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 #include <linux/vs_context.h>
  41 #include <linux/vs_limit.h>
  42
  43 #include <asm/uaccess.h>
  44
  45 #include "util.h"
  46
  47 static struct file_operations shm_file_operations;
  48 static struct vm_operations_struct shm_vm_ops;
  49
  50 static struct ipc_ids init_shm_ids;
  51
  52 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
  53
  54 #define shm_lock(ns, id)                \
  55         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
  56 #define shm_unlock(shp)                 \
  57         ipc_unlock(&(shp)->shm_perm)
  58 #define shm_get(ns, id)                 \
  59         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
  60 #define shm_buildid(ns, id, seq)        \
  61         ipc_buildid(&shm_ids(ns), id, seq)
  62
  63 static int newseg (struct ipc_namespace *ns, key_t key,
  64                 int shmflg, size_t size);
  65 static void shm_open (struct vm_area_struct *shmd);
  66 static void shm_close (struct vm_area_struct *shmd);
  67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  68 #ifdef CONFIG_PROC_FS
  69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  70 #endif
  71
  72 static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  73 {
  74         ns->ids[IPC_SHM_IDS] = ids;
  75         ns->shm_ctlmax = SHMMAX;
  76         ns->shm_ctlall = SHMALL;
  77         ns->shm_ctlmni = SHMMNI;
  78         ns->shm_tot = 0;
  79         ipc_init_ids(ids, 1);
  80 }
  81
  82 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  83 {
  84         if (shp->shm_nattch){
  85                 shp->shm_perm.mode |= SHM_DEST;
  86                 /* Do not find it any more */
  87                 shp->shm_perm.key = IPC_PRIVATE;
  88                 shm_unlock(shp);
  89         } else
  90                 shm_destroy(ns, shp);
  91 }
  92
  93 #ifdef CONFIG_IPC_NS
  94 int shm_init_ns(struct ipc_namespace *ns)
  95 {
  96         struct ipc_ids *ids;
  97
  98         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
  99         if (ids == NULL)
 100                 return -ENOMEM;
 101
 102         __shm_init_ns(ns, ids);
 103         return 0;
 104 }
 105
 106 void shm_exit_ns(struct ipc_namespace *ns)
 107 {
 108         int i;
 109         struct shmid_kernel *shp;
 110
 111         mutex_lock(&shm_ids(ns).mutex);
 112         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 113                 shp = shm_lock(ns, i);
 114                 if (shp == NULL)
 115                         continue;
 116
 117                 do_shm_rmid(ns, shp);
 118         }
 119         mutex_unlock(&shm_ids(ns).mutex);
 120
 121         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
 122         kfree(ns->ids[IPC_SHM_IDS]);
 123         ns->ids[IPC_SHM_IDS] = NULL;
 124 }
 125 #endif
 126
 127 void __init shm_init (void)
 128 {
 129         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 130         ipc_init_proc_interface("sysvipc/shm",
 131                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 132                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 133 }
 134
 135 static inline int shm_checkid(struct ipc_namespace *ns,
 136                 struct shmid_kernel *s, int id)
 137 {
 138         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 139                 return -EIDRM;
 140         return 0;
 141 }
 142
 143 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
 144 {
 145         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
 146 }
 147
 148 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 149 {
 150         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 151 }
 152
 153
 154
 155 static inline void shm_inc(struct ipc_namespace *ns, int id)
 156 {
 157         struct shmid_kernel *shp;
 158
 159         shp = shm_lock(ns, id);
 160         BUG_ON(!shp);
 161         shp->shm_atim = get_seconds();
 162         shp->shm_lprid = current->tgid;
 163         shp->shm_nattch++;
 164         shm_unlock(shp);
 165 }
 166
 167 #define shm_file_ns(file) (*((struct ipc_namespace **)&(file)->private_data))
 168
 169 /* This is called by fork, once for every shm attach. */
 170 static void shm_open(struct vm_area_struct *shmd)
 171 {
 172         shm_inc(shm_file_ns(shmd->vm_file),
 173                         shmd->vm_file->f_path.dentry->d_inode->i_ino);
 174 }
 175
 176 /*
 177  * shm_destroy - free the struct shmid_kernel
 178  *
 179  * @shp: struct to free
 180  *
 181  * It has to be called with shp and shm_ids.mutex locked,
 182  * but returns with shp unlocked and freed.
 183  */
 184 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 185 {
 186         struct vx_info *vxi = lookup_vx_info(shp->shm_perm.xid);
 187         int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 188
 189         vx_ipcshm_sub(vxi, shp, numpages);
 190         ns->shm_tot -= numpages;
 191
 192         shm_rmid(ns, shp->id);
 193         shm_unlock(shp);
 194         if (!is_file_hugepages(shp->shm_file))
 195                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 196         else
 197                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 198                                                 shp->mlock_user);
 199         fput (shp->shm_file);
 200         security_shm_free(shp);
 201         put_vx_info(vxi);
 202         ipc_rcu_putref(shp);
 203 }
 204
 205 /*
 206  * remove the attach descriptor shmd.
 207  * free memory for segment if it is marked destroyed.
 208  * The descriptor has already been removed from the current->mm->mmap list
 209  * and will later be kfree()d.
 210  */
 211 static void shm_close (struct vm_area_struct *shmd)
 212 {
 213         struct file * file = shmd->vm_file;
 214         int id = file->f_path.dentry->d_inode->i_ino;
 215         struct shmid_kernel *shp;
 216         struct ipc_namespace *ns;
 217
 218         ns = shm_file_ns(file);
 219
 220         mutex_lock(&shm_ids(ns).mutex);
 221         /* remove from the list of attaches of the shm segment */
 222         shp = shm_lock(ns, id);
 223         BUG_ON(!shp);
 224         shp->shm_lprid = current->tgid;
 225         shp->shm_dtim = get_seconds();
 226         shp->shm_nattch--;
 227         if(shp->shm_nattch == 0 &&
 228            shp->shm_perm.mode & SHM_DEST)
 229                 shm_destroy(ns, shp);
 230         else
 231                 shm_unlock(shp);
 232         mutex_unlock(&shm_ids(ns).mutex);
 233 }
 234
 235 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 236 {
 237         int ret;
 238
 239         ret = shmem_mmap(file, vma);
 240         if (ret == 0) {
 241                 vma->vm_ops = &shm_vm_ops;
 242                 if (!(vma->vm_flags & VM_WRITE))
 243                         vma->vm_flags &= ~VM_MAYWRITE;
 244                 shm_inc(shm_file_ns(file), file->f_path.dentry->d_inode->i_ino);
 245         }
 246
 247         return ret;
 248 }
 249
 250 static int shm_release(struct inode *ino, struct file *file)
 251 {
 252         struct ipc_namespace *ns;
 253
 254         ns = shm_file_ns(file);
 255         put_ipc_ns(ns);
 256         shm_file_ns(file) = NULL;
 257         return 0;
 258 }
 259
 260 static struct file_operations shm_file_operations = {
 261         .mmap           = shm_mmap,
 262         .release        = shm_release,
 263 #ifndef CONFIG_MMU
 264         .get_unmapped_area = shmem_get_unmapped_area,
 265 #endif
 266 };
 267
 268 static struct vm_operations_struct shm_vm_ops = {
 269         .open   = shm_open,     /* callback for a new vm-area open */
 270         .close  = shm_close,    /* callback for when the vm-area is released */
 271         .nopage = shmem_nopage,
 272 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 273         .set_policy = shmem_set_policy,
 274         .get_policy = shmem_get_policy,
 275 #endif
 276 };
 277
 278 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 279 {
 280         int error;
 281         struct shmid_kernel *shp;
 282         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 283         struct file * file;
 284         char name[13];
 285         int id;
 286
 287         if (size < SHMMIN || size > ns->shm_ctlmax)
 288                 return -EINVAL;
 289
 290         if (ns->shm_tot + numpages > ns->shm_ctlall)
 291                 return -ENOSPC;
 292
 293         if (!vx_ipcshm_avail(current->vx_info, numpages))
 294                 return -ENOSPC;
 295
 296         shp = ipc_rcu_alloc(sizeof(*shp));
 297         if (!shp)
 298                 return -ENOMEM;
 299
 300         shp->shm_perm.key = key;
 301         shp->shm_perm.xid = vx_current_xid();
 302         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 303         shp->mlock_user = NULL;
 304
 305         shp->shm_perm.security = NULL;
 306         error = security_shm_alloc(shp);
 307         if (error) {
 308                 ipc_rcu_putref(shp);
 309                 return error;
 310         }
 311
 312         if (shmflg & SHM_HUGETLB) {
 313                 /* hugetlb_zero_setup takes care of mlock user accounting */
 314                 file = hugetlb_zero_setup(size);
 315                 shp->mlock_user = current->user;
 316         } else {
 317                 int acctflag = VM_ACCOUNT;
 318                 /*
 319                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 320                  * if it's asked for.
 321                  */
 322                 if  ((shmflg & SHM_NORESERVE) &&
 323                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 324                         acctflag = 0;
 325                 sprintf (name, "SYSV%08x", key);
 326                 file = shmem_file_setup(name, size, acctflag);
 327         }
 328         error = PTR_ERR(file);
 329         if (IS_ERR(file))
 330                 goto no_file;
 331
 332         error = -ENOSPC;
 333         id = shm_addid(ns, shp);
 334         if(id == -1)
 335                 goto no_id;
 336
 337         shp->shm_cprid = current->tgid;
 338         shp->shm_lprid = 0;
 339         shp->shm_atim = shp->shm_dtim = 0;
 340         shp->shm_ctim = get_seconds();
 341         shp->shm_segsz = size;
 342         shp->shm_nattch = 0;
 343         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 344         shp->shm_file = file;
 345         file->f_path.dentry->d_inode->i_ino = shp->id;
 346
 347         shm_file_ns(file) = get_ipc_ns(ns);
 348
 349         /* Hugetlb ops would have already been assigned. */
 350         if (!(shmflg & SHM_HUGETLB))
 351                 file->f_op = &shm_file_operations;
 352
 353         ns->shm_tot += numpages;
 354         vx_ipcshm_add(current->vx_info, key, numpages);
 355         shm_unlock(shp);
 356         return shp->id;
 357
 358 no_id:
 359         fput(file);
 360 no_file:
 361         security_shm_free(shp);
 362         ipc_rcu_putref(shp);
 363         return error;
 364 }
 365
 366 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 367 {
 368         struct shmid_kernel *shp;
 369         int err, id = 0;
 370         struct ipc_namespace *ns;
 371
 372         ns = current->nsproxy->ipc_ns;
 373
 374         mutex_lock(&shm_ids(ns).mutex);
 375         if (key == IPC_PRIVATE) {
 376                 err = newseg(ns, key, shmflg, size);
 377         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 378                 if (!(shmflg & IPC_CREAT))
 379                         err = -ENOENT;
 380                 else
 381                         err = newseg(ns, key, shmflg, size);
 382         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 383                 err = -EEXIST;
 384         } else {
 385                 shp = shm_lock(ns, id);
 386                 BUG_ON(shp==NULL);
 387                 if (shp->shm_segsz < size)
 388                         err = -EINVAL;
 389                 else if (ipcperms(&shp->shm_perm, shmflg))
 390                         err = -EACCES;
 391                 else {
 392                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 393                         err = security_shm_associate(shp, shmflg);
 394                         if (!err)
 395                                 err = shmid;
 396                 }
 397                 shm_unlock(shp);
 398         }
 399         mutex_unlock(&shm_ids(ns).mutex);
 400
 401         return err;
 402 }
 403
 404 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 405 {
 406         switch(version) {
 407         case IPC_64:
 408                 return copy_to_user(buf, in, sizeof(*in));
 409         case IPC_OLD:
 410             {
 411                 struct shmid_ds out;
 412
 413                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 414                 out.shm_segsz   = in->shm_segsz;
 415                 out.shm_atime   = in->shm_atime;
 416                 out.shm_dtime   = in->shm_dtime;
 417                 out.shm_ctime   = in->shm_ctime;
 418                 out.shm_cpid    = in->shm_cpid;
 419                 out.shm_lpid    = in->shm_lpid;
 420                 out.shm_nattch  = in->shm_nattch;
 421
 422                 return copy_to_user(buf, &out, sizeof(out));
 423             }
 424         default:
 425                 return -EINVAL;
 426         }
 427 }
 428
 429 struct shm_setbuf {
 430         uid_t   uid;
 431         gid_t   gid;
 432         mode_t  mode;
 433 };
 434
 435 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 436 {
 437         switch(version) {
 438         case IPC_64:
 439             {
 440                 struct shmid64_ds tbuf;
 441
 442                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 443                         return -EFAULT;
 444
 445                 out->uid        = tbuf.shm_perm.uid;
 446                 out->gid        = tbuf.shm_perm.gid;
 447                 out->mode       = tbuf.shm_perm.mode;
 448
 449                 return 0;
 450             }
 451         case IPC_OLD:
 452             {
 453                 struct shmid_ds tbuf_old;
 454
 455                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 456                         return -EFAULT;
 457
 458                 out->uid        = tbuf_old.shm_perm.uid;
 459                 out->gid        = tbuf_old.shm_perm.gid;
 460                 out->mode       = tbuf_old.shm_perm.mode;
 461
 462                 return 0;
 463             }
 464         default:
 465                 return -EINVAL;
 466         }
 467 }
 468
 469 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 470 {
 471         switch(version) {
 472         case IPC_64:
 473                 return copy_to_user(buf, in, sizeof(*in));
 474         case IPC_OLD:
 475             {
 476                 struct shminfo out;
 477
 478                 if(in->shmmax > INT_MAX)
 479                         out.shmmax = INT_MAX;
 480                 else
 481                         out.shmmax = (int)in->shmmax;
 482
 483                 out.shmmin      = in->shmmin;
 484                 out.shmmni      = in->shmmni;
 485                 out.shmseg      = in->shmseg;
 486                 out.shmall      = in->shmall;
 487
 488                 return copy_to_user(buf, &out, sizeof(out));
 489             }
 490         default:
 491                 return -EINVAL;
 492         }
 493 }
 494
 495 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 496                 unsigned long *swp)
 497 {
 498         int i;
 499
 500         *rss = 0;
 501         *swp = 0;
 502
 503         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 504                 struct shmid_kernel *shp;
 505                 struct inode *inode;
 506
 507                 shp = shm_get(ns, i);
 508                 if(!shp)
 509                         continue;
 510
 511                 inode = shp->shm_file->f_path.dentry->d_inode;
 512
 513                 if (is_file_hugepages(shp->shm_file)) {
 514                         struct address_space *mapping = inode->i_mapping;
 515                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 516                 } else {
 517                         struct shmem_inode_info *info = SHMEM_I(inode);
 518                         spin_lock(&info->lock);
 519                         *rss += inode->i_mapping->nrpages;
 520                         *swp += info->swapped;
 521                         spin_unlock(&info->lock);
 522                 }
 523         }
 524 }
 525
 526 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 527 {
 528         struct shm_setbuf setbuf;
 529         struct shmid_kernel *shp;
 530         int err, version;
 531         struct ipc_namespace *ns;
 532
 533         if (cmd < 0 || shmid < 0) {
 534                 err = -EINVAL;
 535                 goto out;
 536         }
 537
 538         version = ipc_parse_version(&cmd);
 539         ns = current->nsproxy->ipc_ns;
 540
 541         switch (cmd) { /* replace with proc interface ? */
 542         case IPC_INFO:
 543         {
 544                 struct shminfo64 shminfo;
 545
 546                 err = security_shm_shmctl(NULL, cmd);
 547                 if (err)
 548                         return err;
 549
 550                 memset(&shminfo,0,sizeof(shminfo));
 551                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 552                 shminfo.shmmax = ns->shm_ctlmax;
 553                 shminfo.shmall = ns->shm_ctlall;
 554
 555                 shminfo.shmmin = SHMMIN;
 556                 if(copy_shminfo_to_user (buf, &shminfo, version))
 557                         return -EFAULT;
 558                 /* reading a integer is always atomic */
 559                 err= shm_ids(ns).max_id;
 560                 if(err<0)
 561                         err = 0;
 562                 goto out;
 563         }
 564         case SHM_INFO:
 565         {
 566                 struct shm_info shm_info;
 567
 568                 err = security_shm_shmctl(NULL, cmd);
 569                 if (err)
 570                         return err;
 571
 572                 memset(&shm_info,0,sizeof(shm_info));
 573                 mutex_lock(&shm_ids(ns).mutex);
 574                 shm_info.used_ids = shm_ids(ns).in_use;
 575                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 576                 shm_info.shm_tot = ns->shm_tot;
 577                 shm_info.swap_attempts = 0;
 578                 shm_info.swap_successes = 0;
 579                 err = shm_ids(ns).max_id;
 580                 mutex_unlock(&shm_ids(ns).mutex);
 581                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 582                         err = -EFAULT;
 583                         goto out;
 584                 }
 585
 586                 err = err < 0 ? 0 : err;
 587                 goto out;
 588         }
 589         case SHM_STAT:
 590         case IPC_STAT:
 591         {
 592                 struct shmid64_ds tbuf;
 593                 int result;
 594                 memset(&tbuf, 0, sizeof(tbuf));
 595                 shp = shm_lock(ns, shmid);
 596                 if(shp==NULL) {
 597                         err = -EINVAL;
 598                         goto out;
 599                 } else if(cmd==SHM_STAT) {
 600                         err = -EINVAL;
 601                         if (shmid > shm_ids(ns).max_id)
 602                                 goto out_unlock;
 603                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 604                 } else {
 605                         err = shm_checkid(ns, shp,shmid);
 606                         if(err)
 607                                 goto out_unlock;
 608                         result = 0;
 609                 }
 610                 err=-EACCES;
 611                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 612                         goto out_unlock;
 613                 err = security_shm_shmctl(shp, cmd);
 614                 if (err)
 615                         goto out_unlock;
 616                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 617                 tbuf.shm_segsz  = shp->shm_segsz;
 618                 tbuf.shm_atime  = shp->shm_atim;
 619                 tbuf.shm_dtime  = shp->shm_dtim;
 620                 tbuf.shm_ctime  = shp->shm_ctim;
 621                 tbuf.shm_cpid   = shp->shm_cprid;
 622                 tbuf.shm_lpid   = shp->shm_lprid;
 623                 if (!is_file_hugepages(shp->shm_file))
 624                         tbuf.shm_nattch = shp->shm_nattch;
 625                 else
 626                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 627                 shm_unlock(shp);
 628                 if(copy_shmid_to_user (buf, &tbuf, version))
 629                         err = -EFAULT;
 630                 else
 631                         err = result;
 632                 goto out;
 633         }
 634         case SHM_LOCK:
 635         case SHM_UNLOCK:
 636         {
 637                 shp = shm_lock(ns, shmid);
 638                 if(shp==NULL) {
 639                         err = -EINVAL;
 640                         goto out;
 641                 }
 642                 err = shm_checkid(ns, shp,shmid);
 643                 if(err)
 644                         goto out_unlock;
 645
 646                 err = audit_ipc_obj(&(shp->shm_perm));
 647                 if (err)
 648                         goto out_unlock;
 649
 650                 if (!capable(CAP_IPC_LOCK)) {
 651                         err = -EPERM;
 652                         if (current->euid != shp->shm_perm.uid &&
 653                             current->euid != shp->shm_perm.cuid)
 654                                 goto out_unlock;
 655                         if (cmd == SHM_LOCK &&
 656                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 657                                 goto out_unlock;
 658                 }
 659
 660                 err = security_shm_shmctl(shp, cmd);
 661                 if (err)
 662                         goto out_unlock;
 663
 664                 if(cmd==SHM_LOCK) {
 665                         struct user_struct * user = current->user;
 666                         if (!is_file_hugepages(shp->shm_file)) {
 667                                 err = shmem_lock(shp->shm_file, 1, user);
 668                                 if (!err) {
 669                                         shp->shm_perm.mode |= SHM_LOCKED;
 670                                         shp->mlock_user = user;
 671                                 }
 672                         }
 673                 } else if (!is_file_hugepages(shp->shm_file)) {
 674                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 675                         shp->shm_perm.mode &= ~SHM_LOCKED;
 676                         shp->mlock_user = NULL;
 677                 }
 678                 shm_unlock(shp);
 679                 goto out;
 680         }
 681         case IPC_RMID:
 682         {
 683                 /*
 684                  *      We cannot simply remove the file. The SVID states
 685                  *      that the block remains until the last person
 686                  *      detaches from it, then is deleted. A shmat() on
 687                  *      an RMID segment is legal in older Linux and if
 688                  *      we change it apps break...
 689                  *
 690                  *      Instead we set a destroyed flag, and then blow
 691                  *      the name away when the usage hits zero.
 692                  */
 693                 mutex_lock(&shm_ids(ns).mutex);
 694                 shp = shm_lock(ns, shmid);
 695                 err = -EINVAL;
 696                 if (shp == NULL)
 697                         goto out_up;
 698                 err = shm_checkid(ns, shp, shmid);
 699                 if(err)
 700                         goto out_unlock_up;
 701
 702                 err = audit_ipc_obj(&(shp->shm_perm));
 703                 if (err)
 704                         goto out_unlock_up;
 705
 706                 if (current->euid != shp->shm_perm.uid &&
 707                     current->euid != shp->shm_perm.cuid &&
 708                     !capable(CAP_SYS_ADMIN)) {
 709                         err=-EPERM;
 710                         goto out_unlock_up;
 711                 }
 712
 713                 err = security_shm_shmctl(shp, cmd);
 714                 if (err)
 715                         goto out_unlock_up;
 716
 717                 do_shm_rmid(ns, shp);
 718                 mutex_unlock(&shm_ids(ns).mutex);
 719                 goto out;
 720         }
 721
 722         case IPC_SET:
 723         {
 724                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 725                         err = -EFAULT;
 726                         goto out;
 727                 }
 728                 mutex_lock(&shm_ids(ns).mutex);
 729                 shp = shm_lock(ns, shmid);
 730                 err=-EINVAL;
 731                 if(shp==NULL)
 732                         goto out_up;
 733                 err = shm_checkid(ns, shp,shmid);
 734                 if(err)
 735                         goto out_unlock_up;
 736                 err = audit_ipc_obj(&(shp->shm_perm));
 737                 if (err)
 738                         goto out_unlock_up;
 739                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 740                 if (err)
 741                         goto out_unlock_up;
 742                 err=-EPERM;
 743                 if (current->euid != shp->shm_perm.uid &&
 744                     current->euid != shp->shm_perm.cuid &&
 745                     !capable(CAP_SYS_ADMIN)) {
 746                         goto out_unlock_up;
 747                 }
 748
 749                 err = security_shm_shmctl(shp, cmd);
 750                 if (err)
 751                         goto out_unlock_up;
 752
 753                 shp->shm_perm.uid = setbuf.uid;
 754                 shp->shm_perm.gid = setbuf.gid;
 755                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 756                         | (setbuf.mode & S_IRWXUGO);
 757                 shp->shm_ctim = get_seconds();
 758                 break;
 759         }
 760
 761         default:
 762                 err = -EINVAL;
 763                 goto out;
 764         }
 765
 766         err = 0;
 767 out_unlock_up:
 768         shm_unlock(shp);
 769 out_up:
 770         mutex_unlock(&shm_ids(ns).mutex);
 771         goto out;
 772 out_unlock:
 773         shm_unlock(shp);
 774 out:
 775         return err;
 776 }
 777
 778 /*
 779  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 780  *
 781  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 782  * "raddr" thing points to kernel space, and there has to be a wrapper around
 783  * this.
 784  */
 785 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 786 {
 787         struct shmid_kernel *shp;
 788         unsigned long addr;
 789         unsigned long size;
 790         struct file * file;
 791         int    err;
 792         unsigned long flags;
 793         unsigned long prot;
 794         int acc_mode;
 795         void *user_addr;
 796         struct ipc_namespace *ns;
 797
 798         if (shmid < 0) {
 799                 err = -EINVAL;
 800                 goto out;
 801         } else if ((addr = (ulong)shmaddr)) {
 802                 if (addr & (SHMLBA-1)) {
 803                         if (shmflg & SHM_RND)
 804                                 addr &= ~(SHMLBA-1);       /* round down */
 805                         else
 806 #ifndef __ARCH_FORCE_SHMLBA
 807                                 if (addr & ~PAGE_MASK)
 808 #endif
 809                                         return -EINVAL;
 810                 }
 811                 flags = MAP_SHARED | MAP_FIXED;
 812         } else {
 813                 if ((shmflg & SHM_REMAP))
 814                         return -EINVAL;
 815
 816                 flags = MAP_SHARED;
 817         }
 818
 819         if (shmflg & SHM_RDONLY) {
 820                 prot = PROT_READ;
 821                 acc_mode = S_IRUGO;
 822         } else {
 823                 prot = PROT_READ | PROT_WRITE;
 824                 acc_mode = S_IRUGO | S_IWUGO;
 825         }
 826         if (shmflg & SHM_EXEC) {
 827                 prot |= PROT_EXEC;
 828                 acc_mode |= S_IXUGO;
 829         }
 830
 831         /*
 832          * We cannot rely on the fs check since SYSV IPC does have an
 833          * additional creator id...
 834          */
 835         ns = current->nsproxy->ipc_ns;
 836         shp = shm_lock(ns, shmid);
 837         if(shp == NULL) {
 838                 err = -EINVAL;
 839                 goto out;
 840         }
 841         err = shm_checkid(ns, shp,shmid);
 842         if (err) {
 843                 shm_unlock(shp);
 844                 goto out;
 845         }
 846         if (ipcperms(&shp->shm_perm, acc_mode)) {
 847                 shm_unlock(shp);
 848                 err = -EACCES;
 849                 goto out;
 850         }
 851
 852         err = security_shm_shmat(shp, shmaddr, shmflg);
 853         if (err) {
 854                 shm_unlock(shp);
 855                 return err;
 856         }
 857
 858         file = shp->shm_file;
 859         size = i_size_read(file->f_path.dentry->d_inode);
 860         shp->shm_nattch++;
 861         shm_unlock(shp);
 862
 863         down_write(&current->mm->mmap_sem);
 864         if (addr && !(shmflg & SHM_REMAP)) {
 865                 user_addr = ERR_PTR(-EINVAL);
 866                 if (find_vma_intersection(current->mm, addr, addr + size))
 867                         goto invalid;
 868                 /*
 869                  * If shm segment goes below stack, make sure there is some
 870                  * space left for the stack to grow (at least 4 pages).
 871                  */
 872                 if (addr < current->mm->start_stack &&
 873                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 874                         goto invalid;
 875         }
 876
 877         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 878
 879 invalid:
 880         up_write(&current->mm->mmap_sem);
 881
 882         mutex_lock(&shm_ids(ns).mutex);
 883         shp = shm_lock(ns, shmid);
 884         BUG_ON(!shp);
 885         shp->shm_nattch--;
 886         if(shp->shm_nattch == 0 &&
 887            shp->shm_perm.mode & SHM_DEST)
 888                 shm_destroy(ns, shp);
 889         else
 890                 shm_unlock(shp);
 891         mutex_unlock(&shm_ids(ns).mutex);
 892
 893         *raddr = (unsigned long) user_addr;
 894         err = 0;
 895         if (IS_ERR(user_addr))
 896                 err = PTR_ERR(user_addr);
 897 out:
 898         return err;
 899 }
 900
 901 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 902 {
 903         unsigned long ret;
 904         long err;
 905
 906         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 907         if (err)
 908                 return err;
 909         force_successful_syscall_return();
 910         return (long)ret;
 911 }
 912
 913 /*
 914  * detach and kill segment if marked destroyed.
 915  * The work is done in shm_close.
 916  */
 917 asmlinkage long sys_shmdt(char __user *shmaddr)
 918 {
 919         struct mm_struct *mm = current->mm;
 920         struct vm_area_struct *vma, *next;
 921         unsigned long addr = (unsigned long)shmaddr;
 922         loff_t size = 0;
 923         int retval = -EINVAL;
 924
 925         if (addr & ~PAGE_MASK)
 926                 return retval;
 927
 928         down_write(&mm->mmap_sem);
 929
 930         /*
 931          * This function tries to be smart and unmap shm segments that
 932          * were modified by partial mlock or munmap calls:
 933          * - It first determines the size of the shm segment that should be
 934          *   unmapped: It searches for a vma that is backed by shm and that
 935          *   started at address shmaddr. It records it's size and then unmaps
 936          *   it.
 937          * - Then it unmaps all shm vmas that started at shmaddr and that
 938          *   are within the initially determined size.
 939          * Errors from do_munmap are ignored: the function only fails if
 940          * it's called with invalid parameters or if it's called to unmap
 941          * a part of a vma. Both calls in this function are for full vmas,
 942          * the parameters are directly copied from the vma itself and always
 943          * valid - therefore do_munmap cannot fail. (famous last words?)
 944          */
 945         /*
 946          * If it had been mremap()'d, the starting address would not
 947          * match the usual checks anyway. So assume all vma's are
 948          * above the starting address given.
 949          */
 950         vma = find_vma(mm, addr);
 951
 952         while (vma) {
 953                 next = vma->vm_next;
 954
 955                 /*
 956                  * Check if the starting address would match, i.e. it's
 957                  * a fragment created by mprotect() and/or munmap(), or it
 958                  * otherwise it starts at this address with no hassles.
 959                  */
 960                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 961                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 962
 963
 964                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
 965                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 966                         /*
 967                          * We discovered the size of the shm segment, so
 968                          * break out of here and fall through to the next
 969                          * loop that uses the size information to stop
 970                          * searching for matching vma's.
 971                          */
 972                         retval = 0;
 973                         vma = next;
 974                         break;
 975                 }
 976                 vma = next;
 977         }
 978
 979         /*
 980          * We need look no further than the maximum address a fragment
 981          * could possibly have landed at. Also cast things to loff_t to
 982          * prevent overflows and make comparisions vs. equal-width types.
 983          */
 984         size = PAGE_ALIGN(size);
 985         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 986                 next = vma->vm_next;
 987
 988                 /* finding a matching vma now does not alter retval */
 989                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 990                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 991
 992                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 993                 vma = next;
 994         }
 995
 996         up_write(&mm->mmap_sem);
 997         return retval;
 998 }
 999
1000 #ifdef CONFIG_PROC_FS
1001 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1002 {
1003         struct shmid_kernel *shp = it;
1004         char *format;
1005
1006 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1007 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1008
1009         if (sizeof(size_t) <= sizeof(int))
1010                 format = SMALL_STRING;
1011         else
1012                 format = BIG_STRING;
1013         return seq_printf(s, format,
1014                           shp->shm_perm.key,
1015                           shp->id,
1016                           shp->shm_perm.mode,
1017                           shp->shm_segsz,
1018                           shp->shm_cprid,
1019                           shp->shm_lprid,
1020                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
1021                           shp->shm_perm.uid,
1022                           shp->shm_perm.gid,
1023                           shp->shm_perm.cuid,
1024                           shp->shm_perm.cgid,
1025                           shp->shm_atim,
1026                           shp->shm_dtim,
1027                           shp->shm_ctim);
1028 }
1029 #endif