SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 struct page *shm_fault(struct vm_area_struct *vma,
 228                                         struct fault_data *fdata)
 229 {
 230         struct file *file = vma->vm_file;
 231         struct shm_file_data *sfd = shm_file_data(file);
 232
 233         return sfd->vm_ops->fault(vma, fdata);
 234 }
 235
 236 #ifdef CONFIG_NUMA
 237 int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 238 {
 239         struct file *file = vma->vm_file;
 240         struct shm_file_data *sfd = shm_file_data(file);
 241         int err = 0;
 242         if (sfd->vm_ops->set_policy)
 243                 err = sfd->vm_ops->set_policy(vma, new);
 244         return err;
 245 }
 246
 247 struct mempolicy *shm_get_policy(struct vm_area_struct *vma, 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         BUG_ON(!sfd->vm_ops->fault);
 273         vma->vm_ops = &shm_vm_ops;
 274         shm_open(vma);
 275
 276         return ret;
 277 }
 278
 279 static int shm_release(struct inode *ino, struct file *file)
 280 {
 281         struct shm_file_data *sfd = shm_file_data(file);
 282
 283         put_ipc_ns(sfd->ns);
 284         shm_file_data(file) = NULL;
 285         kfree(sfd);
 286         return 0;
 287 }
 288
 289 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 290 {
 291         int (*fsync) (struct file *, struct dentry *, int datasync);
 292         struct shm_file_data *sfd = shm_file_data(file);
 293         int ret = -EINVAL;
 294
 295         fsync = sfd->file->f_op->fsync;
 296         if (fsync)
 297                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 298         return ret;
 299 }
 300
 301 static unsigned long shm_get_unmapped_area(struct file *file,
 302         unsigned long addr, unsigned long len, unsigned long pgoff,
 303         unsigned long flags)
 304 {
 305         struct shm_file_data *sfd = shm_file_data(file);
 306         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 307 }
 308
 309 int is_file_shm_hugepages(struct file *file)
 310 {
 311         int ret = 0;
 312
 313         if (file->f_op == &shm_file_operations) {
 314                 struct shm_file_data *sfd;
 315                 sfd = shm_file_data(file);
 316                 ret = is_file_hugepages(sfd->file);
 317         }
 318         return ret;
 319 }
 320
 321 static const struct file_operations shm_file_operations = {
 322         .mmap           = shm_mmap,
 323         .fsync          = shm_fsync,
 324         .release        = shm_release,
 325         .get_unmapped_area      = shm_get_unmapped_area,
 326 };
 327
 328 static struct vm_operations_struct shm_vm_ops = {
 329         .open   = shm_open,     /* callback for a new vm-area open */
 330         .close  = shm_close,    /* callback for when the vm-area is released */
 331         .fault  = shm_fault,
 332 #if defined(CONFIG_NUMA)
 333         .set_policy = shm_set_policy,
 334         .get_policy = shm_get_policy,
 335 #endif
 336 };
 337
 338 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 339 {
 340         int error;
 341         struct shmid_kernel *shp;
 342         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 343         struct file * file;
 344         char name[13];
 345         int id;
 346
 347         if (size < SHMMIN || size > ns->shm_ctlmax)
 348                 return -EINVAL;
 349
 350         if (ns->shm_tot + numpages > ns->shm_ctlall)
 351                 return -ENOSPC;
 352
 353         shp = ipc_rcu_alloc(sizeof(*shp));
 354         if (!shp)
 355                 return -ENOMEM;
 356
 357         shp->shm_perm.key = key;
 358         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 359         shp->mlock_user = NULL;
 360
 361         shp->shm_perm.security = NULL;
 362         error = security_shm_alloc(shp);
 363         if (error) {
 364                 ipc_rcu_putref(shp);
 365                 return error;
 366         }
 367
 368         sprintf (name, "SYSV%08x", key);
 369         if (shmflg & SHM_HUGETLB) {
 370                 /* hugetlb_file_setup takes care of mlock user accounting */
 371                 file = hugetlb_file_setup(name, 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                 file = shmem_file_setup(name, size, acctflag);
 383         }
 384         error = PTR_ERR(file);
 385         if (IS_ERR(file))
 386                 goto no_file;
 387
 388         error = -ENOSPC;
 389         id = shm_addid(ns, shp);
 390         if(id == -1)
 391                 goto no_id;
 392
 393         shp->shm_cprid = current->tgid;
 394         shp->shm_lprid = 0;
 395         shp->shm_atim = shp->shm_dtim = 0;
 396         shp->shm_ctim = get_seconds();
 397         shp->shm_segsz = size;
 398         shp->shm_nattch = 0;
 399         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
 400         shp->shm_file = file;
 401         /*
 402          * shmid gets reported as "inode#" in /proc/pid/maps.
 403          * proc-ps tools use this. Changing this will break them.
 404          */
 405         file->f_dentry->d_inode->i_ino = shp->id;
 406
 407         ns->shm_tot += numpages;
 408         shm_unlock(shp);
 409         return shp->id;
 410
 411 no_id:
 412         fput(file);
 413 no_file:
 414         security_shm_free(shp);
 415         ipc_rcu_putref(shp);
 416         return error;
 417 }
 418
 419 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 420 {
 421         struct shmid_kernel *shp;
 422         int err, id = 0;
 423         struct ipc_namespace *ns;
 424
 425         ns = current->nsproxy->ipc_ns;
 426
 427         mutex_lock(&shm_ids(ns).mutex);
 428         if (key == IPC_PRIVATE) {
 429                 err = newseg(ns, key, shmflg, size);
 430         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
 431                 if (!(shmflg & IPC_CREAT))
 432                         err = -ENOENT;
 433                 else
 434                         err = newseg(ns, key, shmflg, size);
 435         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 436                 err = -EEXIST;
 437         } else {
 438                 shp = shm_lock(ns, id);
 439                 BUG_ON(shp==NULL);
 440                 if (shp->shm_segsz < size)
 441                         err = -EINVAL;
 442                 else if (ipcperms(&shp->shm_perm, shmflg))
 443                         err = -EACCES;
 444                 else {
 445                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
 446                         err = security_shm_associate(shp, shmflg);
 447                         if (!err)
 448                                 err = shmid;
 449                 }
 450                 shm_unlock(shp);
 451         }
 452         mutex_unlock(&shm_ids(ns).mutex);
 453
 454         return err;
 455 }
 456
 457 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 458 {
 459         switch(version) {
 460         case IPC_64:
 461                 return copy_to_user(buf, in, sizeof(*in));
 462         case IPC_OLD:
 463             {
 464                 struct shmid_ds out;
 465
 466                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 467                 out.shm_segsz   = in->shm_segsz;
 468                 out.shm_atime   = in->shm_atime;
 469                 out.shm_dtime   = in->shm_dtime;
 470                 out.shm_ctime   = in->shm_ctime;
 471                 out.shm_cpid    = in->shm_cpid;
 472                 out.shm_lpid    = in->shm_lpid;
 473                 out.shm_nattch  = in->shm_nattch;
 474
 475                 return copy_to_user(buf, &out, sizeof(out));
 476             }
 477         default:
 478                 return -EINVAL;
 479         }
 480 }
 481
 482 struct shm_setbuf {
 483         uid_t   uid;
 484         gid_t   gid;
 485         mode_t  mode;
 486 };
 487
 488 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 489 {
 490         switch(version) {
 491         case IPC_64:
 492             {
 493                 struct shmid64_ds tbuf;
 494
 495                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 496                         return -EFAULT;
 497
 498                 out->uid        = tbuf.shm_perm.uid;
 499                 out->gid        = tbuf.shm_perm.gid;
 500                 out->mode       = tbuf.shm_perm.mode;
 501
 502                 return 0;
 503             }
 504         case IPC_OLD:
 505             {
 506                 struct shmid_ds tbuf_old;
 507
 508                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 509                         return -EFAULT;
 510
 511                 out->uid        = tbuf_old.shm_perm.uid;
 512                 out->gid        = tbuf_old.shm_perm.gid;
 513                 out->mode       = tbuf_old.shm_perm.mode;
 514
 515                 return 0;
 516             }
 517         default:
 518                 return -EINVAL;
 519         }
 520 }
 521
 522 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 523 {
 524         switch(version) {
 525         case IPC_64:
 526                 return copy_to_user(buf, in, sizeof(*in));
 527         case IPC_OLD:
 528             {
 529                 struct shminfo out;
 530
 531                 if(in->shmmax > INT_MAX)
 532                         out.shmmax = INT_MAX;
 533                 else
 534                         out.shmmax = (int)in->shmmax;
 535
 536                 out.shmmin      = in->shmmin;
 537                 out.shmmni      = in->shmmni;
 538                 out.shmseg      = in->shmseg;
 539                 out.shmall      = in->shmall;
 540
 541                 return copy_to_user(buf, &out, sizeof(out));
 542             }
 543         default:
 544                 return -EINVAL;
 545         }
 546 }
 547
 548 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 549                 unsigned long *swp)
 550 {
 551         int i;
 552
 553         *rss = 0;
 554         *swp = 0;
 555
 556         for (i = 0; i <= shm_ids(ns).max_id; i++) {
 557                 struct shmid_kernel *shp;
 558                 struct inode *inode;
 559
 560                 shp = shm_get(ns, i);
 561                 if(!shp)
 562                         continue;
 563
 564                 inode = shp->shm_file->f_path.dentry->d_inode;
 565
 566                 if (is_file_hugepages(shp->shm_file)) {
 567                         struct address_space *mapping = inode->i_mapping;
 568                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 569                 } else {
 570                         struct shmem_inode_info *info = SHMEM_I(inode);
 571                         spin_lock(&info->lock);
 572                         *rss += inode->i_mapping->nrpages;
 573                         *swp += info->swapped;
 574                         spin_unlock(&info->lock);
 575                 }
 576         }
 577 }
 578
 579 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 580 {
 581         struct shm_setbuf setbuf;
 582         struct shmid_kernel *shp;
 583         int err, version;
 584         struct ipc_namespace *ns;
 585
 586         if (cmd < 0 || shmid < 0) {
 587                 err = -EINVAL;
 588                 goto out;
 589         }
 590
 591         version = ipc_parse_version(&cmd);
 592         ns = current->nsproxy->ipc_ns;
 593
 594         switch (cmd) { /* replace with proc interface ? */
 595         case IPC_INFO:
 596         {
 597                 struct shminfo64 shminfo;
 598
 599                 err = security_shm_shmctl(NULL, cmd);
 600                 if (err)
 601                         return err;
 602
 603                 memset(&shminfo,0,sizeof(shminfo));
 604                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 605                 shminfo.shmmax = ns->shm_ctlmax;
 606                 shminfo.shmall = ns->shm_ctlall;
 607
 608                 shminfo.shmmin = SHMMIN;
 609                 if(copy_shminfo_to_user (buf, &shminfo, version))
 610                         return -EFAULT;
 611                 /* reading a integer is always atomic */
 612                 err= shm_ids(ns).max_id;
 613                 if(err<0)
 614                         err = 0;
 615                 goto out;
 616         }
 617         case SHM_INFO:
 618         {
 619                 struct shm_info shm_info;
 620
 621                 err = security_shm_shmctl(NULL, cmd);
 622                 if (err)
 623                         return err;
 624
 625                 memset(&shm_info,0,sizeof(shm_info));
 626                 mutex_lock(&shm_ids(ns).mutex);
 627                 shm_info.used_ids = shm_ids(ns).in_use;
 628                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 629                 shm_info.shm_tot = ns->shm_tot;
 630                 shm_info.swap_attempts = 0;
 631                 shm_info.swap_successes = 0;
 632                 err = shm_ids(ns).max_id;
 633                 mutex_unlock(&shm_ids(ns).mutex);
 634                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 635                         err = -EFAULT;
 636                         goto out;
 637                 }
 638
 639                 err = err < 0 ? 0 : err;
 640                 goto out;
 641         }
 642         case SHM_STAT:
 643         case IPC_STAT:
 644         {
 645                 struct shmid64_ds tbuf;
 646                 int result;
 647                 memset(&tbuf, 0, sizeof(tbuf));
 648                 shp = shm_lock(ns, shmid);
 649                 if(shp==NULL) {
 650                         err = -EINVAL;
 651                         goto out;
 652                 } else if(cmd==SHM_STAT) {
 653                         err = -EINVAL;
 654                         if (shmid > shm_ids(ns).max_id)
 655                                 goto out_unlock;
 656                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
 657                 } else {
 658                         err = shm_checkid(ns, shp,shmid);
 659                         if(err)
 660                                 goto out_unlock;
 661                         result = 0;
 662                 }
 663                 err=-EACCES;
 664                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 665                         goto out_unlock;
 666                 err = security_shm_shmctl(shp, cmd);
 667                 if (err)
 668                         goto out_unlock;
 669                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 670                 tbuf.shm_segsz  = shp->shm_segsz;
 671                 tbuf.shm_atime  = shp->shm_atim;
 672                 tbuf.shm_dtime  = shp->shm_dtim;
 673                 tbuf.shm_ctime  = shp->shm_ctim;
 674                 tbuf.shm_cpid   = shp->shm_cprid;
 675                 tbuf.shm_lpid   = shp->shm_lprid;
 676                 tbuf.shm_nattch = shp->shm_nattch;
 677                 shm_unlock(shp);
 678                 if(copy_shmid_to_user (buf, &tbuf, version))
 679                         err = -EFAULT;
 680                 else
 681                         err = result;
 682                 goto out;
 683         }
 684         case SHM_LOCK:
 685         case SHM_UNLOCK:
 686         {
 687                 shp = shm_lock(ns, shmid);
 688                 if(shp==NULL) {
 689                         err = -EINVAL;
 690                         goto out;
 691                 }
 692                 err = shm_checkid(ns, shp,shmid);
 693                 if(err)
 694                         goto out_unlock;
 695
 696                 err = audit_ipc_obj(&(shp->shm_perm));
 697                 if (err)
 698                         goto out_unlock;
 699
 700                 if (!capable(CAP_IPC_LOCK)) {
 701                         err = -EPERM;
 702                         if (current->euid != shp->shm_perm.uid &&
 703                             current->euid != shp->shm_perm.cuid)
 704                                 goto out_unlock;
 705                         if (cmd == SHM_LOCK &&
 706                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 707                                 goto out_unlock;
 708                 }
 709
 710                 err = security_shm_shmctl(shp, cmd);
 711                 if (err)
 712                         goto out_unlock;
 713
 714                 if(cmd==SHM_LOCK) {
 715                         struct user_struct * user = current->user;
 716                         if (!is_file_hugepages(shp->shm_file)) {
 717                                 err = shmem_lock(shp->shm_file, 1, user);
 718                                 if (!err) {
 719                                         shp->shm_perm.mode |= SHM_LOCKED;
 720                                         shp->mlock_user = user;
 721                                 }
 722                         }
 723                 } else if (!is_file_hugepages(shp->shm_file)) {
 724                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 725                         shp->shm_perm.mode &= ~SHM_LOCKED;
 726                         shp->mlock_user = NULL;
 727                 }
 728                 shm_unlock(shp);
 729                 goto out;
 730         }
 731         case IPC_RMID:
 732         {
 733                 /*
 734                  *      We cannot simply remove the file. The SVID states
 735                  *      that the block remains until the last person
 736                  *      detaches from it, then is deleted. A shmat() on
 737                  *      an RMID segment is legal in older Linux and if
 738                  *      we change it apps break...
 739                  *
 740                  *      Instead we set a destroyed flag, and then blow
 741                  *      the name away when the usage hits zero.
 742                  */
 743                 mutex_lock(&shm_ids(ns).mutex);
 744                 shp = shm_lock(ns, shmid);
 745                 err = -EINVAL;
 746                 if (shp == NULL)
 747                         goto out_up;
 748                 err = shm_checkid(ns, shp, shmid);
 749                 if(err)
 750                         goto out_unlock_up;
 751
 752                 err = audit_ipc_obj(&(shp->shm_perm));
 753                 if (err)
 754                         goto out_unlock_up;
 755
 756                 if (current->euid != shp->shm_perm.uid &&
 757                     current->euid != shp->shm_perm.cuid &&
 758                     !capable(CAP_SYS_ADMIN)) {
 759                         err=-EPERM;
 760                         goto out_unlock_up;
 761                 }
 762
 763                 err = security_shm_shmctl(shp, cmd);
 764                 if (err)
 765                         goto out_unlock_up;
 766
 767                 do_shm_rmid(ns, shp);
 768                 mutex_unlock(&shm_ids(ns).mutex);
 769                 goto out;
 770         }
 771
 772         case IPC_SET:
 773         {
 774                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 775                         err = -EFAULT;
 776                         goto out;
 777                 }
 778                 mutex_lock(&shm_ids(ns).mutex);
 779                 shp = shm_lock(ns, shmid);
 780                 err=-EINVAL;
 781                 if(shp==NULL)
 782                         goto out_up;
 783                 err = shm_checkid(ns, shp,shmid);
 784                 if(err)
 785                         goto out_unlock_up;
 786                 err = audit_ipc_obj(&(shp->shm_perm));
 787                 if (err)
 788                         goto out_unlock_up;
 789                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 790                 if (err)
 791                         goto out_unlock_up;
 792                 err=-EPERM;
 793                 if (current->euid != shp->shm_perm.uid &&
 794                     current->euid != shp->shm_perm.cuid &&
 795                     !capable(CAP_SYS_ADMIN)) {
 796                         goto out_unlock_up;
 797                 }
 798
 799                 err = security_shm_shmctl(shp, cmd);
 800                 if (err)
 801                         goto out_unlock_up;
 802
 803                 shp->shm_perm.uid = setbuf.uid;
 804                 shp->shm_perm.gid = setbuf.gid;
 805                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 806                         | (setbuf.mode & S_IRWXUGO);
 807                 shp->shm_ctim = get_seconds();
 808                 break;
 809         }
 810
 811         default:
 812                 err = -EINVAL;
 813                 goto out;
 814         }
 815
 816         err = 0;
 817 out_unlock_up:
 818         shm_unlock(shp);
 819 out_up:
 820         mutex_unlock(&shm_ids(ns).mutex);
 821         goto out;
 822 out_unlock:
 823         shm_unlock(shp);
 824 out:
 825         return err;
 826 }
 827
 828 /*
 829  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 830  *
 831  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 832  * "raddr" thing points to kernel space, and there has to be a wrapper around
 833  * this.
 834  */
 835 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 836 {
 837         struct shmid_kernel *shp;
 838         unsigned long addr;
 839         unsigned long size;
 840         struct file * file;
 841         int    err;
 842         unsigned long flags;
 843         unsigned long prot;
 844         int acc_mode;
 845         unsigned long user_addr;
 846         struct ipc_namespace *ns;
 847         struct shm_file_data *sfd;
 848         struct path path;
 849         mode_t f_mode;
 850
 851         err = -EINVAL;
 852         if (shmid < 0)
 853                 goto out;
 854         else if ((addr = (ulong)shmaddr)) {
 855                 if (addr & (SHMLBA-1)) {
 856                         if (shmflg & SHM_RND)
 857                                 addr &= ~(SHMLBA-1);       /* round down */
 858                         else
 859 #ifndef __ARCH_FORCE_SHMLBA
 860                                 if (addr & ~PAGE_MASK)
 861 #endif
 862                                         goto out;
 863                 }
 864                 flags = MAP_SHARED | MAP_FIXED;
 865         } else {
 866                 if ((shmflg & SHM_REMAP))
 867                         goto out;
 868
 869                 flags = MAP_SHARED;
 870         }
 871
 872         if (shmflg & SHM_RDONLY) {
 873                 prot = PROT_READ;
 874                 acc_mode = S_IRUGO;
 875                 f_mode = FMODE_READ;
 876         } else {
 877                 prot = PROT_READ | PROT_WRITE;
 878                 acc_mode = S_IRUGO | S_IWUGO;
 879                 f_mode = FMODE_READ | FMODE_WRITE;
 880         }
 881         if (shmflg & SHM_EXEC) {
 882                 prot |= PROT_EXEC;
 883                 acc_mode |= S_IXUGO;
 884         }
 885
 886         /*
 887          * We cannot rely on the fs check since SYSV IPC does have an
 888          * additional creator id...
 889          */
 890         ns = current->nsproxy->ipc_ns;
 891         shp = shm_lock(ns, shmid);
 892         if(shp == NULL)
 893                 goto out;
 894
 895         err = shm_checkid(ns, shp,shmid);
 896         if (err)
 897                 goto out_unlock;
 898
 899         err = -EACCES;
 900         if (ipcperms(&shp->shm_perm, acc_mode))
 901                 goto out_unlock;
 902
 903         err = security_shm_shmat(shp, shmaddr, shmflg);
 904         if (err)
 905                 goto out_unlock;
 906
 907         path.dentry = dget(shp->shm_file->f_path.dentry);
 908         path.mnt    = mntget(shp->shm_file->f_path.mnt);
 909         shp->shm_nattch++;
 910         size = i_size_read(path.dentry->d_inode);
 911         shm_unlock(shp);
 912
 913         err = -ENOMEM;
 914         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 915         if (!sfd)
 916                 goto out_put_path;
 917
 918         err = -ENOMEM;
 919         file = get_empty_filp();
 920         if (!file)
 921                 goto out_free;
 922
 923         file->f_op = &shm_file_operations;
 924         file->private_data = sfd;
 925         file->f_path = path;
 926         file->f_mapping = shp->shm_file->f_mapping;
 927         file->f_mode = f_mode;
 928         sfd->id = shp->id;
 929         sfd->ns = get_ipc_ns(ns);
 930         sfd->file = shp->shm_file;
 931         sfd->vm_ops = NULL;
 932
 933         down_write(&current->mm->mmap_sem);
 934         if (addr && !(shmflg & SHM_REMAP)) {
 935                 err = -EINVAL;
 936                 if (find_vma_intersection(current->mm, addr, addr + size))
 937                         goto invalid;
 938                 /*
 939                  * If shm segment goes below stack, make sure there is some
 940                  * space left for the stack to grow (at least 4 pages).
 941                  */
 942                 if (addr < current->mm->start_stack &&
 943                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 944                         goto invalid;
 945         }
 946
 947         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 948         *raddr = user_addr;
 949         err = 0;
 950         if (IS_ERR_VALUE(user_addr))
 951                 err = (long)user_addr;
 952 invalid:
 953         up_write(&current->mm->mmap_sem);
 954
 955         fput(file);
 956
 957 out_nattch:
 958         mutex_lock(&shm_ids(ns).mutex);
 959         shp = shm_lock(ns, shmid);
 960         BUG_ON(!shp);
 961         shp->shm_nattch--;
 962         if(shp->shm_nattch == 0 &&
 963            shp->shm_perm.mode & SHM_DEST)
 964                 shm_destroy(ns, shp);
 965         else
 966                 shm_unlock(shp);
 967         mutex_unlock(&shm_ids(ns).mutex);
 968
 969 out:
 970         return err;
 971
 972 out_unlock:
 973         shm_unlock(shp);
 974         goto out;
 975
 976 out_free:
 977         kfree(sfd);
 978 out_put_path:
 979         dput(path.dentry);
 980         mntput(path.mnt);
 981         goto out_nattch;
 982 }
 983
 984 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 985 {
 986         unsigned long ret;
 987         long err;
 988
 989         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 990         if (err)
 991                 return err;
 992         force_successful_syscall_return();
 993         return (long)ret;
 994 }
 995
 996 /*
 997  * detach and kill segment if marked destroyed.
 998  * The work is done in shm_close.
 999  */
1000 asmlinkage long sys_shmdt(char __user *shmaddr)
1001 {
1002         struct mm_struct *mm = current->mm;
1003         struct vm_area_struct *vma, *next;
1004         unsigned long addr = (unsigned long)shmaddr;
1005         loff_t size = 0;
1006         int retval = -EINVAL;
1007
1008         if (addr & ~PAGE_MASK)
1009                 return retval;
1010
1011         down_write(&mm->mmap_sem);
1012
1013         /*
1014          * This function tries to be smart and unmap shm segments that
1015          * were modified by partial mlock or munmap calls:
1016          * - It first determines the size of the shm segment that should be
1017          *   unmapped: It searches for a vma that is backed by shm and that
1018          *   started at address shmaddr. It records it's size and then unmaps
1019          *   it.
1020          * - Then it unmaps all shm vmas that started at shmaddr and that
1021          *   are within the initially determined size.
1022          * Errors from do_munmap are ignored: the function only fails if
1023          * it's called with invalid parameters or if it's called to unmap
1024          * a part of a vma. Both calls in this function are for full vmas,
1025          * the parameters are directly copied from the vma itself and always
1026          * valid - therefore do_munmap cannot fail. (famous last words?)
1027          */
1028         /*
1029          * If it had been mremap()'d, the starting address would not
1030          * match the usual checks anyway. So assume all vma's are
1031          * above the starting address given.
1032          */
1033         vma = find_vma(mm, addr);
1034
1035         while (vma) {
1036                 next = vma->vm_next;
1037
1038                 /*
1039                  * Check if the starting address would match, i.e. it's
1040                  * a fragment created by mprotect() and/or munmap(), or it
1041                  * otherwise it starts at this address with no hassles.
1042                  */
1043                 if ((vma->vm_ops == &shm_vm_ops) &&
1044                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1045
1046
1047                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1048                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1049                         /*
1050                          * We discovered the size of the shm segment, so
1051                          * break out of here and fall through to the next
1052                          * loop that uses the size information to stop
1053                          * searching for matching vma's.
1054                          */
1055                         retval = 0;
1056                         vma = next;
1057                         break;
1058                 }
1059                 vma = next;
1060         }
1061
1062         /*
1063          * We need look no further than the maximum address a fragment
1064          * could possibly have landed at. Also cast things to loff_t to
1065          * prevent overflows and make comparisions vs. equal-width types.
1066          */
1067         size = PAGE_ALIGN(size);
1068         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1069                 next = vma->vm_next;
1070
1071                 /* finding a matching vma now does not alter retval */
1072                 if ((vma->vm_ops == &shm_vm_ops) &&
1073                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1074
1075                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1076                 vma = next;
1077         }
1078
1079         up_write(&mm->mmap_sem);
1080         return retval;
1081 }
1082
1083 #ifdef CONFIG_PROC_FS
1084 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1085 {
1086         struct shmid_kernel *shp = it;
1087         char *format;
1088
1089 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1090 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1091
1092         if (sizeof(size_t) <= sizeof(int))
1093                 format = SMALL_STRING;
1094         else
1095                 format = BIG_STRING;
1096         return seq_printf(s, format,
1097                           shp->shm_perm.key,
1098                           shp->id,
1099                           shp->shm_perm.mode,
1100                           shp->shm_segsz,
1101                           shp->shm_cprid,
1102                           shp->shm_lprid,
1103                           shp->shm_nattch,
1104                           shp->shm_perm.uid,
1105                           shp->shm_perm.gid,
1106                           shp->shm_perm.cuid,
1107                           shp->shm_perm.cgid,
1108                           shp->shm_atim,
1109                           shp->shm_dtim,
1110                           shp->shm_ctim);
1111 }
1112 #endif