0 ? 0 : err;
 573                 goto out;
 574         }
 575         case SHM_STAT:
 576         case IPC_STAT:
 577         {
 578                 struct shmid64_ds tbuf;
 579                 int result;
 580                 memset(&tbuf, 0, sizeof(tbuf));
 581                 shp = shm_lock(ns, shmid);
 582                 if(shp==NULL) {
 583                         err = -EINVAL;
 584                         goto out;
 585                 } else if(cmd==SHM_STAT) {
 586                         err = -EINVAL;
 587                         if (shmid > shm_ids(ns).max_id)
 588                                 goto out_unlock;
 589                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 590                 } else {
 591                         err = shm_checkid(ns, shp,shmid);
 592                         if(err)
 593                                 goto out_unlock;
 594                         result = 0;
 595                 }
 596                 err=-EACCES;
 597                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 598                         goto out_unlock;
 599                 err = security_shm_shmctl(shp, cmd);
 600                 if (err)
 601                         goto out_unlock;
 602                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 603                 tbuf.shm_segsz  = shp->shm_segsz;
 604                 tbuf.shm_atime  = shp->shm_atim;
 605                 tbuf.shm_dtime  = shp->shm_dtim;
 606                 tbuf.shm_ctime  = shp->shm_ctim;
 607                 tbuf.shm_cpid   = shp->shm_cprid;
 608                 tbuf.shm_lpid   = shp->shm_lprid;
 609                 if (!is_file_hugepages(shp->shm_file))
 610                         tbuf.shm_nattch = shp->shm_nattch;
 611                 else
 612                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 613                 shm_unlock(shp);
 614                 if(copy_shmid_to_user (buf, &tbuf, version))
 615                         err = -EFAULT;
 616                 else
 617                         err = result;
 618                 goto out;
 619         }
 620         case SHM_LOCK:
 621         case SHM_UNLOCK:
 622         {
 623                 shp = shm_lock(ns, shmid);
 624                 if(shp==NULL) {
 625                         err = -EINVAL;
 626                         goto out;
 627                 }
 628                 err = shm_checkid(ns, shp,shmid);
 629                 if(err)
 630                         goto out_unlock;
 631
 632                 err = audit_ipc_obj(&(shp->shm_perm));
 633                 if (err)
 634                         goto out_unlock;
 635
 636                 if (!capable(CAP_IPC_LOCK)) {
 637                         err = -EPERM;
 638                         if (current->euid != shp->shm_perm.uid &&
 639                             current->euid != shp->shm_perm.cuid)
 640                                 goto out_unlock;
 641                         if (cmd == SHM_LOCK &&
 642                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 643                                 goto out_unlock;
 644                 }
 645
 646                 err = security_shm_shmctl(shp, cmd);
 647                 if (err)
 648                         goto out_unlock;
 649
 650                 if(cmd==SHM_LOCK) {
 651                         struct user_struct * user = current->user;
 652                         if (!is_file_hugepages(shp->shm_file)) {
 653                                 err = shmem_lock(shp->shm_file, 1, user);
 654                                 if (!err) {
 655                                         shp->shm_perm.mode |= SHM_LOCKED;
 656                                         shp->mlock_user = user;
 657                                 }
 658                         }
 659                 } else if (!is_file_hugepages(shp->shm_file)) {
 660                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 661                         shp->shm_perm.mode &= ~SHM_LOCKED;
 662                         shp->mlock_user = NULL;
 663                 }
 664                 shm_unlock(shp);
 665                 goto out;
 666         }
 667         case IPC_RMID:
 668         {
 669                 /*
 670                  *      We cannot simply remove the file. The SVID states
 671                  *      that the block remains until the last person
 672                  *      detaches from it, then is deleted. A shmat() on
 673                  *      an RMID segment is legal in older Linux and if
 674                  *      we change it apps break...
 675                  *
 676                  *      Instead we set a destroyed flag, and then blow
 677                  *      the name away when the usage hits zero.
 678                  */
 679                 mutex_lock(&shm_ids(ns).mutex);
 680                 shp = shm_lock(ns, shmid);
 681                 err = -EINVAL;
 682                 if (shp == NULL)
 683                         goto out_up;
 684                 err = shm_checkid(ns, shp, shmid);
 685                 if(err)
 686                         goto out_unlock_up;
 687
 688                 err = audit_ipc_obj(&(shp->shm_perm));
 689                 if (err)
 690                         goto out_unlock_up;
 691
 692                 if (current->euid != shp->shm_perm.uid &&
 693                     current->euid != shp->shm_perm.cuid &&
 694                     !capable(CAP_SYS_ADMIN)) {
 695                         err=-EPERM;
 696                         goto out_unlock_up;
 697                 }
 698
 699                 err = security_shm_shmctl(shp, cmd);
 700                 if (err)
 701                         goto out_unlock_up;
 702
 703                 do_shm_rmid(ns, shp);
 704                 mutex_unlock(&shm_ids(ns).mutex);
 705                 goto out;
 706         }
 707
 708         case IPC_SET:
 709         {
 710                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 711                         err = -EFAULT;
 712                         goto out;
 713                 }
 714                 mutex_lock(&shm_ids(ns).mutex);
 715                 shp = shm_lock(ns, shmid);
 716                 err=-EINVAL;
 717                 if(shp==NULL)
 718                         goto out_up;
 719                 err = shm_checkid(ns, shp,shmid);
 720                 if(err)
 721                         goto out_unlock_up;
 722                 err = audit_ipc_obj(&(shp->shm_perm));
 723                 if (err)
 724                         goto out_unlock_up;
 725                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 726                 if (err)
 727                         goto out_unlock_up;
 728                 err=-EPERM;
 729                 if (current->euid != shp->shm_perm.uid &&
 730                     current->euid != shp->shm_perm.cuid &&
 731                     !capable(CAP_SYS_ADMIN)) {
 732                         goto out_unlock_up;
 733                 }
 734
 735                 err = security_shm_shmctl(shp, cmd);
 736                 if (err)
 737                         goto out_unlock_up;
 738
 739                 shp->shm_perm.uid = setbuf.uid;
 740                 shp->shm_perm.gid = setbuf.gid;
 741                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 742                         | (setbuf.mode & S_IRWXUGO);
 743                 shp->shm_ctim = get_seconds();
 744                 break;
 745         }
 746
 747         default:
 748                 err = -EINVAL;
 749                 goto out;
 750         }
 751
 752         err = 0;
 753 out_unlock_up:
 754         shm_unlock(shp);
 755 out_up:
 756         mutex_unlock(&shm_ids(ns).mutex);
 757         goto out;
 758 out_unlock:
 759         shm_unlock(shp);
 760 out:
 761         return err;
 762 }
 763
 764 /*
 765  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 766  *
 767  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 768  * "raddr" thing points to kernel space, and there has to be a wrapper around
 769  * this.
 770  */
 771 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 772 {
 773         struct shmid_kernel *shp;
 774         unsigned long addr;
 775         unsigned long size;
 776         struct file * file;
 777         int    err;
 778         unsigned long flags;
 779         unsigned long prot;
 780         int acc_mode;
 781         void *user_addr;
 782         struct ipc_namespace *ns;
 783
 784         if (shmid < 0) {
 785                 err = -EINVAL;
 786                 goto out;
 787         } else if ((addr = (ulong)shmaddr)) {
 788                 if (addr & (SHMLBA-1)) {
 789                         if (shmflg & SHM_RND)
 790                                 addr &= ~(SHMLBA-1);       /* round down */
 791                         else
 792 #ifndef __ARCH_FORCE_SHMLBA
 793                                 if (addr & ~PAGE_MASK)
 794 #endif
 795                                         return -EINVAL;
 796                 }
 797                 flags = MAP_SHARED | MAP_FIXED;
 798         } else {
 799                 if ((shmflg & SHM_REMAP))
 800                         return -EINVAL;
 801
 802                 flags = MAP_SHARED;
 803         }
 804
 805         if (shmflg & SHM_RDONLY) {
 806                 prot = PROT_READ;
 807                 acc_mode = S_IRUGO;
 808         } else {
 809                 prot = PROT_READ | PROT_WRITE;
 810                 acc_mode = S_IRUGO | S_IWUGO;
 811         }
 812         if (shmflg & SHM_EXEC) {
 813                 prot |= PROT_EXEC;
 814                 acc_mode |= S_IXUGO;
 815         }
 816
 817         /*
 818          * We cannot rely on the fs check since SYSV IPC does have an
 819          * additional creator id...
 820          */
 821         ns = current->nsproxy->ipc_ns;
 822         shp = shm_lock(ns, shmid);
 823         if(shp == NULL) {
 824                 err = -EINVAL;
 825                 goto out;
 826         }
 827         err = shm_checkid(ns, shp,shmid);
 828         if (err) {
 829                 shm_unlock(shp);
 830                 goto out;
 831         }
 832         if (ipcperms(&shp->shm_perm, acc_mode)) {
 833                 shm_unlock(shp);
 834                 err = -EACCES;
 835                 goto out;
 836         }
 837
 838         err = security_shm_shmat(shp, shmaddr, shmflg);
 839         if (err) {
 840                 shm_unlock(shp);
 841                 return err;
 842         }
 843
 844         file = shp->shm_file;
 845         size = i_size_read(file->f_dentry->d_inode);
 846         shp->shm_nattch++;
 847         shm_unlock(shp);
 848
 849         down_write(&current->mm->mmap_sem);
 850         if (addr && !(shmflg & SHM_REMAP)) {
 851                 user_addr = ERR_PTR(-EINVAL);
 852                 if (find_vma_intersection(current->mm, addr, addr + size))
 853                         goto invalid;
 854                 /*
 855                  * If shm segment goes below stack, make sure there is some
 856                  * space left for the stack to grow (at least 4 pages).
 857                  */
 858                 if (addr < current->mm->start_stack &&
 859                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 860                         goto invalid;
 861         }
 862
 863         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 864
 865 invalid:
 866         up_write(&current->mm->mmap_sem);
 867
 868         mutex_lock(&shm_ids(ns).mutex);
 869         shp = shm_lock(ns, shmid);
 870         BUG_ON(!shp);
 871         shp->shm_nattch--;
 872         if(shp->shm_nattch == 0 &&
 873            shp->shm_perm.mode & SHM_DEST)
 874                 shm_destroy(ns, shp);
 875         else
 876                 shm_unlock(shp);
 877         mutex_unlock(&shm_ids(ns).mutex);
 878
 879         *raddr = (unsigned long) user_addr;
 880         err = 0;
 881         if (IS_ERR(user_addr))
 882                 err = PTR_ERR(user_addr);
 883 out:
 884         return err;
 885 }
 886
 887 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 888 {
 889         unsigned long ret;
 890         long err;
 891
 892         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 893         if (err)
 894                 return err;
 895         force_successful_syscall_return();
 896         return (long)ret;
 897 }
 898
 899 /*
 900  * detach and kill segment if marked destroyed.
 901  * The work is done in shm_close.
 902  */
 903 asmlinkage long sys_shmdt(char __user *shmaddr)
 904 {
 905         struct mm_struct *mm = current->mm;
 906         struct vm_area_struct *vma, *next;
 907         unsigned long addr = (unsigned long)shmaddr;
 908         loff_t size = 0;
 909         int retval = -EINVAL;
 910
 911         if (addr & ~PAGE_MASK)
 912                 return retval;
 913
 914         down_write(&mm->mmap_sem);
 915
 916         /*
 917          * This function tries to be smart and unmap shm segments that
 918          * were modified by partial mlock or munmap calls:
 919          * - It first determines the size of the shm segment that should be
 920          *   unmapped: It searches for a vma that is backed by shm and that
 921          *   started at address shmaddr. It records it's size and then unmaps
 922          *   it.
 923          * - Then it unmaps all shm vmas that started at shmaddr and that
 924          *   are within the initially determined size.
 925          * Errors from do_munmap are ignored: the function only fails if
 926          * it's called with invalid parameters or if it's called to unmap
 927          * a part of a vma. Both calls in this function are for full vmas,
 928          * the parameters are directly copied from the vma itself and always
 929          * valid - therefore do_munmap cannot fail. (famous last words?)
 930          */
 931         /*
 932          * If it had been mremap()'d, the starting address would not
 933          * match the usual checks anyway. So assume all vma's are
 934          * above the starting address given.
 935          */
 936         vma = find_vma(mm, addr);
 937
 938         while (vma) {
 939                 next = vma->vm_next;
 940
 941                 /*
 942                  * Check if the starting address would match, i.e. it's
 943                  * a fragment created by mprotect() and/or munmap(), or it
 944                  * otherwise it starts at this address with no hassles.
 945                  */
 946                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 947                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 948
 949
 950                         size = vma->vm_file->f_dentry->d_inode->i_size;
 951                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 952                         /*
 953                          * We discovered the size of the shm segment, so
 954                          * break out of here and fall through to the next
 955                          * loop that uses the size information to stop
 956                          * searching for matching vma's.
 957                          */
 958                         retval = 0;
 959                         vma = next;
 960                         break;
 961                 }
 962                 vma = next;
 963         }
 964
 965         /*
 966          * We need look no further than the maximum address a fragment
 967          * could possibly have landed at. Also cast things to loff_t to
 968          * prevent overflows and make comparisions vs. equal-width types.
 969          */
 970         size = PAGE_ALIGN(size);
 971         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 972                 next = vma->vm_next;
 973
 974                 /* finding a matching vma now does not alter retval */
 975                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 976                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 977
 978                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 979                 vma = next;
 980         }
 981
 982         up_write(&mm->mmap_sem);
 983         return retval;
 984 }
 985
 986 #ifdef CONFIG_PROC_FS
 987 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 988 {
 989         struct shmid_kernel *shp = it;
 990         char *format;
 991
 992 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 993 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 994
 995         if (sizeof(size_t) <= sizeof(int))
 996                 format = SMALL_STRING;
 997         else
 998                 format = BIG_STRING;
 999         return seq_printf(s, format,
1000                           shp->shm_perm.key,
1001                           shp->id,
1002                           shp->shm_perm.mode,
1003                           shp->shm_segsz,
1004                           shp->shm_cprid,
1005                           shp->shm_lprid,
1006                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
1007                           shp->shm_perm.uid,
1008                           shp->shm_perm.gid,
1009                           shp->shm_perm.cuid,
1010                           shp->shm_perm.cgid,
1011                           shp->shm_atim,
1012                           shp->shm_dtim,
1013                           shp->shm_ctim);
1014 }
1015 #endif