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 static struct page *shm_nopage(struct vm_area_struct *vma,
 230                                unsigned long address, 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 if (vma->vm_policy)
 258                 pol = vma->vm_policy;
 259         else
 260                 pol = current->mempolicy;
 261         return pol;
 262 }
 263 #endif
 264
 265 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 266 {
 267         struct shm_file_data *sfd = shm_file_data(file);
 268         int ret;
 269
 270         ret = sfd->file->f_op->mmap(sfd->file, vma);
 271         if (ret != 0)
 272                 return ret;
 273         sfd->vm_ops = vma->vm_ops;
 274         vma->vm_ops = &shm_vm_ops;
 275         shm_open(vma);
 276
 277         return ret;
 278 }
 279
 280 static int shm_release(struct inode *ino, struct file *file)
 281 {
 282         struct shm_file_data *sfd = shm_file_data(file);
 283
 284         put_ipc_ns(sfd->ns);
 285         shm_file_data(file) = NULL;
 286         kfree(sfd);
 287         return 0;
 288 }
 289
 290 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 291 {
 292         int (*fsync) (struct file *, struct dentry *, int datasync);
 293         struct shm_file_data *sfd = shm_file_data(file);
 294         int ret = -EINVAL;
 295
 296         fsync = sfd->file->f_op->fsync;
 297         if (fsync)
 298                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 299         return ret;
 300 }
 301
 302 static unsigned long shm_get_unmapped_area(struct file *file,
 303         unsigned long addr, unsigned long len, unsigned long pgoff,
 304         unsigned long flags)
 305 {
 306         struct shm_file_data *sfd = shm_file_data(file);
 307         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 308 }
 309
 310 int is_file_shm_hugepages(struct file *file)
 311 {
 312         int ret = 0;
 313
 314         if (file->f_op == &shm_file_operations) {
 315                 struct shm_file_data *sfd;
 316                 sfd = shm_file_data(file);
 317                 ret = is_file_hugepages(sfd->file);
 318         }
 319         return ret;
 320 }
 321
 322 static const struct file_operations shm_file_operations = {
 323         .mmap           = shm_mmap,
 324         .fsync          = shm_fsync,
 325         .release        = shm_release,
 326         .get_unmapped_area      = shm_get_unmapped_area,
 327 };
 328
 329 static struct vm_operations_struct shm_vm_ops = {
 330         .open   = shm_open,     /* callback for a new vm-area open */
 331         .close  = shm_close,    /* callback for when the vm-area is released */
 332         .nopage = shm_nopage,
 333 #if defined(CONFIG_NUMA)
 334         .set_policy = shm_set_policy,
 335         .get_policy = shm_get_policy,
 336 #endif
 337 };
 338
 339 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 340 {
 341         int error;
 342         struct shmid_kernel *shp;
 343         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 344         struct file * file;
 345         char name[13];
 346         int id;
 347
 348         if (size < SHMMIN || size > ns->shm_ctlmax)
 349                 return -EINVAL;
 350
 351         if (ns->shm_tot + numpages > ns->shm_ctlall)
 352                 return -ENOSPC;
 353
 354         shp = ipc_rcu_alloc(sizeof(*shp));
 355         if (!shp)
 356                 return -ENOMEM;
 357
 358         shp->shm_perm.key = key;
 359         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 360         shp->mlock_user = NULL;
 361
 362         shp->shm_perm.security = NULL;
 363         error = security_shm_alloc(shp);
 364         if (error) {
 365                 ipc_rcu_putref(shp);
 366                 return error;
 367         }
 368
 369         if (shmflg & SHM_HUGETLB) {
 370                 /* hugetlb_zero_setup takes care of mlock user accounting */
 371                 file = hugetlb_zero_setup(size);
 372                 shp->mlock_user = current->user;
 373         } else {
 374                 int acctflag = VM_ACCOUNT;
 375                 /*
 376                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 377                  * if it's asked for.
 378                  */
 379                 if  ((shmflg & SHM_NORESERVE) &&
 380                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 381                         acctflag = 0;
 382                 sprintf (name, "SYSV%08x", key);
 383                 file = shmem_file_setup(name, size, acctflag);
 384         }
 385         error = PTR_ERR(file);
 386         if (IS_ERR(file))
 387                 goto no_file;
 388
 389         error = -ENOSPC;
 390         id = shm_addid(ns, shp);
 391         if(id == -1)
 392                 goto no_id;
 393
 394         shp->shm_cprid = current->tgid;
 395         shp->shm_lprid = 0;
 396         shp->shm_atim = shp->shm_dtim = 0;
 397         shp->shm_ctim = get_seconds();
 398         shp->shm_segsz = size;
 399         shp->shm_nattch = 0;
 400         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 401         shp->shm_file = file;
 402
 403         ns->shm_tot += numpages;
 404         shm_unlock(shp);
 405         return shp->id;
 406
 407 no_id:
 408         fput(file);
 409 no_file:
 410         security_shm_free(shp);
 411         ipc_rcu_putref(shp);
 412         return error;
 413 }
 414
 415 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 416 {
 417         struct shmid_kernel *shp;
 418         int err, id = 0;
 419         struct ipc_namespace *ns;
 420
 421         ns = current->nsproxy->ipc_ns;
 422
 423         mutex_lock(&shm_ids(ns).mutex);
 424         if (key == IPC_PRIVATE) {
 425                 err = newseg(ns, key, shmflg, size);
 426         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 427                 if (!(shmflg & IPC_CREAT))
 428                         err = -ENOENT;
 429                 else
 430                         err = newseg(ns, key, shmflg, size);
 431         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 432                 err = -EEXIST;
 433         } else {
 434                 shp = shm_lock(ns, id);
 435                 BUG_ON(shp==NULL);
 436                 if (shp->shm_segsz < size)
 437                         err = -EINVAL;
 438                 else if (ipcperms(&shp->shm_perm, shmflg))
 439                         err = -EACCES;
 440                 else {
 441                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 442                         err = security_shm_associate(shp, shmflg);
 443                         if (!err)
 444                                 err = shmid;
 445                 }
 446                 shm_unlock(shp);
 447         }
 448         mutex_unlock(&shm_ids(ns).mutex);
 449
 450         return err;
 451 }
 452
 453 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 454 {
 455         switch(version) {
 456         case IPC_64:
 457                 return copy_to_user(buf, in, sizeof(*in));
 458         case IPC_OLD:
 459             {
 460                 struct shmid_ds out;
 461
 462                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 463                 out.shm_segsz   = in->shm_segsz;
 464                 out.shm_atime   = in->shm_atime;
 465                 out.shm_dtime   = in->shm_dtime;
 466                 out.shm_ctime   = in->shm_ctime;
 467                 out.shm_cpid    = in->shm_cpid;
 468                 out.shm_lpid    = in->shm_lpid;
 469                 out.shm_nattch  = in->shm_nattch;
 470
 471                 return copy_to_user(buf, &out, sizeof(out));
 472             }
 473         default:
 474                 return -EINVAL;
 475         }
 476 }
 477
 478 struct shm_setbuf {
 479         uid_t   uid;
 480         gid_t   gid;
 481         mode_t  mode;
 482 };
 483
 484 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 485 {
 486         switch(version) {
 487         case IPC_64:
 488             {
 489                 struct shmid64_ds tbuf;
 490
 491                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 492                         return -EFAULT;
 493
 494                 out->uid        = tbuf.shm_perm.uid;
 495                 out->gid        = tbuf.shm_perm.gid;
 496                 out->mode       = tbuf.shm_perm.mode;
 497
 498                 return 0;
 499             }
 500         case IPC_OLD:
 501             {
 502                 struct shmid_ds tbuf_old;
 503
 504                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 505                         return -EFAULT;
 506
 507                 out->uid        = tbuf_old.shm_perm.uid;
 508                 out->gid        = tbuf_old.shm_perm.gid;
 509                 out->mode       = tbuf_old.shm_perm.mode;
 510
 511                 return 0;
 512             }
 513         default:
 514                 return -EINVAL;
 515         }
 516 }
 517
 518 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 519 {
 520         switch(version) {
 521         case IPC_64:
 522                 return copy_to_user(buf, in, sizeof(*in));
 523         case IPC_OLD:
 524             {
 525                 struct shminfo out;
 526
 527                 if(in->shmmax > INT_MAX)
 528                         out.shmmax = INT_MAX;
 529                 else
 530                         out.shmmax = (int)in->shmmax;
 531
 532                 out.shmmin      = in->shmmin;
 533                 out.shmmni      = in->shmmni;
 534                 out.shmseg      = in->shmseg;
 535                 out.shmall      = in->shmall;
 536
 537                 return copy_to_user(buf, &out, sizeof(out));
 538             }
 539         default:
 540                 return -EINVAL;
 541         }
 542 }
 543
 544 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 545                 unsigned long *swp)
 546 {
 547         int i;
 548
 549         *rss = 0;
 550         *swp = 0;
 551
 552         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 553                 struct shmid_kernel *shp;
 554                 struct inode *inode;
 555
 556                 shp = shm_get(ns, i);
 557                 if(!shp)
 558                         continue;
 559
 560                 inode = shp->shm_file->f_path.dentry->d_inode;
 561
 562                 if (is_file_hugepages(shp->shm_file)) {
 563                         struct address_space *mapping = inode->i_mapping;
 564                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 565                 } else {
 566                         struct shmem_inode_info *info = SHMEM_I(inode);
 567                         spin_lock(&info->lock);
 568                         *rss += inode->i_mapping->nrpages;
 569                         *swp += info->swapped;
 570                         spin_unlock(&info->lock);
 571                 }
 572         }
 573 }
 574
 575 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 576 {
 577         struct shm_setbuf setbuf;
 578         struct shmid_kernel *shp;
 579         int err, version;
 580         struct ipc_namespace *ns;
 581
 582         if (cmd < 0 || shmid < 0) {
 583                 err = -EINVAL;
 584                 goto out;
 585         }
 586
 587         version = ipc_parse_version(&cmd);
 588         ns = current->nsproxy->ipc_ns;
 589
 590         switch (cmd) { /* replace with proc interface ? */
 591         case IPC_INFO:
 592         {
 593                 struct shminfo64 shminfo;
 594
 595                 err = security_shm_shmctl(NULL, cmd);
 596                 if (err)
 597                         return err;
 598
 599                 memset(&shminfo,0,sizeof(shminfo));
 600                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 601                 shminfo.shmmax = ns->shm_ctlmax;
 602                 shminfo.shmall = ns->shm_ctlall;
 603
 604                 shminfo.shmmin = SHMMIN;
 605                 if(copy_shminfo_to_user (buf, &shminfo, version))
 606                         return -EFAULT;
 607                 /* reading a integer is always atomic */
 608                 err= shm_ids(ns).max_id;
 609                 if(err<0)
 610                         err = 0;
 611                 goto out;
 612         }
 613         case SHM_INFO:
 614         {
 615                 struct shm_info shm_info;
 616
 617                 err = security_shm_shmctl(NULL, cmd);
 618                 if (err)
 619                         return err;
 620
 621                 memset(&shm_info,0,sizeof(shm_info));
 622                 mutex_lock(&shm_ids(ns).mutex);
 623                 shm_info.used_ids = shm_ids(ns).in_use;
 624                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 625                 shm_info.shm_tot = ns->shm_tot;
 626                 shm_info.swap_attempts = 0;
 627                 shm_info.swap_successes = 0;
 628                 err = shm_ids(ns).max_id;
 629                 mutex_unlock(&shm_ids(ns).mutex);
 630                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 631                         err = -EFAULT;
 632                         goto out;
 633                 }
 634
 635                 err = err < 0 ? 0 : err;
 636                 goto out;
 637         }
 638         case SHM_STAT:
 639         case IPC_STAT:
 640         {
 641                 struct shmid64_ds tbuf;
 642                 int result;
 643                 memset(&tbuf, 0, sizeof(tbuf));
 644                 shp = shm_lock(ns, shmid);
 645                 if(shp==NULL) {
 646                         err = -EINVAL;
 647                         goto out;
 648                 } else if(cmd==SHM_STAT) {
 649                         err = -EINVAL;
 650                         if (shmid > shm_ids(ns).max_id)
 651                                 goto out_unlock;
 652                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 653                 } else {
 654                         err = shm_checkid(ns, shp,shmid);
 655                         if(err)
 656                                 goto out_unlock;
 657                         result = 0;
 658                 }
 659                 err=-EACCES;
 660                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 661                         goto out_unlock;
 662                 err = security_shm_shmctl(shp, cmd);
 663                 if (err)
 664                         goto out_unlock;
 665                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 666                 tbuf.shm_segsz  = shp->shm_segsz;
 667                 tbuf.shm_atime  = shp->shm_atim;
 668                 tbuf.shm_dtime  = shp->shm_dtim;
 669                 tbuf.shm_ctime  = shp->shm_ctim;
 670                 tbuf.shm_cpid   = shp->shm_cprid;
 671                 tbuf.shm_lpid   = shp->shm_lprid;
 672                 tbuf.shm_nattch = shp->shm_nattch;
 673                 shm_unlock(shp);
 674                 if(copy_shmid_to_user (buf, &tbuf, version))
 675                         err = -EFAULT;
 676                 else
 677                         err = result;
 678                 goto out;
 679         }
 680         case SHM_LOCK:
 681         case SHM_UNLOCK:
 682         {
 683                 shp = shm_lock(ns, shmid);
 684                 if(shp==NULL) {
 685                         err = -EINVAL;
 686                         goto out;
 687                 }
 688                 err = shm_checkid(ns, shp,shmid);
 689                 if(err)
 690                         goto out_unlock;
 691
 692                 err = audit_ipc_obj(&(shp->shm_perm));
 693                 if (err)
 694                         goto out_unlock;
 695
 696                 if (!capable(CAP_IPC_LOCK)) {
 697                         err = -EPERM;
 698                         if (current->euid != shp->shm_perm.uid &&
 699                             current->euid != shp->shm_perm.cuid)
 700                                 goto out_unlock;
 701                         if (cmd == SHM_LOCK &&
 702                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 703                                 goto out_unlock;
 704                 }
 705
 706                 err = security_shm_shmctl(shp, cmd);
 707                 if (err)
 708                         goto out_unlock;
 709
 710                 if(cmd==SHM_LOCK) {
 711                         struct user_struct * user = current->user;
 712                         if (!is_file_hugepages(shp->shm_file)) {
 713                                 err = shmem_lock(shp->shm_file, 1, user);
 714                                 if (!err) {
 715                                         shp->shm_perm.mode |= SHM_LOCKED;
 716                                         shp->mlock_user = user;
 717                                 }
 718                         }
 719                 } else if (!is_file_hugepages(shp->shm_file)) {
 720                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 721                         shp->shm_perm.mode &= ~SHM_LOCKED;
 722                         shp->mlock_user = NULL;
 723                 }
 724                 shm_unlock(shp);
 725                 goto out;
 726         }
 727         case IPC_RMID:
 728         {
 729                 /*
 730                  *      We cannot simply remove the file. The SVID states
 731                  *      that the block remains until the last person
 732                  *      detaches from it, then is deleted. A shmat() on
 733                  *      an RMID segment is legal in older Linux and if
 734                  *      we change it apps break...
 735                  *
 736                  *      Instead we set a destroyed flag, and then blow
 737                  *      the name away when the usage hits zero.
 738                  */
 739                 mutex_lock(&shm_ids(ns).mutex);
 740                 shp = shm_lock(ns, shmid);
 741                 err = -EINVAL;
 742                 if (shp == NULL)
 743                         goto out_up;
 744                 err = shm_checkid(ns, shp, shmid);
 745                 if(err)
 746                         goto out_unlock_up;
 747
 748                 err = audit_ipc_obj(&(shp->shm_perm));
 749                 if (err)
 750                         goto out_unlock_up;
 751
 752                 if (current->euid != shp->shm_perm.uid &&
 753                     current->euid != shp->shm_perm.cuid &&
 754                     !capable(CAP_SYS_ADMIN)) {
 755                         err=-EPERM;
 756                         goto out_unlock_up;
 757                 }
 758
 759                 err = security_shm_shmctl(shp, cmd);
 760                 if (err)
 761                         goto out_unlock_up;
 762
 763                 do_shm_rmid(ns, shp);
 764                 mutex_unlock(&shm_ids(ns).mutex);
 765                 goto out;
 766         }
 767
 768         case IPC_SET:
 769         {
 770                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 771                         err = -EFAULT;
 772                         goto out;
 773                 }
 774                 mutex_lock(&shm_ids(ns).mutex);
 775                 shp = shm_lock(ns, shmid);
 776                 err=-EINVAL;
 777                 if(shp==NULL)
 778                         goto out_up;
 779                 err = shm_checkid(ns, shp,shmid);
 780                 if(err)
 781                         goto out_unlock_up;
 782                 err = audit_ipc_obj(&(shp->shm_perm));
 783                 if (err)
 784                         goto out_unlock_up;
 785                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 786                 if (err)
 787                         goto out_unlock_up;
 788                 err=-EPERM;
 789                 if (current->euid != shp->shm_perm.uid &&
 790                     current->euid != shp->shm_perm.cuid &&
 791                     !capable(CAP_SYS_ADMIN)) {
 792                         goto out_unlock_up;
 793                 }
 794
 795                 err = security_shm_shmctl(shp, cmd);
 796                 if (err)
 797                         goto out_unlock_up;
 798
 799                 shp->shm_perm.uid = setbuf.uid;
 800                 shp->shm_perm.gid = setbuf.gid;
 801                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 802                         | (setbuf.mode & S_IRWXUGO);
 803                 shp->shm_ctim = get_seconds();
 804                 break;
 805         }
 806
 807         default:
 808                 err = -EINVAL;
 809                 goto out;
 810         }
 811
 812         err = 0;
 813 out_unlock_up:
 814         shm_unlock(shp);
 815 out_up:
 816         mutex_unlock(&shm_ids(ns).mutex);
 817         goto out;
 818 out_unlock:
 819         shm_unlock(shp);
 820 out:
 821         return err;
 822 }
 823
 824 /*
 825  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 826  *
 827  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 828  * "raddr" thing points to kernel space, and there has to be a wrapper around
 829  * this.
 830  */
 831 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 832 {
 833         struct shmid_kernel *shp;
 834         unsigned long addr;
 835         unsigned long size;
 836         struct file * file;
 837         int    err;
 838         unsigned long flags;
 839         unsigned long prot;
 840         int acc_mode;
 841         unsigned long user_addr;
 842         struct ipc_namespace *ns;
 843         struct shm_file_data *sfd;
 844         struct path path;
 845         mode_t f_mode;
 846
 847         err = -EINVAL;
 848         if (shmid < 0)
 849                 goto out;
 850         else if ((addr = (ulong)shmaddr)) {
 851                 if (addr & (SHMLBA-1)) {
 852                         if (shmflg & SHM_RND)
 853                                 addr &= ~(SHMLBA-1);       /* round down */
 854                         else
 855 #ifndef __ARCH_FORCE_SHMLBA
 856                                 if (addr & ~PAGE_MASK)
 857 #endif
 858                                         goto out;
 859                 }
 860                 flags = MAP_SHARED | MAP_FIXED;
 861         } else {
 862                 if ((shmflg & SHM_REMAP))
 863                         goto out;
 864
 865                 flags = MAP_SHARED;
 866         }
 867
 868         if (shmflg & SHM_RDONLY) {
 869                 prot = PROT_READ;
 870                 acc_mode = S_IRUGO;
 871                 f_mode = FMODE_READ;
 872         } else {
 873                 prot = PROT_READ | PROT_WRITE;
 874                 acc_mode = S_IRUGO | S_IWUGO;
 875                 f_mode = FMODE_READ | FMODE_WRITE;
 876         }
 877         if (shmflg & SHM_EXEC) {
 878                 prot |= PROT_EXEC;
 879                 acc_mode |= S_IXUGO;
 880         }
 881
 882         /*
 883          * We cannot rely on the fs check since SYSV IPC does have an
 884          * additional creator id...
 885          */
 886         ns = current->nsproxy->ipc_ns;
 887         shp = shm_lock(ns, shmid);
 888         if(shp == NULL)
 889                 goto out;
 890
 891         err = shm_checkid(ns, shp,shmid);
 892         if (err)
 893                 goto out_unlock;
 894
 895         err = -EACCES;
 896         if (ipcperms(&shp->shm_perm, acc_mode))
 897                 goto out_unlock;
 898
 899         err = security_shm_shmat(shp, shmaddr, shmflg);
 900         if (err)
 901                 goto out_unlock;
 902
 903         path.dentry = dget(shp->shm_file->f_path.dentry);
 904         path.mnt    = mntget(shp->shm_file->f_path.mnt);
 905         shp->shm_nattch++;
 906         size = i_size_read(path.dentry->d_inode);
 907         shm_unlock(shp);
 908
 909         err = -ENOMEM;
 910         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 911         if (!sfd)
 912                 goto out_put_path;
 913
 914         err = -ENOMEM;
 915         file = get_empty_filp();
 916         if (!file)
 917                 goto out_free;
 918
 919         file->f_op = &shm_file_operations;
 920         file->private_data = sfd;
 921         file->f_path = path;
 922         file->f_mapping = shp->shm_file->f_mapping;
 923         file->f_mode = f_mode;
 924         sfd->id = shp->id;
 925         sfd->ns = get_ipc_ns(ns);
 926         sfd->file = shp->shm_file;
 927         sfd->vm_ops = NULL;
 928
 929         down_write(&current->mm->mmap_sem);
 930         if (addr && !(shmflg & SHM_REMAP)) {
 931                 err = -EINVAL;
 932                 if (find_vma_intersection(current->mm, addr, addr + size))
 933                         goto invalid;
 934                 /*
 935                  * If shm segment goes below stack, make sure there is some
 936                  * space left for the stack to grow (at least 4 pages).
 937                  */
 938                 if (addr < current->mm->start_stack &&
 939                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 940                         goto invalid;
 941         }
 942
 943         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 944         *raddr = user_addr;
 945         err = 0;
 946         if (IS_ERR_VALUE(user_addr))
 947                 err = (long)user_addr;
 948 invalid:
 949         up_write(&current->mm->mmap_sem);
 950
 951         fput(file);
 952
 953 out_nattch:
 954         mutex_lock(&shm_ids(ns).mutex);
 955         shp = shm_lock(ns, shmid);
 956         BUG_ON(!shp);
 957         shp->shm_nattch--;
 958         if(shp->shm_nattch == 0 &&
 959            shp->shm_perm.mode & SHM_DEST)
 960                 shm_destroy(ns, shp);
 961         else
 962                 shm_unlock(shp);
 963         mutex_unlock(&shm_ids(ns).mutex);
 964
 965 out:
 966         return err;
 967
 968 out_unlock:
 969         shm_unlock(shp);
 970         goto out;
 971
 972 out_free:
 973         kfree(sfd);
 974 out_put_path:
 975         dput(path.dentry);
 976         mntput(path.mnt);
 977         goto out_nattch;
 978 }
 979
 980 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 981 {
 982         unsigned long ret;
 983         long err;
 984
 985         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 986         if (err)
 987                 return err;
 988         force_successful_syscall_return();
 989         return (long)ret;
 990 }
 991
 992 /*
 993  * detach and kill segment if marked destroyed.
 994  * The work is done in shm_close.
 995  */
 996 asmlinkage long sys_shmdt(char __user *shmaddr)
 997 {
 998         struct mm_struct *mm = current->mm;
 999         struct vm_area_struct *vma, *next;
1000         unsigned long addr = (unsigned long)shmaddr;
1001         loff_t size = 0;
1002         int retval = -EINVAL;
1003
1004         if (addr & ~PAGE_MASK)
1005                 return retval;
1006
1007         down_write(&mm->mmap_sem);
1008
1009         /*
1010          * This function tries to be smart and unmap shm segments that
1011          * were modified by partial mlock or munmap calls:
1012          * - It first determines the size of the shm segment that should be
1013          *   unmapped: It searches for a vma that is backed by shm and that
1014          *   started at address shmaddr. It records it's size and then unmaps
1015          *   it.
1016          * - Then it unmaps all shm vmas that started at shmaddr and that
1017          *   are within the initially determined size.
1018          * Errors from do_munmap are ignored: the function only fails if
1019          * it's called with invalid parameters or if it's called to unmap
1020          * a part of a vma. Both calls in this function are for full vmas,
1021          * the parameters are directly copied from the vma itself and always
1022          * valid - therefore do_munmap cannot fail. (famous last words?)
1023          */
1024         /*
1025          * If it had been mremap()'d, the starting address would not
1026          * match the usual checks anyway. So assume all vma's are
1027          * above the starting address given.
1028          */
1029         vma = find_vma(mm, addr);
1030
1031         while (vma) {
1032                 next = vma->vm_next;
1033
1034                 /*
1035                  * Check if the starting address would match, i.e. it's
1036                  * a fragment created by mprotect() and/or munmap(), or it
1037                  * otherwise it starts at this address with no hassles.
1038                  */
1039                 if ((vma->vm_ops == &shm_vm_ops) &&
1040                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1041
1042
1043                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1044                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1045                         /*
1046                          * We discovered the size of the shm segment, so
1047                          * break out of here and fall through to the next
1048                          * loop that uses the size information to stop
1049                          * searching for matching vma's.
1050                          */
1051                         retval = 0;
1052                         vma = next;
1053                         break;
1054                 }
1055                 vma = next;
1056         }
1057
1058         /*
1059          * We need look no further than the maximum address a fragment
1060          * could possibly have landed at. Also cast things to loff_t to
1061          * prevent overflows and make comparisions vs. equal-width types.
1062          */
1063         size = PAGE_ALIGN(size);
1064         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1065                 next = vma->vm_next;
1066
1067                 /* finding a matching vma now does not alter retval */
1068                 if ((vma->vm_ops == &shm_vm_ops) &&
1069                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1070
1071                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1072                 vma = next;
1073         }
1074
1075         up_write(&mm->mmap_sem);
1076         return retval;
1077 }
1078
1079 #ifdef CONFIG_PROC_FS
1080 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1081 {
1082         struct shmid_kernel *shp = it;
1083         char *format;
1084
1085 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1086 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1087
1088         if (sizeof(size_t) <= sizeof(int))
1089                 format = SMALL_STRING;
1090         else
1091                 format = BIG_STRING;
1092         return seq_printf(s, format,
1093                           shp->shm_perm.key,
1094                           shp->id,
1095                           shp->shm_perm.mode,
1096                           shp->shm_segsz,
1097                           shp->shm_cprid,
1098                           shp->shm_lprid,
1099                           shp->shm_nattch,
1100                           shp->shm_perm.uid,
1101                           shp->shm_perm.gid,
1102                           shp->shm_perm.cuid,
1103                           shp->shm_perm.cgid,
1104                           shp->shm_atim,
1105                           shp->shm_dtim,
1106                           shp->shm_ctim);
1107 }
1108 #endif