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/mount.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "util.h"
  45
  46 struct shm_file_data {
  47         int id;
  48         struct ipc_namespace *ns;
  49         struct file *file;
  50         const struct vm_operations_struct *vm_ops;
  51 };
  52
  53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  54
  55 static const struct file_operations shm_file_operations;
  56 static struct vm_operations_struct shm_vm_ops;
  57
  58 static struct ipc_ids init_shm_ids;
  59
  60 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
  61
  62 #define shm_lock(ns, id)                \
  63         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
  64 #define shm_unlock(shp)                 \
  65         ipc_unlock(&(shp)->shm_perm)
  66 #define shm_get(ns, id)                 \
  67         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
  68 #define shm_buildid(ns, id, seq)        \
  69         ipc_buildid(&shm_ids(ns), id, seq)
  70
  71 static int newseg (struct ipc_namespace *ns, key_t key,
  72                 int shmflg, size_t size);
  73 static void shm_open(struct vm_area_struct *vma);
  74 static void shm_close(struct vm_area_struct *vma);
  75 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  76 #ifdef CONFIG_PROC_FS
  77 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  78 #endif
  79
  80 static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  81 {
  82         ns->ids[IPC_SHM_IDS] = ids;
  83         ns->shm_ctlmax = SHMMAX;
  84         ns->shm_ctlall = SHMALL;
  85         ns->shm_ctlmni = SHMMNI;
  86         ns->shm_tot = 0;
  87         ipc_init_ids(ids, 1);
  88 }
  89
  90 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  91 {
  92         if (shp->shm_nattch){
  93                 shp->shm_perm.mode |= SHM_DEST;
  94                 /* Do not find it any more */
  95                 shp->shm_perm.key = IPC_PRIVATE;
  96                 shm_unlock(shp);
  97         } else
  98                 shm_destroy(ns, shp);
  99 }
 100
 101 #ifdef CONFIG_IPC_NS
 102 int shm_init_ns(struct ipc_namespace *ns)
 103 {
 104         struct ipc_ids *ids;
 105
 106         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 107         if (ids == NULL)
 108                 return -ENOMEM;
 109
 110         __shm_init_ns(ns, ids);
 111         return 0;
 112 }
 113
 114 void shm_exit_ns(struct ipc_namespace *ns)
 115 {
 116         int i;
 117         struct shmid_kernel *shp;
 118
 119         mutex_lock(&shm_ids(ns).mutex);
 120         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 121                 shp = shm_lock(ns, i);
 122                 if (shp == NULL)
 123                         continue;
 124
 125                 do_shm_rmid(ns, shp);
 126         }
 127         mutex_unlock(&shm_ids(ns).mutex);
 128
 129         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
 130         kfree(ns->ids[IPC_SHM_IDS]);
 131         ns->ids[IPC_SHM_IDS] = NULL;
 132 }
 133 #endif
 134
 135 void __init shm_init (void)
 136 {
 137         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 138         ipc_init_proc_interface("sysvipc/shm",
 139                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 140                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 141 }
 142
 143 static inline int shm_checkid(struct ipc_namespace *ns,
 144                 struct shmid_kernel *s, int id)
 145 {
 146         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 147                 return -EIDRM;
 148         return 0;
 149 }
 150
 151 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
 152 {
 153         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
 154 }
 155
 156 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 157 {
 158         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 159 }
 160
 161
 162
 163 /* This is called by fork, once for every shm attach. */
 164 static void shm_open(struct vm_area_struct *vma)
 165 {
 166         struct file *file = vma->vm_file;
 167         struct shm_file_data *sfd = shm_file_data(file);
 168         struct shmid_kernel *shp;
 169
 170         shp = shm_lock(sfd->ns, sfd->id);
 171         BUG_ON(!shp);
 172         shp->shm_atim = get_seconds();
 173         shp->shm_lprid = current->tgid;
 174         shp->shm_nattch++;
 175         shm_unlock(shp);
 176 }
 177
 178 /*
 179  * shm_destroy - free the struct shmid_kernel
 180  *
 181  * @shp: struct to free
 182  *
 183  * It has to be called with shp and shm_ids.mutex locked,
 184  * but returns with shp unlocked and freed.
 185  */
 186 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 187 {
 188         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 189         shm_rmid(ns, shp->id);
 190         shm_unlock(shp);
 191         if (!is_file_hugepages(shp->shm_file))
 192                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 193         else
 194                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 195                                                 shp->mlock_user);
 196         fput (shp->shm_file);
 197         security_shm_free(shp);
 198         ipc_rcu_putref(shp);
 199 }
 200
 201 /*
 202  * remove the attach descriptor vma.
 203  * free memory for segment if it is marked destroyed.
 204  * The descriptor has already been removed from the current->mm->mmap list
 205  * and will later be kfree()d.
 206  */
 207 static void shm_close(struct vm_area_struct *vma)
 208 {
 209         struct file * file = vma->vm_file;
 210         struct shm_file_data *sfd = shm_file_data(file);
 211         struct shmid_kernel *shp;
 212         struct ipc_namespace *ns = sfd->ns;
 213
 214         mutex_lock(&shm_ids(ns).mutex);
 215         /* remove from the list of attaches of the shm segment */
 216         shp = shm_lock(ns, sfd->id);
 217         BUG_ON(!shp);
 218         shp->shm_lprid = current->tgid;
 219         shp->shm_dtim = get_seconds();
 220         shp->shm_nattch--;
 221         if(shp->shm_nattch == 0 &&
 222            shp->shm_perm.mode & SHM_DEST)
 223                 shm_destroy(ns, shp);
 224         else
 225                 shm_unlock(shp);
 226         mutex_unlock(&shm_ids(ns).mutex);
 227 }
 228
 229 struct page *shm_nopage(struct vm_area_struct *vma, unsigned long address,
 230                         int *type)
 231 {
 232         struct file *file = vma->vm_file;
 233         struct shm_file_data *sfd = shm_file_data(file);
 234
 235         return sfd->vm_ops->nopage(vma, address, type);
 236 }
 237
 238 #ifdef CONFIG_NUMA
 239 int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 240 {
 241         struct file *file = vma->vm_file;
 242         struct shm_file_data *sfd = shm_file_data(file);
 243         int err = 0;
 244         if (sfd->vm_ops->set_policy)
 245                 err = sfd->vm_ops->set_policy(vma, new);
 246         return err;
 247 }
 248
 249 struct mempolicy *shm_get_policy(struct vm_area_struct *vma, unsigned long addr)
 250 {
 251         struct file *file = vma->vm_file;
 252         struct shm_file_data *sfd = shm_file_data(file);
 253         struct mempolicy *pol = NULL;
 254
 255         if (sfd->vm_ops->get_policy)
 256                 pol = sfd->vm_ops->get_policy(vma, addr);
 257         else
 258                 pol = vma->vm_policy;
 259         return pol;
 260 }
 261 #endif
 262
 263 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 264 {
 265         struct shm_file_data *sfd = shm_file_data(file);
 266         int ret;
 267
 268         ret = sfd->file->f_op->mmap(sfd->file, vma);
 269         if (ret != 0)
 270                 return ret;
 271         sfd->vm_ops = vma->vm_ops;
 272         vma->vm_ops = &shm_vm_ops;
 273         shm_open(vma);
 274
 275         return ret;
 276 }
 277
 278 static int shm_release(struct inode *ino, struct file *file)
 279 {
 280         struct shm_file_data *sfd = shm_file_data(file);
 281
 282         put_ipc_ns(sfd->ns);
 283         shm_file_data(file) = NULL;
 284         kfree(sfd);
 285         return 0;
 286 }
 287
 288 #ifndef CONFIG_MMU
 289 static unsigned long shm_get_unmapped_area(struct file *file,
 290         unsigned long addr, unsigned long len, unsigned long pgoff,
 291         unsigned long flags)
 292 {
 293         struct shm_file_data *sfd = shm_file_data(file);
 294         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len, pgoff,
 295                                                         flags);
 296 }
 297 #else
 298 #define shm_get_unmapped_area NULL
 299 #endif
 300
 301 static const struct file_operations shm_file_operations = {
 302         .mmap           = shm_mmap,
 303         .release        = shm_release,
 304         .get_unmapped_area      = shm_get_unmapped_area,
 305 };
 306
 307 static struct vm_operations_struct shm_vm_ops = {
 308         .open   = shm_open,     /* callback for a new vm-area open */
 309         .close  = shm_close,    /* callback for when the vm-area is released */
 310         .nopage = shm_nopage,
 311 #if defined(CONFIG_NUMA)
 312         .set_policy = shm_set_policy,
 313         .get_policy = shm_get_policy,
 314 #endif
 315 };
 316
 317 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 318 {
 319         int error;
 320         struct shmid_kernel *shp;
 321         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 322         struct file * file;
 323         char name[13];
 324         int id;
 325
 326         if (size < SHMMIN || size > ns->shm_ctlmax)
 327                 return -EINVAL;
 328
 329         if (ns->shm_tot + numpages > ns->shm_ctlall)
 330                 return -ENOSPC;
 331
 332         shp = ipc_rcu_alloc(sizeof(*shp));
 333         if (!shp)
 334                 return -ENOMEM;
 335
 336         shp->shm_perm.key = key;
 337         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 338         shp->mlock_user = NULL;
 339
 340         shp->shm_perm.security = NULL;
 341         error = security_shm_alloc(shp);
 342         if (error) {
 343                 ipc_rcu_putref(shp);
 344                 return error;
 345         }
 346
 347         if (shmflg & SHM_HUGETLB) {
 348                 /* hugetlb_zero_setup takes care of mlock user accounting */
 349                 file = hugetlb_zero_setup(size);
 350                 shp->mlock_user = current->user;
 351         } else {
 352                 int acctflag = VM_ACCOUNT;
 353                 /*
 354                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 355                  * if it's asked for.
 356                  */
 357                 if  ((shmflg & SHM_NORESERVE) &&
 358                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 359                         acctflag = 0;
 360                 sprintf (name, "SYSV%08x", key);
 361                 file = shmem_file_setup(name, size, acctflag);
 362         }
 363         error = PTR_ERR(file);
 364         if (IS_ERR(file))
 365                 goto no_file;
 366
 367         error = -ENOSPC;
 368         id = shm_addid(ns, shp);
 369         if(id == -1)
 370                 goto no_id;
 371
 372         shp->shm_cprid = current->tgid;
 373         shp->shm_lprid = 0;
 374         shp->shm_atim = shp->shm_dtim = 0;
 375         shp->shm_ctim = get_seconds();
 376         shp->shm_segsz = size;
 377         shp->shm_nattch = 0;
 378         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 379         shp->shm_file = file;
 380
 381         ns->shm_tot += numpages;
 382         shm_unlock(shp);
 383         return shp->id;
 384
 385 no_id:
 386         fput(file);
 387 no_file:
 388         security_shm_free(shp);
 389         ipc_rcu_putref(shp);
 390         return error;
 391 }
 392
 393 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 394 {
 395         struct shmid_kernel *shp;
 396         int err, id = 0;
 397         struct ipc_namespace *ns;
 398
 399         ns = current->nsproxy->ipc_ns;
 400
 401         mutex_lock(&shm_ids(ns).mutex);
 402         if (key == IPC_PRIVATE) {
 403                 err = newseg(ns, key, shmflg, size);
 404         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 405                 if (!(shmflg & IPC_CREAT))
 406                         err = -ENOENT;
 407                 else
 408                         err = newseg(ns, key, shmflg, size);
 409         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 410                 err = -EEXIST;
 411         } else {
 412                 shp = shm_lock(ns, id);
 413                 BUG_ON(shp==NULL);
 414                 if (shp->shm_segsz < size)
 415                         err = -EINVAL;
 416                 else if (ipcperms(&shp->shm_perm, shmflg))
 417                         err = -EACCES;
 418                 else {
 419                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 420                         err = security_shm_associate(shp, shmflg);
 421                         if (!err)
 422                                 err = shmid;
 423                 }
 424                 shm_unlock(shp);
 425         }
 426         mutex_unlock(&shm_ids(ns).mutex);
 427
 428         return err;
 429 }
 430
 431 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 432 {
 433         switch(version) {
 434         case IPC_64:
 435                 return copy_to_user(buf, in, sizeof(*in));
 436         case IPC_OLD:
 437             {
 438                 struct shmid_ds out;
 439
 440                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 441                 out.shm_segsz   = in->shm_segsz;
 442                 out.shm_atime   = in->shm_atime;
 443                 out.shm_dtime   = in->shm_dtime;
 444                 out.shm_ctime   = in->shm_ctime;
 445                 out.shm_cpid    = in->shm_cpid;
 446                 out.shm_lpid    = in->shm_lpid;
 447                 out.shm_nattch  = in->shm_nattch;
 448
 449                 return copy_to_user(buf, &out, sizeof(out));
 450             }
 451         default:
 452                 return -EINVAL;
 453         }
 454 }
 455
 456 struct shm_setbuf {
 457         uid_t   uid;
 458         gid_t   gid;
 459         mode_t  mode;
 460 };
 461
 462 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 463 {
 464         switch(version) {
 465         case IPC_64:
 466             {
 467                 struct shmid64_ds tbuf;
 468
 469                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 470                         return -EFAULT;
 471
 472                 out->uid        = tbuf.shm_perm.uid;
 473                 out->gid        = tbuf.shm_perm.gid;
 474                 out->mode       = tbuf.shm_perm.mode;
 475
 476                 return 0;
 477             }
 478         case IPC_OLD:
 479             {
 480                 struct shmid_ds tbuf_old;
 481
 482                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 483                         return -EFAULT;
 484
 485                 out->uid        = tbuf_old.shm_perm.uid;
 486                 out->gid        = tbuf_old.shm_perm.gid;
 487                 out->mode       = tbuf_old.shm_perm.mode;
 488
 489                 return 0;
 490             }
 491         default:
 492                 return -EINVAL;
 493         }
 494 }
 495
 496 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 497 {
 498         switch(version) {
 499         case IPC_64:
 500                 return copy_to_user(buf, in, sizeof(*in));
 501         case IPC_OLD:
 502             {
 503                 struct shminfo out;
 504
 505                 if(in->shmmax > INT_MAX)
 506                         out.shmmax = INT_MAX;
 507                 else
 508                         out.shmmax = (int)in->shmmax;
 509
 510                 out.shmmin      = in->shmmin;
 511                 out.shmmni      = in->shmmni;
 512                 out.shmseg      = in->shmseg;
 513                 out.shmall      = in->shmall;
 514
 515                 return copy_to_user(buf, &out, sizeof(out));
 516             }
 517         default:
 518                 return -EINVAL;
 519         }
 520 }
 521
 522 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 523                 unsigned long *swp)
 524 {
 525         int i;
 526
 527         *rss = 0;
 528         *swp = 0;
 529
 530         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 531                 struct shmid_kernel *shp;
 532                 struct inode *inode;
 533
 534                 shp = shm_get(ns, i);
 535                 if(!shp)
 536                         continue;
 537
 538                 inode = shp->shm_file->f_path.dentry->d_inode;
 539
 540                 if (is_file_hugepages(shp->shm_file)) {
 541                         struct address_space *mapping = inode->i_mapping;
 542                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 543                 } else {
 544                         struct shmem_inode_info *info = SHMEM_I(inode);
 545                         spin_lock(&info->lock);
 546                         *rss += inode->i_mapping->nrpages;
 547                         *swp += info->swapped;
 548                         spin_unlock(&info->lock);
 549                 }
 550         }
 551 }
 552
 553 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 554 {
 555         struct shm_setbuf setbuf;
 556         struct shmid_kernel *shp;
 557         int err, version;
 558         struct ipc_namespace *ns;
 559
 560         if (cmd < 0 || shmid < 0) {
 561                 err = -EINVAL;
 562                 goto out;
 563         }
 564
 565         version = ipc_parse_version(&cmd);
 566         ns = current->nsproxy->ipc_ns;
 567
 568         switch (cmd) { /* replace with proc interface ? */
 569         case IPC_INFO:
 570         {
 571                 struct shminfo64 shminfo;
 572
 573                 err = security_shm_shmctl(NULL, cmd);
 574                 if (err)
 575                         return err;
 576
 577                 memset(&shminfo,0,sizeof(shminfo));
 578                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 579                 shminfo.shmmax = ns->shm_ctlmax;
 580                 shminfo.shmall = ns->shm_ctlall;
 581
 582                 shminfo.shmmin = SHMMIN;
 583                 if(copy_shminfo_to_user (buf, &shminfo, version))
 584                         return -EFAULT;
 585                 /* reading a integer is always atomic */
 586                 err= shm_ids(ns).max_id;
 587                 if(err<0)
 588                         err = 0;
 589                 goto out;
 590         }
 591         case SHM_INFO:
 592         {
 593                 struct shm_info shm_info;
 594
 595                 err = security_shm_shmctl(NULL, cmd);
 596                 if (err)
 597                         return err;
 598
 599                 memset(&shm_info,0,sizeof(shm_info));
 600                 mutex_lock(&shm_ids(ns).mutex);
 601                 shm_info.used_ids = shm_ids(ns).in_use;
 602                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 603                 shm_info.shm_tot = ns->shm_tot;
 604                 shm_info.swap_attempts = 0;
 605                 shm_info.swap_successes = 0;
 606                 err = shm_ids(ns).max_id;
 607                 mutex_unlock(&shm_ids(ns).mutex);
 608                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 609                         err = -EFAULT;
 610                         goto out;
 611                 }
 612
 613                 err = err < 0 ? 0 : err;
 614                 goto out;
 615         }
 616         case SHM_STAT:
 617         case IPC_STAT:
 618         {
 619                 struct shmid64_ds tbuf;
 620                 int result;
 621                 memset(&tbuf, 0, sizeof(tbuf));
 622                 shp = shm_lock(ns, shmid);
 623                 if(shp==NULL) {
 624                         err = -EINVAL;
 625                         goto out;
 626                 } else if(cmd==SHM_STAT) {
 627                         err = -EINVAL;
 628                         if (shmid > shm_ids(ns).max_id)
 629                                 goto out_unlock;
 630                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 631                 } else {
 632                         err = shm_checkid(ns, shp,shmid);
 633                         if(err)
 634                                 goto out_unlock;
 635                         result = 0;
 636                 }
 637                 err=-EACCES;
 638                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 639                         goto out_unlock;
 640                 err = security_shm_shmctl(shp, cmd);
 641                 if (err)
 642                         goto out_unlock;
 643                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 644                 tbuf.shm_segsz  = shp->shm_segsz;
 645                 tbuf.shm_atime  = shp->shm_atim;
 646                 tbuf.shm_dtime  = shp->shm_dtim;
 647                 tbuf.shm_ctime  = shp->shm_ctim;
 648                 tbuf.shm_cpid   = shp->shm_cprid;
 649                 tbuf.shm_lpid   = shp->shm_lprid;
 650                 tbuf.shm_nattch = shp->shm_nattch;
 651                 shm_unlock(shp);
 652                 if(copy_shmid_to_user (buf, &tbuf, version))
 653                         err = -EFAULT;
 654                 else
 655                         err = result;
 656                 goto out;
 657         }
 658         case SHM_LOCK:
 659         case SHM_UNLOCK:
 660         {
 661                 shp = shm_lock(ns, shmid);
 662                 if(shp==NULL) {
 663                         err = -EINVAL;
 664                         goto out;
 665                 }
 666                 err = shm_checkid(ns, shp,shmid);
 667                 if(err)
 668                         goto out_unlock;
 669
 670                 err = audit_ipc_obj(&(shp->shm_perm));
 671                 if (err)
 672                         goto out_unlock;
 673
 674                 if (!capable(CAP_IPC_LOCK)) {
 675                         err = -EPERM;
 676                         if (current->euid != shp->shm_perm.uid &&
 677                             current->euid != shp->shm_perm.cuid)
 678                                 goto out_unlock;
 679                         if (cmd == SHM_LOCK &&
 680                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 681                                 goto out_unlock;
 682                 }
 683
 684                 err = security_shm_shmctl(shp, cmd);
 685                 if (err)
 686                         goto out_unlock;
 687
 688                 if(cmd==SHM_LOCK) {
 689                         struct user_struct * user = current->user;
 690                         if (!is_file_hugepages(shp->shm_file)) {
 691                                 err = shmem_lock(shp->shm_file, 1, user);
 692                                 if (!err) {
 693                                         shp->shm_perm.mode |= SHM_LOCKED;
 694                                         shp->mlock_user = user;
 695                                 }
 696                         }
 697                 } else if (!is_file_hugepages(shp->shm_file)) {
 698                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 699                         shp->shm_perm.mode &= ~SHM_LOCKED;
 700                         shp->mlock_user = NULL;
 701                 }
 702                 shm_unlock(shp);
 703                 goto out;
 704         }
 705         case IPC_RMID:
 706         {
 707                 /*
 708                  *      We cannot simply remove the file. The SVID states
 709                  *      that the block remains until the last person
 710                  *      detaches from it, then is deleted. A shmat() on
 711                  *      an RMID segment is legal in older Linux and if
 712                  *      we change it apps break...
 713                  *
 714                  *      Instead we set a destroyed flag, and then blow
 715                  *      the name away when the usage hits zero.
 716                  */
 717                 mutex_lock(&shm_ids(ns).mutex);
 718                 shp = shm_lock(ns, shmid);
 719                 err = -EINVAL;
 720                 if (shp == NULL)
 721                         goto out_up;
 722                 err = shm_checkid(ns, shp, shmid);
 723                 if(err)
 724                         goto out_unlock_up;
 725
 726                 err = audit_ipc_obj(&(shp->shm_perm));
 727                 if (err)
 728                         goto out_unlock_up;
 729
 730                 if (current->euid != shp->shm_perm.uid &&
 731                     current->euid != shp->shm_perm.cuid &&
 732                     !capable(CAP_SYS_ADMIN)) {
 733                         err=-EPERM;
 734                         goto out_unlock_up;
 735                 }
 736
 737                 err = security_shm_shmctl(shp, cmd);
 738                 if (err)
 739                         goto out_unlock_up;
 740
 741                 do_shm_rmid(ns, shp);
 742                 mutex_unlock(&shm_ids(ns).mutex);
 743                 goto out;
 744         }
 745
 746         case IPC_SET:
 747         {
 748                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 749                         err = -EFAULT;
 750                         goto out;
 751                 }
 752                 mutex_lock(&shm_ids(ns).mutex);
 753                 shp = shm_lock(ns, shmid);
 754                 err=-EINVAL;
 755                 if(shp==NULL)
 756                         goto out_up;
 757                 err = shm_checkid(ns, shp,shmid);
 758                 if(err)
 759                         goto out_unlock_up;
 760                 err = audit_ipc_obj(&(shp->shm_perm));
 761                 if (err)
 762                         goto out_unlock_up;
 763                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 764                 if (err)
 765                         goto out_unlock_up;
 766                 err=-EPERM;
 767                 if (current->euid != shp->shm_perm.uid &&
 768                     current->euid != shp->shm_perm.cuid &&
 769                     !capable(CAP_SYS_ADMIN)) {
 770                         goto out_unlock_up;
 771                 }
 772
 773                 err = security_shm_shmctl(shp, cmd);
 774                 if (err)
 775                         goto out_unlock_up;
 776
 777                 shp->shm_perm.uid = setbuf.uid;
 778                 shp->shm_perm.gid = setbuf.gid;
 779                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 780                         | (setbuf.mode & S_IRWXUGO);
 781                 shp->shm_ctim = get_seconds();
 782                 break;
 783         }
 784
 785         default:
 786                 err = -EINVAL;
 787                 goto out;
 788         }
 789
 790         err = 0;
 791 out_unlock_up:
 792         shm_unlock(shp);
 793 out_up:
 794         mutex_unlock(&shm_ids(ns).mutex);
 795         goto out;
 796 out_unlock:
 797         shm_unlock(shp);
 798 out:
 799         return err;
 800 }
 801
 802 /*
 803  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 804  *
 805  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 806  * "raddr" thing points to kernel space, and there has to be a wrapper around
 807  * this.
 808  */
 809 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 810 {
 811         struct shmid_kernel *shp;
 812         unsigned long addr;
 813         unsigned long size;
 814         struct file * file;
 815         int    err;
 816         unsigned long flags;
 817         unsigned long prot;
 818         int acc_mode;
 819         unsigned long user_addr;
 820         struct ipc_namespace *ns;
 821         struct shm_file_data *sfd;
 822         struct path path;
 823         mode_t f_mode;
 824
 825         err = -EINVAL;
 826         if (shmid < 0)
 827                 goto out;
 828         else if ((addr = (ulong)shmaddr)) {
 829                 if (addr & (SHMLBA-1)) {
 830                         if (shmflg & SHM_RND)
 831                                 addr &= ~(SHMLBA-1);       /* round down */
 832                         else
 833 #ifndef __ARCH_FORCE_SHMLBA
 834                                 if (addr & ~PAGE_MASK)
 835 #endif
 836                                         goto out;
 837                 }
 838                 flags = MAP_SHARED | MAP_FIXED;
 839         } else {
 840                 if ((shmflg & SHM_REMAP))
 841                         goto out;
 842
 843                 flags = MAP_SHARED;
 844         }
 845
 846         if (shmflg & SHM_RDONLY) {
 847                 prot = PROT_READ;
 848                 acc_mode = S_IRUGO;
 849                 f_mode = FMODE_READ;
 850         } else {
 851                 prot = PROT_READ | PROT_WRITE;
 852                 acc_mode = S_IRUGO | S_IWUGO;
 853                 f_mode = FMODE_READ | FMODE_WRITE;
 854         }
 855         if (shmflg & SHM_EXEC) {
 856                 prot |= PROT_EXEC;
 857                 acc_mode |= S_IXUGO;
 858         }
 859
 860         /*
 861          * We cannot rely on the fs check since SYSV IPC does have an
 862          * additional creator id...
 863          */
 864         ns = current->nsproxy->ipc_ns;
 865         shp = shm_lock(ns, shmid);
 866         if(shp == NULL)
 867                 goto out;
 868
 869         err = shm_checkid(ns, shp,shmid);
 870         if (err)
 871                 goto out_unlock;
 872
 873         err = -EACCES;
 874         if (ipcperms(&shp->shm_perm, acc_mode))
 875                 goto out_unlock;
 876
 877         err = security_shm_shmat(shp, shmaddr, shmflg);
 878         if (err)
 879                 goto out_unlock;
 880
 881         path.dentry = dget(shp->shm_file->f_path.dentry);
 882         path.mnt    = mntget(shp->shm_file->f_path.mnt);
 883         shp->shm_nattch++;
 884         size = i_size_read(path.dentry->d_inode);
 885         shm_unlock(shp);
 886
 887         err = -ENOMEM;
 888         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 889         if (!sfd)
 890                 goto out_put_path;
 891
 892         err = -ENOMEM;
 893         file = get_empty_filp();
 894         if (!file)
 895                 goto out_free;
 896
 897         file->f_op = &shm_file_operations;
 898         file->private_data = sfd;
 899         file->f_path = path;
 900         file->f_mapping = shp->shm_file->f_mapping;
 901         file->f_mode = f_mode;
 902         sfd->id = shp->id;
 903         sfd->ns = get_ipc_ns(ns);
 904         sfd->file = shp->shm_file;
 905         sfd->vm_ops = NULL;
 906
 907         down_write(&current->mm->mmap_sem);
 908         if (addr && !(shmflg & SHM_REMAP)) {
 909                 err = -EINVAL;
 910                 if (find_vma_intersection(current->mm, addr, addr + size))
 911                         goto invalid;
 912                 /*
 913                  * If shm segment goes below stack, make sure there is some
 914                  * space left for the stack to grow (at least 4 pages).
 915                  */
 916                 if (addr < current->mm->start_stack &&
 917                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 918                         goto invalid;
 919         }
 920
 921         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 922         *raddr = user_addr;
 923         err = 0;
 924         if (IS_ERR_VALUE(user_addr))
 925                 err = (long)user_addr;
 926 invalid:
 927         up_write(&current->mm->mmap_sem);
 928
 929         fput(file);
 930
 931 out_nattch:
 932         mutex_lock(&shm_ids(ns).mutex);
 933         shp = shm_lock(ns, shmid);
 934         BUG_ON(!shp);
 935         shp->shm_nattch--;
 936         if(shp->shm_nattch == 0 &&
 937            shp->shm_perm.mode & SHM_DEST)
 938                 shm_destroy(ns, shp);
 939         else
 940                 shm_unlock(shp);
 941         mutex_unlock(&shm_ids(ns).mutex);
 942
 943 out:
 944         return err;
 945
 946 out_unlock:
 947         shm_unlock(shp);
 948         goto out;
 949
 950 out_free:
 951         kfree(sfd);
 952 out_put_path:
 953         dput(path.dentry);
 954         mntput(path.mnt);
 955         goto out_nattch;
 956 }
 957
 958 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 959 {
 960         unsigned long ret;
 961         long err;
 962
 963         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 964         if (err)
 965                 return err;
 966         force_successful_syscall_return();
 967         return (long)ret;
 968 }
 969
 970 /*
 971  * detach and kill segment if marked destroyed.
 972  * The work is done in shm_close.
 973  */
 974 asmlinkage long sys_shmdt(char __user *shmaddr)
 975 {
 976         struct mm_struct *mm = current->mm;
 977         struct vm_area_struct *vma, *next;
 978         unsigned long addr = (unsigned long)shmaddr;
 979         loff_t size = 0;
 980         int retval = -EINVAL;
 981
 982         if (addr & ~PAGE_MASK)
 983                 return retval;
 984
 985         down_write(&mm->mmap_sem);
 986
 987         /*
 988          * This function tries to be smart and unmap shm segments that
 989          * were modified by partial mlock or munmap calls:
 990          * - It first determines the size of the shm segment that should be
 991          *   unmapped: It searches for a vma that is backed by shm and that
 992          *   started at address shmaddr. It records it's size and then unmaps
 993          *   it.
 994          * - Then it unmaps all shm vmas that started at shmaddr and that
 995          *   are within the initially determined size.
 996          * Errors from do_munmap are ignored: the function only fails if
 997          * it's called with invalid parameters or if it's called to unmap
 998          * a part of a vma. Both calls in this function are for full vmas,
 999          * the parameters are directly copied from the vma itself and always
1000          * valid - therefore do_munmap cannot fail. (famous last words?)
1001          */
1002         /*
1003          * If it had been mremap()'d, the starting address would not
1004          * match the usual checks anyway. So assume all vma's are
1005          * above the starting address given.
1006          */
1007         vma = find_vma(mm, addr);
1008
1009         while (vma) {
1010                 next = vma->vm_next;
1011
1012                 /*
1013                  * Check if the starting address would match, i.e. it's
1014                  * a fragment created by mprotect() and/or munmap(), or it
1015                  * otherwise it starts at this address with no hassles.
1016                  */
1017                 if ((vma->vm_ops == &shm_vm_ops) &&
1018                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1019
1020
1021                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1022                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1023                         /*
1024                          * We discovered the size of the shm segment, so
1025                          * break out of here and fall through to the next
1026                          * loop that uses the size information to stop
1027                          * searching for matching vma's.
1028                          */
1029                         retval = 0;
1030                         vma = next;
1031                         break;
1032                 }
1033                 vma = next;
1034         }
1035
1036         /*
1037          * We need look no further than the maximum address a fragment
1038          * could possibly have landed at. Also cast things to loff_t to
1039          * prevent overflows and make comparisions vs. equal-width types.
1040          */
1041         size = PAGE_ALIGN(size);
1042         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1043                 next = vma->vm_next;
1044
1045                 /* finding a matching vma now does not alter retval */
1046                 if ((vma->vm_ops == &shm_vm_ops) &&
1047                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1048
1049                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1050                 vma = next;
1051         }
1052
1053         up_write(&mm->mmap_sem);
1054         return retval;
1055 }
1056
1057 #ifdef CONFIG_PROC_FS
1058 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1059 {
1060         struct shmid_kernel *shp = it;
1061         char *format;
1062
1063 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1064 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1065
1066         if (sizeof(size_t) <= sizeof(int))
1067                 format = SMALL_STRING;
1068         else
1069                 format = BIG_STRING;
1070         return seq_printf(s, format,
1071                           shp->shm_perm.key,
1072                           shp->id,
1073                           shp->shm_perm.mode,
1074                           shp->shm_segsz,
1075                           shp->shm_cprid,
1076                           shp->shm_lprid,
1077                           shp->shm_nattch,
1078                           shp->shm_perm.uid,
1079                           shp->shm_perm.gid,
1080                           shp->shm_perm.cuid,
1081                           shp->shm_perm.cgid,
1082                           shp->shm_atim,
1083                           shp->shm_dtim,
1084                           shp->shm_ctim);
1085 }
1086 #endif