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 __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 int shm_init_ns(struct ipc_namespace *ns)
 102 {
 103         struct ipc_ids *ids;
 104
 105         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 106         if (ids == NULL)
 107                 return -ENOMEM;
 108
 109         __shm_init_ns(ns, ids);
 110         return 0;
 111 }
 112
 113 void shm_exit_ns(struct ipc_namespace *ns)
 114 {
 115         int i;
 116         struct shmid_kernel *shp;
 117
 118         mutex_lock(&shm_ids(ns).mutex);
 119         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 120                 shp = shm_lock(ns, i);
 121                 if (shp == NULL)
 122                         continue;
 123
 124                 do_shm_rmid(ns, shp);
 125         }
 126         mutex_unlock(&shm_ids(ns).mutex);
 127
 128         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
 129         kfree(ns->ids[IPC_SHM_IDS]);
 130         ns->ids[IPC_SHM_IDS] = NULL;
 131 }
 132
 133 void __init shm_init (void)
 134 {
 135         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 136         ipc_init_proc_interface("sysvipc/shm",
 137                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 138                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 139 }
 140
 141 static inline int shm_checkid(struct ipc_namespace *ns,
 142                 struct shmid_kernel *s, int id)
 143 {
 144         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 145                 return -EIDRM;
 146         return 0;
 147 }
 148
 149 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
 150 {
 151         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
 152 }
 153
 154 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 155 {
 156         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 157 }
 158
 159
 160
 161 /* This is called by fork, once for every shm attach. */
 162 static void shm_open(struct vm_area_struct *vma)
 163 {
 164         struct file *file = vma->vm_file;
 165         struct shm_file_data *sfd = shm_file_data(file);
 166         struct shmid_kernel *shp;
 167
 168         shp = shm_lock(sfd->ns, sfd->id);
 169         BUG_ON(!shp);
 170         shp->shm_atim = get_seconds();
 171         shp->shm_lprid = current->tgid;
 172         shp->shm_nattch++;
 173         shm_unlock(shp);
 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         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 187         shm_rmid(ns, shp->id);
 188         shm_unlock(shp);
 189         if (!is_file_hugepages(shp->shm_file))
 190                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 191         else
 192                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 193                                                 shp->mlock_user);
 194         fput (shp->shm_file);
 195         security_shm_free(shp);
 196         ipc_rcu_putref(shp);
 197 }
 198
 199 /*
 200  * remove the attach descriptor vma.
 201  * free memory for segment if it is marked destroyed.
 202  * The descriptor has already been removed from the current->mm->mmap list
 203  * and will later be kfree()d.
 204  */
 205 static void shm_close(struct vm_area_struct *vma)
 206 {
 207         struct file * file = vma->vm_file;
 208         struct shm_file_data *sfd = shm_file_data(file);
 209         struct shmid_kernel *shp;
 210         struct ipc_namespace *ns = sfd->ns;
 211
 212         mutex_lock(&shm_ids(ns).mutex);
 213         /* remove from the list of attaches of the shm segment */
 214         shp = shm_lock(ns, sfd->id);
 215         BUG_ON(!shp);
 216         shp->shm_lprid = current->tgid;
 217         shp->shm_dtim = get_seconds();
 218         shp->shm_nattch--;
 219         if(shp->shm_nattch == 0 &&
 220            shp->shm_perm.mode & SHM_DEST)
 221                 shm_destroy(ns, shp);
 222         else
 223                 shm_unlock(shp);
 224         mutex_unlock(&shm_ids(ns).mutex);
 225 }
 226
 227 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 228 {
 229         struct file *file = vma->vm_file;
 230         struct shm_file_data *sfd = shm_file_data(file);
 231
 232         return sfd->vm_ops->fault(vma, vmf);
 233 }
 234
 235 #ifdef CONFIG_NUMA
 236 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 237 {
 238         struct file *file = vma->vm_file;
 239         struct shm_file_data *sfd = shm_file_data(file);
 240         int err = 0;
 241         if (sfd->vm_ops->set_policy)
 242                 err = sfd->vm_ops->set_policy(vma, new);
 243         return err;
 244 }
 245
 246 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 247                                         unsigned long addr)
 248 {
 249         struct file *file = vma->vm_file;
 250         struct shm_file_data *sfd = shm_file_data(file);
 251         struct mempolicy *pol = NULL;
 252
 253         if (sfd->vm_ops->get_policy)
 254                 pol = sfd->vm_ops->get_policy(vma, addr);
 255         else if (vma->vm_policy)
 256                 pol = vma->vm_policy;
 257         else
 258                 pol = current->mempolicy;
 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 #ifdef CONFIG_MMU
 273         BUG_ON(!sfd->vm_ops->fault);
 274 #endif
 275         vma->vm_ops = &shm_vm_ops;
 276         shm_open(vma);
 277
 278         return ret;
 279 }
 280
 281 static int shm_release(struct inode *ino, struct file *file)
 282 {
 283         struct shm_file_data *sfd = shm_file_data(file);
 284
 285         put_ipc_ns(sfd->ns);
 286         shm_file_data(file) = NULL;
 287         kfree(sfd);
 288         return 0;
 289 }
 290
 291 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 292 {
 293         int (*fsync) (struct file *, struct dentry *, int datasync);
 294         struct shm_file_data *sfd = shm_file_data(file);
 295         int ret = -EINVAL;
 296
 297         fsync = sfd->file->f_op->fsync;
 298         if (fsync)
 299                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 300         return ret;
 301 }
 302
 303 static unsigned long shm_get_unmapped_area(struct file *file,
 304         unsigned long addr, unsigned long len, unsigned long pgoff,
 305         unsigned long flags)
 306 {
 307         struct shm_file_data *sfd = shm_file_data(file);
 308         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 309 }
 310
 311 int is_file_shm_hugepages(struct file *file)
 312 {
 313         int ret = 0;
 314
 315         if (file->f_op == &shm_file_operations) {
 316                 struct shm_file_data *sfd;
 317                 sfd = shm_file_data(file);
 318                 ret = is_file_hugepages(sfd->file);
 319         }
 320         return ret;
 321 }
 322
 323 static const struct file_operations shm_file_operations = {
 324         .mmap           = shm_mmap,
 325         .fsync          = shm_fsync,
 326         .release        = shm_release,
 327         .get_unmapped_area      = shm_get_unmapped_area,
 328 };
 329
 330 static struct vm_operations_struct shm_vm_ops = {
 331         .open   = shm_open,     /* callback for a new vm-area open */
 332         .close  = shm_close,    /* callback for when the vm-area is released */
 333         .fault  = shm_fault,
 334 #if defined(CONFIG_NUMA)
 335         .set_policy = shm_set_policy,
 336         .get_policy = shm_get_policy,
 337 #endif
 338 };
 339
 340 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 341 {
 342         int error;
 343         struct shmid_kernel *shp;
 344         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 345         struct file * file;
 346         char name[13];
 347         int id;
 348
 349         if (size < SHMMIN || size > ns->shm_ctlmax)
 350                 return -EINVAL;
 351
 352         if (ns->shm_tot + numpages > ns->shm_ctlall)
 353                 return -ENOSPC;
 354
 355         shp = ipc_rcu_alloc(sizeof(*shp));
 356         if (!shp)
 357                 return -ENOMEM;
 358
 359         shp->shm_perm.key = key;
 360         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 361         shp->mlock_user = NULL;
 362
 363         shp->shm_perm.security = NULL;
 364         error = security_shm_alloc(shp);
 365         if (error) {
 366                 ipc_rcu_putref(shp);
 367                 return error;
 368         }
 369
 370         sprintf (name, "SYSV%08x", key);
 371         if (shmflg & SHM_HUGETLB) {
 372                 /* hugetlb_file_setup takes care of mlock user accounting */
 373                 file = hugetlb_file_setup(name, size);
 374                 shp->mlock_user = current->user;
 375         } else {
 376                 int acctflag = VM_ACCOUNT;
 377                 /*
 378                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 379                  * if it's asked for.
 380                  */
 381                 if  ((shmflg & SHM_NORESERVE) &&
 382                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 383                         acctflag = 0;
 384                 file = shmem_file_setup(name, size, acctflag);
 385         }
 386         error = PTR_ERR(file);
 387         if (IS_ERR(file))
 388                 goto no_file;
 389
 390         error = -ENOSPC;
 391         id = shm_addid(ns, shp);
 392         if(id == -1)
 393                 goto no_id;
 394
 395         shp->shm_cprid = current->tgid;
 396         shp->shm_lprid = 0;
 397         shp->shm_atim = shp->shm_dtim = 0;
 398         shp->shm_ctim = get_seconds();
 399         shp->shm_segsz = size;
 400         shp->shm_nattch = 0;
 401         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 402         shp->shm_file = file;
 403         /*
 404          * shmid gets reported as "inode#" in /proc/pid/maps.
 405          * proc-ps tools use this. Changing this will break them.
 406          */
 407         file->f_dentry->d_inode->i_ino = shp->id;
 408
 409         ns->shm_tot += numpages;
 410         shm_unlock(shp);
 411         return shp->id;
 412
 413 no_id:
 414         fput(file);
 415 no_file:
 416         security_shm_free(shp);
 417         ipc_rcu_putref(shp);
 418         return error;
 419 }
 420
 421 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 422 {
 423         struct shmid_kernel *shp;
 424         int err, id = 0;
 425         struct ipc_namespace *ns;
 426
 427         ns = current->nsproxy->ipc_ns;
 428
 429         mutex_lock(&shm_ids(ns).mutex);
 430         if (key == IPC_PRIVATE) {
 431                 err = newseg(ns, key, shmflg, size);
 432         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 433                 if (!(shmflg & IPC_CREAT))
 434                         err = -ENOENT;
 435                 else
 436                         err = newseg(ns, key, shmflg, size);
 437         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 438                 err = -EEXIST;
 439         } else {
 440                 shp = shm_lock(ns, id);
 441                 BUG_ON(shp==NULL);
 442                 if (shp->shm_segsz < size)
 443                         err = -EINVAL;
 444                 else if (ipcperms(&shp->shm_perm, shmflg))
 445                         err = -EACCES;
 446                 else {
 447                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 448                         err = security_shm_associate(shp, shmflg);
 449                         if (!err)
 450                                 err = shmid;
 451                 }
 452                 shm_unlock(shp);
 453         }
 454         mutex_unlock(&shm_ids(ns).mutex);
 455
 456         return err;
 457 }
 458
 459 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 460 {
 461         switch(version) {
 462         case IPC_64:
 463                 return copy_to_user(buf, in, sizeof(*in));
 464         case IPC_OLD:
 465             {
 466                 struct shmid_ds out;
 467
 468                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 469                 out.shm_segsz   = in->shm_segsz;
 470                 out.shm_atime   = in->shm_atime;
 471                 out.shm_dtime   = in->shm_dtime;
 472                 out.shm_ctime   = in->shm_ctime;
 473                 out.shm_cpid    = in->shm_cpid;
 474                 out.shm_lpid    = in->shm_lpid;
 475                 out.shm_nattch  = in->shm_nattch;
 476
 477                 return copy_to_user(buf, &out, sizeof(out));
 478             }
 479         default:
 480                 return -EINVAL;
 481         }
 482 }
 483
 484 struct shm_setbuf {
 485         uid_t   uid;
 486         gid_t   gid;
 487         mode_t  mode;
 488 };
 489
 490 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 491 {
 492         switch(version) {
 493         case IPC_64:
 494             {
 495                 struct shmid64_ds tbuf;
 496
 497                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 498                         return -EFAULT;
 499
 500                 out->uid        = tbuf.shm_perm.uid;
 501                 out->gid        = tbuf.shm_perm.gid;
 502                 out->mode       = tbuf.shm_perm.mode;
 503
 504                 return 0;
 505             }
 506         case IPC_OLD:
 507             {
 508                 struct shmid_ds tbuf_old;
 509
 510                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 511                         return -EFAULT;
 512
 513                 out->uid        = tbuf_old.shm_perm.uid;
 514                 out->gid        = tbuf_old.shm_perm.gid;
 515                 out->mode       = tbuf_old.shm_perm.mode;
 516
 517                 return 0;
 518             }
 519         default:
 520                 return -EINVAL;
 521         }
 522 }
 523
 524 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 525 {
 526         switch(version) {
 527         case IPC_64:
 528                 return copy_to_user(buf, in, sizeof(*in));
 529         case IPC_OLD:
 530             {
 531                 struct shminfo out;
 532
 533                 if(in->shmmax > INT_MAX)
 534                         out.shmmax = INT_MAX;
 535                 else
 536                         out.shmmax = (int)in->shmmax;
 537
 538                 out.shmmin      = in->shmmin;
 539                 out.shmmni      = in->shmmni;
 540                 out.shmseg      = in->shmseg;
 541                 out.shmall      = in->shmall;
 542
 543                 return copy_to_user(buf, &out, sizeof(out));
 544             }
 545         default:
 546                 return -EINVAL;
 547         }
 548 }
 549
 550 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 551                 unsigned long *swp)
 552 {
 553         int i;
 554
 555         *rss = 0;
 556         *swp = 0;
 557
 558         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 559                 struct shmid_kernel *shp;
 560                 struct inode *inode;
 561
 562                 shp = shm_get(ns, i);
 563                 if(!shp)
 564                         continue;
 565
 566                 inode = shp->shm_file->f_path.dentry->d_inode;
 567
 568                 if (is_file_hugepages(shp->shm_file)) {
 569                         struct address_space *mapping = inode->i_mapping;
 570                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 571                 } else {
 572                         struct shmem_inode_info *info = SHMEM_I(inode);
 573                         spin_lock(&info->lock);
 574                         *rss += inode->i_mapping->nrpages;
 575                         *swp += info->swapped;
 576                         spin_unlock(&info->lock);
 577                 }
 578         }
 579 }
 580
 581 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 582 {
 583         struct shm_setbuf setbuf;
 584         struct shmid_kernel *shp;
 585         int err, version;
 586         struct ipc_namespace *ns;
 587
 588         if (cmd < 0 || shmid < 0) {
 589                 err = -EINVAL;
 590                 goto out;
 591         }
 592
 593         version = ipc_parse_version(&cmd);
 594         ns = current->nsproxy->ipc_ns;
 595
 596         switch (cmd) { /* replace with proc interface ? */
 597         case IPC_INFO:
 598         {
 599                 struct shminfo64 shminfo;
 600
 601                 err = security_shm_shmctl(NULL, cmd);
 602                 if (err)
 603                         return err;
 604
 605                 memset(&shminfo,0,sizeof(shminfo));
 606                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 607                 shminfo.shmmax = ns->shm_ctlmax;
 608                 shminfo.shmall = ns->shm_ctlall;
 609
 610                 shminfo.shmmin = SHMMIN;
 611                 if(copy_shminfo_to_user (buf, &shminfo, version))
 612                         return -EFAULT;
 613                 /* reading a integer is always atomic */
 614                 err= shm_ids(ns).max_id;
 615                 if(err<0)
 616                         err = 0;
 617                 goto out;
 618         }
 619         case SHM_INFO:
 620         {
 621                 struct shm_info shm_info;
 622
 623                 err = security_shm_shmctl(NULL, cmd);
 624                 if (err)
 625                         return err;
 626
 627                 memset(&shm_info,0,sizeof(shm_info));
 628                 mutex_lock(&shm_ids(ns).mutex);
 629                 shm_info.used_ids = shm_ids(ns).in_use;
 630                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 631                 shm_info.shm_tot = ns->shm_tot;
 632                 shm_info.swap_attempts = 0;
 633                 shm_info.swap_successes = 0;
 634                 err = shm_ids(ns).max_id;
 635                 mutex_unlock(&shm_ids(ns).mutex);
 636                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 637                         err = -EFAULT;
 638                         goto out;
 639                 }
 640
 641                 err = err < 0 ? 0 : err;
 642                 goto out;
 643         }
 644         case SHM_STAT:
 645         case IPC_STAT:
 646         {
 647                 struct shmid64_ds tbuf;
 648                 int result;
 649                 memset(&tbuf, 0, sizeof(tbuf));
 650                 shp = shm_lock(ns, shmid);
 651                 if(shp==NULL) {
 652                         err = -EINVAL;
 653                         goto out;
 654                 } else if(cmd==SHM_STAT) {
 655                         err = -EINVAL;
 656                         if (shmid > shm_ids(ns).max_id)
 657                                 goto out_unlock;
 658                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 659                 } else {
 660                         err = shm_checkid(ns, shp,shmid);
 661                         if(err)
 662                                 goto out_unlock;
 663                         result = 0;
 664                 }
 665                 err=-EACCES;
 666                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 667                         goto out_unlock;
 668                 err = security_shm_shmctl(shp, cmd);
 669                 if (err)
 670                         goto out_unlock;
 671                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 672                 tbuf.shm_segsz  = shp->shm_segsz;
 673                 tbuf.shm_atime  = shp->shm_atim;
 674                 tbuf.shm_dtime  = shp->shm_dtim;
 675                 tbuf.shm_ctime  = shp->shm_ctim;
 676                 tbuf.shm_cpid   = shp->shm_cprid;
 677                 tbuf.shm_lpid   = shp->shm_lprid;
 678                 tbuf.shm_nattch = shp->shm_nattch;
 679                 shm_unlock(shp);
 680                 if(copy_shmid_to_user (buf, &tbuf, version))
 681                         err = -EFAULT;
 682                 else
 683                         err = result;
 684                 goto out;
 685         }
 686         case SHM_LOCK:
 687         case SHM_UNLOCK:
 688         {
 689                 shp = shm_lock(ns, shmid);
 690                 if(shp==NULL) {
 691                         err = -EINVAL;
 692                         goto out;
 693                 }
 694                 err = shm_checkid(ns, shp,shmid);
 695                 if(err)
 696                         goto out_unlock;
 697
 698                 err = audit_ipc_obj(&(shp->shm_perm));
 699                 if (err)
 700                         goto out_unlock;
 701
 702                 if (!capable(CAP_IPC_LOCK)) {
 703                         err = -EPERM;
 704                         if (current->euid != shp->shm_perm.uid &&
 705                             current->euid != shp->shm_perm.cuid)
 706                                 goto out_unlock;
 707                         if (cmd == SHM_LOCK &&
 708                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 709                                 goto out_unlock;
 710                 }
 711
 712                 err = security_shm_shmctl(shp, cmd);
 713                 if (err)
 714                         goto out_unlock;
 715
 716                 if(cmd==SHM_LOCK) {
 717                         struct user_struct * user = current->user;
 718                         if (!is_file_hugepages(shp->shm_file)) {
 719                                 err = shmem_lock(shp->shm_file, 1, user);
 720                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 721                                         shp->shm_perm.mode |= SHM_LOCKED;
 722                                         shp->mlock_user = user;
 723                                 }
 724                         }
 725                 } else if (!is_file_hugepages(shp->shm_file)) {
 726                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 727                         shp->shm_perm.mode &= ~SHM_LOCKED;
 728                         shp->mlock_user = NULL;
 729                 }
 730                 shm_unlock(shp);
 731                 goto out;
 732         }
 733         case IPC_RMID:
 734         {
 735                 /*
 736                  *      We cannot simply remove the file. The SVID states
 737                  *      that the block remains until the last person
 738                  *      detaches from it, then is deleted. A shmat() on
 739                  *      an RMID segment is legal in older Linux and if
 740                  *      we change it apps break...
 741                  *
 742                  *      Instead we set a destroyed flag, and then blow
 743                  *      the name away when the usage hits zero.
 744                  */
 745                 mutex_lock(&shm_ids(ns).mutex);
 746                 shp = shm_lock(ns, shmid);
 747                 err = -EINVAL;
 748                 if (shp == NULL)
 749                         goto out_up;
 750                 err = shm_checkid(ns, shp, shmid);
 751                 if(err)
 752                         goto out_unlock_up;
 753
 754                 err = audit_ipc_obj(&(shp->shm_perm));
 755                 if (err)
 756                         goto out_unlock_up;
 757
 758                 if (current->euid != shp->shm_perm.uid &&
 759                     current->euid != shp->shm_perm.cuid &&
 760                     !capable(CAP_SYS_ADMIN)) {
 761                         err=-EPERM;
 762                         goto out_unlock_up;
 763                 }
 764
 765                 err = security_shm_shmctl(shp, cmd);
 766                 if (err)
 767                         goto out_unlock_up;
 768
 769                 do_shm_rmid(ns, shp);
 770                 mutex_unlock(&shm_ids(ns).mutex);
 771                 goto out;
 772         }
 773
 774         case IPC_SET:
 775         {
 776                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 777                         err = -EFAULT;
 778                         goto out;
 779                 }
 780                 mutex_lock(&shm_ids(ns).mutex);
 781                 shp = shm_lock(ns, shmid);
 782                 err=-EINVAL;
 783                 if(shp==NULL)
 784                         goto out_up;
 785                 err = shm_checkid(ns, shp,shmid);
 786                 if(err)
 787                         goto out_unlock_up;
 788                 err = audit_ipc_obj(&(shp->shm_perm));
 789                 if (err)
 790                         goto out_unlock_up;
 791                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 792                 if (err)
 793                         goto out_unlock_up;
 794                 err=-EPERM;
 795                 if (current->euid != shp->shm_perm.uid &&
 796                     current->euid != shp->shm_perm.cuid &&
 797                     !capable(CAP_SYS_ADMIN)) {
 798                         goto out_unlock_up;
 799                 }
 800
 801                 err = security_shm_shmctl(shp, cmd);
 802                 if (err)
 803                         goto out_unlock_up;
 804
 805                 shp->shm_perm.uid = setbuf.uid;
 806                 shp->shm_perm.gid = setbuf.gid;
 807                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 808                         | (setbuf.mode & S_IRWXUGO);
 809                 shp->shm_ctim = get_seconds();
 810                 break;
 811         }
 812
 813         default:
 814                 err = -EINVAL;
 815                 goto out;
 816         }
 817
 818         err = 0;
 819 out_unlock_up:
 820         shm_unlock(shp);
 821 out_up:
 822         mutex_unlock(&shm_ids(ns).mutex);
 823         goto out;
 824 out_unlock:
 825         shm_unlock(shp);
 826 out:
 827         return err;
 828 }
 829
 830 /*
 831  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 832  *
 833  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 834  * "raddr" thing points to kernel space, and there has to be a wrapper around
 835  * this.
 836  */
 837 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 838 {
 839         struct shmid_kernel *shp;
 840         unsigned long addr;
 841         unsigned long size;
 842         struct file * file;
 843         int    err;
 844         unsigned long flags;
 845         unsigned long prot;
 846         int acc_mode;
 847         unsigned long user_addr;
 848         struct ipc_namespace *ns;
 849         struct shm_file_data *sfd;
 850         struct path path;
 851         mode_t f_mode;
 852
 853         err = -EINVAL;
 854         if (shmid < 0)
 855                 goto out;
 856         else if ((addr = (ulong)shmaddr)) {
 857                 if (addr & (SHMLBA-1)) {
 858                         if (shmflg & SHM_RND)
 859                                 addr &= ~(SHMLBA-1);       /* round down */
 860                         else
 861 #ifndef __ARCH_FORCE_SHMLBA
 862                                 if (addr & ~PAGE_MASK)
 863 #endif
 864                                         goto out;
 865                 }
 866                 flags = MAP_SHARED | MAP_FIXED;
 867         } else {
 868                 if ((shmflg & SHM_REMAP))
 869                         goto out;
 870
 871                 flags = MAP_SHARED;
 872         }
 873
 874         if (shmflg & SHM_RDONLY) {
 875                 prot = PROT_READ;
 876                 acc_mode = S_IRUGO;
 877                 f_mode = FMODE_READ;
 878         } else {
 879                 prot = PROT_READ | PROT_WRITE;
 880                 acc_mode = S_IRUGO | S_IWUGO;
 881                 f_mode = FMODE_READ | FMODE_WRITE;
 882         }
 883         if (shmflg & SHM_EXEC) {
 884                 prot |= PROT_EXEC;
 885                 acc_mode |= S_IXUGO;
 886         }
 887
 888         /*
 889          * We cannot rely on the fs check since SYSV IPC does have an
 890          * additional creator id...
 891          */
 892         ns = current->nsproxy->ipc_ns;
 893         shp = shm_lock(ns, shmid);
 894         if(shp == NULL)
 895                 goto out;
 896
 897         err = shm_checkid(ns, shp,shmid);
 898         if (err)
 899                 goto out_unlock;
 900
 901         err = -EACCES;
 902         if (ipcperms(&shp->shm_perm, acc_mode))
 903                 goto out_unlock;
 904
 905         err = security_shm_shmat(shp, shmaddr, shmflg);
 906         if (err)
 907                 goto out_unlock;
 908
 909         path.dentry = dget(shp->shm_file->f_path.dentry);
 910         path.mnt    = mntget(shp->shm_file->f_path.mnt);
 911         shp->shm_nattch++;
 912         size = i_size_read(path.dentry->d_inode);
 913         shm_unlock(shp);
 914
 915         err = -ENOMEM;
 916         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 917         if (!sfd)
 918                 goto out_put_path;
 919
 920         err = -ENOMEM;
 921         file = get_empty_filp();
 922         if (!file)
 923                 goto out_free;
 924
 925         file->f_op = &shm_file_operations;
 926         file->private_data = sfd;
 927         file->f_path = path;
 928         file->f_mapping = shp->shm_file->f_mapping;
 929         file->f_mode = f_mode;
 930         sfd->id = shp->id;
 931         sfd->ns = get_ipc_ns(ns);
 932         sfd->file = shp->shm_file;
 933         sfd->vm_ops = NULL;
 934
 935         down_write(&current->mm->mmap_sem);
 936         if (addr && !(shmflg & SHM_REMAP)) {
 937                 err = -EINVAL;
 938                 if (find_vma_intersection(current->mm, addr, addr + size))
 939                         goto invalid;
 940                 /*
 941                  * If shm segment goes below stack, make sure there is some
 942                  * space left for the stack to grow (at least 4 pages).
 943                  */
 944                 if (addr < current->mm->start_stack &&
 945                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 946                         goto invalid;
 947         }
 948
 949         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 950         *raddr = user_addr;
 951         err = 0;
 952         if (IS_ERR_VALUE(user_addr))
 953                 err = (long)user_addr;
 954 invalid:
 955         up_write(&current->mm->mmap_sem);
 956
 957         fput(file);
 958
 959 out_nattch:
 960         mutex_lock(&shm_ids(ns).mutex);
 961         shp = shm_lock(ns, shmid);
 962         BUG_ON(!shp);
 963         shp->shm_nattch--;
 964         if(shp->shm_nattch == 0 &&
 965            shp->shm_perm.mode & SHM_DEST)
 966                 shm_destroy(ns, shp);
 967         else
 968                 shm_unlock(shp);
 969         mutex_unlock(&shm_ids(ns).mutex);
 970
 971 out:
 972         return err;
 973
 974 out_unlock:
 975         shm_unlock(shp);
 976         goto out;
 977
 978 out_free:
 979         kfree(sfd);
 980 out_put_path:
 981         dput(path.dentry);
 982         mntput(path.mnt);
 983         goto out_nattch;
 984 }
 985
 986 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 987 {
 988         unsigned long ret;
 989         long err;
 990
 991         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 992         if (err)
 993                 return err;
 994         force_successful_syscall_return();
 995         return (long)ret;
 996 }
 997
 998 /*
 999  * detach and kill segment if marked destroyed.
1000  * The work is done in shm_close.
1001  */
1002 asmlinkage long sys_shmdt(char __user *shmaddr)
1003 {
1004         struct mm_struct *mm = current->mm;
1005         struct vm_area_struct *vma, *next;
1006         unsigned long addr = (unsigned long)shmaddr;
1007         loff_t size = 0;
1008         int retval = -EINVAL;
1009
1010         if (addr & ~PAGE_MASK)
1011                 return retval;
1012
1013         down_write(&mm->mmap_sem);
1014
1015         /*
1016          * This function tries to be smart and unmap shm segments that
1017          * were modified by partial mlock or munmap calls:
1018          * - It first determines the size of the shm segment that should be
1019          *   unmapped: It searches for a vma that is backed by shm and that
1020          *   started at address shmaddr. It records it's size and then unmaps
1021          *   it.
1022          * - Then it unmaps all shm vmas that started at shmaddr and that
1023          *   are within the initially determined size.
1024          * Errors from do_munmap are ignored: the function only fails if
1025          * it's called with invalid parameters or if it's called to unmap
1026          * a part of a vma. Both calls in this function are for full vmas,
1027          * the parameters are directly copied from the vma itself and always
1028          * valid - therefore do_munmap cannot fail. (famous last words?)
1029          */
1030         /*
1031          * If it had been mremap()'d, the starting address would not
1032          * match the usual checks anyway. So assume all vma's are
1033          * above the starting address given.
1034          */
1035         vma = find_vma(mm, addr);
1036
1037         while (vma) {
1038                 next = vma->vm_next;
1039
1040                 /*
1041                  * Check if the starting address would match, i.e. it's
1042                  * a fragment created by mprotect() and/or munmap(), or it
1043                  * otherwise it starts at this address with no hassles.
1044                  */
1045                 if ((vma->vm_ops == &shm_vm_ops) &&
1046                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1047
1048
1049                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1050                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051                         /*
1052                          * We discovered the size of the shm segment, so
1053                          * break out of here and fall through to the next
1054                          * loop that uses the size information to stop
1055                          * searching for matching vma's.
1056                          */
1057                         retval = 0;
1058                         vma = next;
1059                         break;
1060                 }
1061                 vma = next;
1062         }
1063
1064         /*
1065          * We need look no further than the maximum address a fragment
1066          * could possibly have landed at. Also cast things to loff_t to
1067          * prevent overflows and make comparisions vs. equal-width types.
1068          */
1069         size = PAGE_ALIGN(size);
1070         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1071                 next = vma->vm_next;
1072
1073                 /* finding a matching vma now does not alter retval */
1074                 if ((vma->vm_ops == &shm_vm_ops) &&
1075                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1076
1077                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1078                 vma = next;
1079         }
1080
1081         up_write(&mm->mmap_sem);
1082         return retval;
1083 }
1084
1085 #ifdef CONFIG_PROC_FS
1086 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1087 {
1088         struct shmid_kernel *shp = it;
1089         char *format;
1090
1091 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1092 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1093
1094         if (sizeof(size_t) <= sizeof(int))
1095                 format = SMALL_STRING;
1096         else
1097                 format = BIG_STRING;
1098         return seq_printf(s, format,
1099                           shp->shm_perm.key,
1100                           shp->id,
1101                           shp->shm_perm.mode,
1102                           shp->shm_segsz,
1103                           shp->shm_cprid,
1104                           shp->shm_lprid,
1105                           shp->shm_nattch,
1106                           shp->shm_perm.uid,
1107                           shp->shm_perm.gid,
1108                           shp->shm_perm.cuid,
1109                           shp->shm_perm.cgid,
1110                           shp->shm_atim,
1111                           shp->shm_dtim,
1112                           shp->shm_ctim);
1113 }
1114 #endif