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_buildid(ns, id, seq)        \
  67         ipc_buildid(&shm_ids(ns), id, seq)
  68
  69 static int newseg(struct ipc_namespace *, struct ipc_params *);
  70 static void shm_open(struct vm_area_struct *vma);
  71 static void shm_close(struct vm_area_struct *vma);
  72 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  73 #ifdef CONFIG_PROC_FS
  74 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  75 #endif
  76
  77 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  78 {
  79         ns->ids[IPC_SHM_IDS] = ids;
  80         ns->shm_ctlmax = SHMMAX;
  81         ns->shm_ctlall = SHMALL;
  82         ns->shm_ctlmni = SHMMNI;
  83         ns->shm_tot = 0;
  84         ipc_init_ids(ids);
  85 }
  86
  87 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  88 {
  89         if (shp->shm_nattch){
  90                 shp->shm_perm.mode |= SHM_DEST;
  91                 /* Do not find it any more */
  92                 shp->shm_perm.key = IPC_PRIVATE;
  93                 shm_unlock(shp);
  94         } else
  95                 shm_destroy(ns, shp);
  96 }
  97
  98 int shm_init_ns(struct ipc_namespace *ns)
  99 {
 100         struct ipc_ids *ids;
 101
 102         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 103         if (ids == NULL)
 104                 return -ENOMEM;
 105
 106         __shm_init_ns(ns, ids);
 107         return 0;
 108 }
 109
 110 void shm_exit_ns(struct ipc_namespace *ns)
 111 {
 112         struct shmid_kernel *shp;
 113         int next_id;
 114         int total, in_use;
 115
 116         mutex_lock(&shm_ids(ns).mutex);
 117
 118         in_use = shm_ids(ns).in_use;
 119
 120         for (total = 0, next_id = 0; total < in_use; next_id++) {
 121                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 122                 if (shp == NULL)
 123                         continue;
 124                 ipc_lock_by_ptr(&shp->shm_perm);
 125                 do_shm_rmid(ns, shp);
 126                 total++;
 127         }
 128         mutex_unlock(&shm_ids(ns).mutex);
 129
 130         kfree(ns->ids[IPC_SHM_IDS]);
 131         ns->ids[IPC_SHM_IDS] = NULL;
 132 }
 133
 134 void __init shm_init (void)
 135 {
 136         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 137         ipc_init_proc_interface("sysvipc/shm",
 138                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 139                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 140 }
 141
 142 static inline int shm_checkid(struct ipc_namespace *ns,
 143                 struct shmid_kernel *s, int id)
 144 {
 145         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 146                 return -EIDRM;
 147         return 0;
 148 }
 149
 150 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 151 {
 152         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 153 }
 154
 155 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 156 {
 157         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 158 }
 159
 160
 161
 162 /* This is called by fork, once for every shm attach. */
 163 static void shm_open(struct vm_area_struct *vma)
 164 {
 165         struct file *file = vma->vm_file;
 166         struct shm_file_data *sfd = shm_file_data(file);
 167         struct shmid_kernel *shp;
 168
 169         shp = shm_lock(sfd->ns, sfd->id);
 170         BUG_ON(!shp);
 171         shp->shm_atim = get_seconds();
 172         shp->shm_lprid = task_tgid_vnr(current);
 173         shp->shm_nattch++;
 174         shm_unlock(shp);
 175 }
 176
 177 /*
 178  * shm_destroy - free the struct shmid_kernel
 179  *
 180  * @shp: struct to free
 181  *
 182  * It has to be called with shp and shm_ids.mutex locked,
 183  * but returns with shp unlocked and freed.
 184  */
 185 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 186 {
 187         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 188         shm_rmid(ns, shp);
 189         shm_unlock(shp);
 190         if (!is_file_hugepages(shp->shm_file))
 191                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 192         else
 193                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 194                                                 shp->mlock_user);
 195         fput (shp->shm_file);
 196         security_shm_free(shp);
 197         ipc_rcu_putref(shp);
 198 }
 199
 200 /*
 201  * remove the attach descriptor vma.
 202  * free memory for segment if it is marked destroyed.
 203  * The descriptor has already been removed from the current->mm->mmap list
 204  * and will later be kfree()d.
 205  */
 206 static void shm_close(struct vm_area_struct *vma)
 207 {
 208         struct file * file = vma->vm_file;
 209         struct shm_file_data *sfd = shm_file_data(file);
 210         struct shmid_kernel *shp;
 211         struct ipc_namespace *ns = sfd->ns;
 212
 213         mutex_lock(&shm_ids(ns).mutex);
 214         /* remove from the list of attaches of the shm segment */
 215         shp = shm_lock(ns, sfd->id);
 216         BUG_ON(!shp);
 217         shp->shm_lprid = task_tgid_vnr(current);
 218         shp->shm_dtim = get_seconds();
 219         shp->shm_nattch--;
 220         if(shp->shm_nattch == 0 &&
 221            shp->shm_perm.mode & SHM_DEST)
 222                 shm_destroy(ns, shp);
 223         else
 224                 shm_unlock(shp);
 225         mutex_unlock(&shm_ids(ns).mutex);
 226 }
 227
 228 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 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, vmf);
 234 }
 235
 236 #ifdef CONFIG_NUMA
 237 static 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 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 248                                         unsigned long addr)
 249 {
 250         struct file *file = vma->vm_file;
 251         struct shm_file_data *sfd = shm_file_data(file);
 252         struct mempolicy *pol = NULL;
 253
 254         if (sfd->vm_ops->get_policy)
 255                 pol = sfd->vm_ops->get_policy(vma, addr);
 256         else if (vma->vm_policy)
 257                 pol = vma->vm_policy;
 258         else
 259                 pol = current->mempolicy;
 260         return pol;
 261 }
 262 #endif
 263
 264 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 265 {
 266         struct shm_file_data *sfd = shm_file_data(file);
 267         int ret;
 268
 269         ret = sfd->file->f_op->mmap(sfd->file, vma);
 270         if (ret != 0)
 271                 return ret;
 272         sfd->vm_ops = vma->vm_ops;
 273 #ifdef CONFIG_MMU
 274         BUG_ON(!sfd->vm_ops->fault);
 275 #endif
 276         vma->vm_ops = &shm_vm_ops;
 277         shm_open(vma);
 278
 279         return ret;
 280 }
 281
 282 static int shm_release(struct inode *ino, struct file *file)
 283 {
 284         struct shm_file_data *sfd = shm_file_data(file);
 285
 286         put_ipc_ns(sfd->ns);
 287         shm_file_data(file) = NULL;
 288         kfree(sfd);
 289         return 0;
 290 }
 291
 292 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 293 {
 294         int (*fsync) (struct file *, struct dentry *, int datasync);
 295         struct shm_file_data *sfd = shm_file_data(file);
 296         int ret = -EINVAL;
 297
 298         fsync = sfd->file->f_op->fsync;
 299         if (fsync)
 300                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 301         return ret;
 302 }
 303
 304 static unsigned long shm_get_unmapped_area(struct file *file,
 305         unsigned long addr, unsigned long len, unsigned long pgoff,
 306         unsigned long flags)
 307 {
 308         struct shm_file_data *sfd = shm_file_data(file);
 309         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 310 }
 311
 312 int is_file_shm_hugepages(struct file *file)
 313 {
 314         int ret = 0;
 315
 316         if (file->f_op == &shm_file_operations) {
 317                 struct shm_file_data *sfd;
 318                 sfd = shm_file_data(file);
 319                 ret = is_file_hugepages(sfd->file);
 320         }
 321         return ret;
 322 }
 323
 324 static const struct file_operations shm_file_operations = {
 325         .mmap           = shm_mmap,
 326         .fsync          = shm_fsync,
 327         .release        = shm_release,
 328         .get_unmapped_area      = shm_get_unmapped_area,
 329 };
 330
 331 static struct vm_operations_struct shm_vm_ops = {
 332         .open   = shm_open,     /* callback for a new vm-area open */
 333         .close  = shm_close,    /* callback for when the vm-area is released */
 334         .fault  = shm_fault,
 335 #if defined(CONFIG_NUMA)
 336         .set_policy = shm_set_policy,
 337         .get_policy = shm_get_policy,
 338 #endif
 339 };
 340
 341 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 342 {
 343         key_t key = params->key;
 344         int shmflg = params->flg;
 345         size_t size = params->u.size;
 346         int error;
 347         struct shmid_kernel *shp;
 348         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 349         struct file * file;
 350         char name[13];
 351         int id;
 352
 353         if (size < SHMMIN || size > ns->shm_ctlmax)
 354                 return -EINVAL;
 355
 356         if (ns->shm_tot + numpages > ns->shm_ctlall)
 357                 return -ENOSPC;
 358
 359         shp = ipc_rcu_alloc(sizeof(*shp));
 360         if (!shp)
 361                 return -ENOMEM;
 362
 363         shp->shm_perm.key = key;
 364         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 365         shp->mlock_user = NULL;
 366
 367         shp->shm_perm.security = NULL;
 368         error = security_shm_alloc(shp);
 369         if (error) {
 370                 ipc_rcu_putref(shp);
 371                 return error;
 372         }
 373
 374         sprintf (name, "SYSV%08x", key);
 375         if (shmflg & SHM_HUGETLB) {
 376                 /* hugetlb_file_setup takes care of mlock user accounting */
 377                 file = hugetlb_file_setup(name, size);
 378                 shp->mlock_user = current->user;
 379         } else {
 380                 int acctflag = VM_ACCOUNT;
 381                 /*
 382                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 383                  * if it's asked for.
 384                  */
 385                 if  ((shmflg & SHM_NORESERVE) &&
 386                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 387                         acctflag = 0;
 388                 file = shmem_file_setup(name, size, acctflag);
 389         }
 390         error = PTR_ERR(file);
 391         if (IS_ERR(file))
 392                 goto no_file;
 393
 394         error = -ENOSPC;
 395         id = shm_addid(ns, shp);
 396         if(id == -1)
 397                 goto no_id;
 398
 399         shp->shm_cprid = task_tgid_vnr(current);
 400         shp->shm_lprid = 0;
 401         shp->shm_atim = shp->shm_dtim = 0;
 402         shp->shm_ctim = get_seconds();
 403         shp->shm_segsz = size;
 404         shp->shm_nattch = 0;
 405         shp->shm_perm.id = shm_buildid(ns, id, shp->shm_perm.seq);
 406         shp->shm_file = file;
 407         /*
 408          * shmid gets reported as "inode#" in /proc/pid/maps.
 409          * proc-ps tools use this. Changing this will break them.
 410          */
 411         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 412
 413         ns->shm_tot += numpages;
 414         error = shp->shm_perm.id;
 415         shm_unlock(shp);
 416         return error;
 417
 418 no_id:
 419         fput(file);
 420 no_file:
 421         security_shm_free(shp);
 422         ipc_rcu_putref(shp);
 423         return error;
 424 }
 425
 426 static inline int shm_security(void *shp, int shmflg)
 427 {
 428         return security_shm_associate((struct shmid_kernel *) shp, shmflg);
 429 }
 430
 431 static inline int shm_more_checks(void *shp, struct ipc_params *params)
 432 {
 433         if (((struct shmid_kernel *)shp)->shm_segsz < params->u.size)
 434                 return -EINVAL;
 435
 436         return 0;
 437 }
 438
 439 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 440 {
 441         struct ipc_namespace *ns;
 442         struct ipc_ops shm_ops;
 443         struct ipc_params shm_params;
 444
 445         ns = current->nsproxy->ipc_ns;
 446
 447         shm_ops.getnew = newseg;
 448         shm_ops.associate = shm_security;
 449         shm_ops.more_checks = shm_more_checks;
 450
 451         shm_params.key = key;
 452         shm_params.flg = shmflg;
 453         shm_params.u.size = size;
 454
 455         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 456 }
 457
 458 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 459 {
 460         switch(version) {
 461         case IPC_64:
 462                 return copy_to_user(buf, in, sizeof(*in));
 463         case IPC_OLD:
 464             {
 465                 struct shmid_ds out;
 466
 467                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 468                 out.shm_segsz   = in->shm_segsz;
 469                 out.shm_atime   = in->shm_atime;
 470                 out.shm_dtime   = in->shm_dtime;
 471                 out.shm_ctime   = in->shm_ctime;
 472                 out.shm_cpid    = in->shm_cpid;
 473                 out.shm_lpid    = in->shm_lpid;
 474                 out.shm_nattch  = in->shm_nattch;
 475
 476                 return copy_to_user(buf, &out, sizeof(out));
 477             }
 478         default:
 479                 return -EINVAL;
 480         }
 481 }
 482
 483 struct shm_setbuf {
 484         uid_t   uid;
 485         gid_t   gid;
 486         mode_t  mode;
 487 };
 488
 489 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 490 {
 491         switch(version) {
 492         case IPC_64:
 493             {
 494                 struct shmid64_ds tbuf;
 495
 496                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 497                         return -EFAULT;
 498
 499                 out->uid        = tbuf.shm_perm.uid;
 500                 out->gid        = tbuf.shm_perm.gid;
 501                 out->mode       = tbuf.shm_perm.mode;
 502
 503                 return 0;
 504             }
 505         case IPC_OLD:
 506             {
 507                 struct shmid_ds tbuf_old;
 508
 509                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 510                         return -EFAULT;
 511
 512                 out->uid        = tbuf_old.shm_perm.uid;
 513                 out->gid        = tbuf_old.shm_perm.gid;
 514                 out->mode       = tbuf_old.shm_perm.mode;
 515
 516                 return 0;
 517             }
 518         default:
 519                 return -EINVAL;
 520         }
 521 }
 522
 523 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 524 {
 525         switch(version) {
 526         case IPC_64:
 527                 return copy_to_user(buf, in, sizeof(*in));
 528         case IPC_OLD:
 529             {
 530                 struct shminfo out;
 531
 532                 if(in->shmmax > INT_MAX)
 533                         out.shmmax = INT_MAX;
 534                 else
 535                         out.shmmax = (int)in->shmmax;
 536
 537                 out.shmmin      = in->shmmin;
 538                 out.shmmni      = in->shmmni;
 539                 out.shmseg      = in->shmseg;
 540                 out.shmall      = in->shmall;
 541
 542                 return copy_to_user(buf, &out, sizeof(out));
 543             }
 544         default:
 545                 return -EINVAL;
 546         }
 547 }
 548
 549 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 550                 unsigned long *swp)
 551 {
 552         int next_id;
 553         int total, in_use;
 554
 555         *rss = 0;
 556         *swp = 0;
 557
 558         in_use = shm_ids(ns).in_use;
 559
 560         for (total = 0, next_id = 0; total < in_use; next_id++) {
 561                 struct shmid_kernel *shp;
 562                 struct inode *inode;
 563
 564                 /*
 565                  * idr_find() is called via shm_get(), so with shm_ids.mutex
 566                  * locked. Since ipc_addid() is also called with
 567                  * shm_ids.mutex down, there is no need to add read barriers
 568                  * here to gurantee the writes in ipc_addid() are seen in
 569                  * order here (for Alpha).
 570                  * However idr_find() itself does not necessary require
 571                  * ipc_ids.mutex down. So if idr_find() is used by other
 572                  * places without ipc_ids.mutex down, then it needs read
 573                  * read memory barriers as ipc_lock() does.
 574                  */
 575
 576                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 577                 if (shp == NULL)
 578                         continue;
 579
 580                 inode = shp->shm_file->f_path.dentry->d_inode;
 581
 582                 if (is_file_hugepages(shp->shm_file)) {
 583                         struct address_space *mapping = inode->i_mapping;
 584                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 585                 } else {
 586                         struct shmem_inode_info *info = SHMEM_I(inode);
 587                         spin_lock(&info->lock);
 588                         *rss += inode->i_mapping->nrpages;
 589                         *swp += info->swapped;
 590                         spin_unlock(&info->lock);
 591                 }
 592
 593                 total++;
 594         }
 595 }
 596
 597 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 598 {
 599         struct shm_setbuf setbuf;
 600         struct shmid_kernel *shp;
 601         int err, version;
 602         struct ipc_namespace *ns;
 603
 604         if (cmd < 0 || shmid < 0) {
 605                 err = -EINVAL;
 606                 goto out;
 607         }
 608
 609         version = ipc_parse_version(&cmd);
 610         ns = current->nsproxy->ipc_ns;
 611
 612         switch (cmd) { /* replace with proc interface ? */
 613         case IPC_INFO:
 614         {
 615                 struct shminfo64 shminfo;
 616
 617                 err = security_shm_shmctl(NULL, cmd);
 618                 if (err)
 619                         return err;
 620
 621                 memset(&shminfo,0,sizeof(shminfo));
 622                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 623                 shminfo.shmmax = ns->shm_ctlmax;
 624                 shminfo.shmall = ns->shm_ctlall;
 625
 626                 shminfo.shmmin = SHMMIN;
 627                 if(copy_shminfo_to_user (buf, &shminfo, version))
 628                         return -EFAULT;
 629                 /* reading a integer is always atomic */
 630                 err = ipc_get_maxid(&shm_ids(ns));
 631                 if(err<0)
 632                         err = 0;
 633                 goto out;
 634         }
 635         case SHM_INFO:
 636         {
 637                 struct shm_info shm_info;
 638
 639                 err = security_shm_shmctl(NULL, cmd);
 640                 if (err)
 641                         return err;
 642
 643                 memset(&shm_info,0,sizeof(shm_info));
 644                 mutex_lock(&shm_ids(ns).mutex);
 645                 shm_info.used_ids = shm_ids(ns).in_use;
 646                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 647                 shm_info.shm_tot = ns->shm_tot;
 648                 shm_info.swap_attempts = 0;
 649                 shm_info.swap_successes = 0;
 650                 err = ipc_get_maxid(&shm_ids(ns));
 651                 mutex_unlock(&shm_ids(ns).mutex);
 652                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 653                         err = -EFAULT;
 654                         goto out;
 655                 }
 656
 657                 err = err < 0 ? 0 : err;
 658                 goto out;
 659         }
 660         case SHM_STAT:
 661         case IPC_STAT:
 662         {
 663                 struct shmid64_ds tbuf;
 664                 int result;
 665                 memset(&tbuf, 0, sizeof(tbuf));
 666                 shp = shm_lock(ns, shmid);
 667                 if(shp==NULL) {
 668                         err = -EINVAL;
 669                         goto out;
 670                 } else if (cmd == SHM_STAT) {
 671                         result = shp->shm_perm.id;
 672                 } else {
 673                         err = shm_checkid(ns, shp,shmid);
 674                         if(err)
 675                                 goto out_unlock;
 676                         result = 0;
 677                 }
 678                 err=-EACCES;
 679                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 680                         goto out_unlock;
 681                 err = security_shm_shmctl(shp, cmd);
 682                 if (err)
 683                         goto out_unlock;
 684                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 685                 tbuf.shm_segsz  = shp->shm_segsz;
 686                 tbuf.shm_atime  = shp->shm_atim;
 687                 tbuf.shm_dtime  = shp->shm_dtim;
 688                 tbuf.shm_ctime  = shp->shm_ctim;
 689                 tbuf.shm_cpid   = shp->shm_cprid;
 690                 tbuf.shm_lpid   = shp->shm_lprid;
 691                 tbuf.shm_nattch = shp->shm_nattch;
 692                 shm_unlock(shp);
 693                 if(copy_shmid_to_user (buf, &tbuf, version))
 694                         err = -EFAULT;
 695                 else
 696                         err = result;
 697                 goto out;
 698         }
 699         case SHM_LOCK:
 700         case SHM_UNLOCK:
 701         {
 702                 shp = shm_lock(ns, shmid);
 703                 if(shp==NULL) {
 704                         err = -EINVAL;
 705                         goto out;
 706                 }
 707                 err = shm_checkid(ns, shp,shmid);
 708                 if(err)
 709                         goto out_unlock;
 710
 711                 err = audit_ipc_obj(&(shp->shm_perm));
 712                 if (err)
 713                         goto out_unlock;
 714
 715                 if (!capable(CAP_IPC_LOCK)) {
 716                         err = -EPERM;
 717                         if (current->euid != shp->shm_perm.uid &&
 718                             current->euid != shp->shm_perm.cuid)
 719                                 goto out_unlock;
 720                         if (cmd == SHM_LOCK &&
 721                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 722                                 goto out_unlock;
 723                 }
 724
 725                 err = security_shm_shmctl(shp, cmd);
 726                 if (err)
 727                         goto out_unlock;
 728
 729                 if(cmd==SHM_LOCK) {
 730                         struct user_struct * user = current->user;
 731                         if (!is_file_hugepages(shp->shm_file)) {
 732                                 err = shmem_lock(shp->shm_file, 1, user);
 733                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 734                                         shp->shm_perm.mode |= SHM_LOCKED;
 735                                         shp->mlock_user = user;
 736                                 }
 737                         }
 738                 } else if (!is_file_hugepages(shp->shm_file)) {
 739                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 740                         shp->shm_perm.mode &= ~SHM_LOCKED;
 741                         shp->mlock_user = NULL;
 742                 }
 743                 shm_unlock(shp);
 744                 goto out;
 745         }
 746         case IPC_RMID:
 747         {
 748                 /*
 749                  *      We cannot simply remove the file. The SVID states
 750                  *      that the block remains until the last person
 751                  *      detaches from it, then is deleted. A shmat() on
 752                  *      an RMID segment is legal in older Linux and if
 753                  *      we change it apps break...
 754                  *
 755                  *      Instead we set a destroyed flag, and then blow
 756                  *      the name away when the usage hits zero.
 757                  */
 758                 mutex_lock(&shm_ids(ns).mutex);
 759                 shp = shm_lock(ns, shmid);
 760                 err = -EINVAL;
 761                 if (shp == NULL)
 762                         goto out_up;
 763                 err = shm_checkid(ns, shp, shmid);
 764                 if(err)
 765                         goto out_unlock_up;
 766
 767                 err = audit_ipc_obj(&(shp->shm_perm));
 768                 if (err)
 769                         goto out_unlock_up;
 770
 771                 if (current->euid != shp->shm_perm.uid &&
 772                     current->euid != shp->shm_perm.cuid &&
 773                     !capable(CAP_SYS_ADMIN)) {
 774                         err=-EPERM;
 775                         goto out_unlock_up;
 776                 }
 777
 778                 err = security_shm_shmctl(shp, cmd);
 779                 if (err)
 780                         goto out_unlock_up;
 781
 782                 do_shm_rmid(ns, shp);
 783                 mutex_unlock(&shm_ids(ns).mutex);
 784                 goto out;
 785         }
 786
 787         case IPC_SET:
 788         {
 789                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 790                         err = -EFAULT;
 791                         goto out;
 792                 }
 793                 mutex_lock(&shm_ids(ns).mutex);
 794                 shp = shm_lock(ns, shmid);
 795                 err=-EINVAL;
 796                 if(shp==NULL)
 797                         goto out_up;
 798                 err = shm_checkid(ns, shp,shmid);
 799                 if(err)
 800                         goto out_unlock_up;
 801                 err = audit_ipc_obj(&(shp->shm_perm));
 802                 if (err)
 803                         goto out_unlock_up;
 804                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 805                 if (err)
 806                         goto out_unlock_up;
 807                 err=-EPERM;
 808                 if (current->euid != shp->shm_perm.uid &&
 809                     current->euid != shp->shm_perm.cuid &&
 810                     !capable(CAP_SYS_ADMIN)) {
 811                         goto out_unlock_up;
 812                 }
 813
 814                 err = security_shm_shmctl(shp, cmd);
 815                 if (err)
 816                         goto out_unlock_up;
 817
 818                 shp->shm_perm.uid = setbuf.uid;
 819                 shp->shm_perm.gid = setbuf.gid;
 820                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 821                         | (setbuf.mode & S_IRWXUGO);
 822                 shp->shm_ctim = get_seconds();
 823                 break;
 824         }
 825
 826         default:
 827                 err = -EINVAL;
 828                 goto out;
 829         }
 830
 831         err = 0;
 832 out_unlock_up:
 833         shm_unlock(shp);
 834 out_up:
 835         mutex_unlock(&shm_ids(ns).mutex);
 836         goto out;
 837 out_unlock:
 838         shm_unlock(shp);
 839 out:
 840         return err;
 841 }
 842
 843 /*
 844  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 845  *
 846  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 847  * "raddr" thing points to kernel space, and there has to be a wrapper around
 848  * this.
 849  */
 850 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 851 {
 852         struct shmid_kernel *shp;
 853         unsigned long addr;
 854         unsigned long size;
 855         struct file * file;
 856         int    err;
 857         unsigned long flags;
 858         unsigned long prot;
 859         int acc_mode;
 860         unsigned long user_addr;
 861         struct ipc_namespace *ns;
 862         struct shm_file_data *sfd;
 863         struct path path;
 864         mode_t f_mode;
 865
 866         err = -EINVAL;
 867         if (shmid < 0)
 868                 goto out;
 869         else if ((addr = (ulong)shmaddr)) {
 870                 if (addr & (SHMLBA-1)) {
 871                         if (shmflg & SHM_RND)
 872                                 addr &= ~(SHMLBA-1);       /* round down */
 873                         else
 874 #ifndef __ARCH_FORCE_SHMLBA
 875                                 if (addr & ~PAGE_MASK)
 876 #endif
 877                                         goto out;
 878                 }
 879                 flags = MAP_SHARED | MAP_FIXED;
 880         } else {
 881                 if ((shmflg & SHM_REMAP))
 882                         goto out;
 883
 884                 flags = MAP_SHARED;
 885         }
 886
 887         if (shmflg & SHM_RDONLY) {
 888                 prot = PROT_READ;
 889                 acc_mode = S_IRUGO;
 890                 f_mode = FMODE_READ;
 891         } else {
 892                 prot = PROT_READ | PROT_WRITE;
 893                 acc_mode = S_IRUGO | S_IWUGO;
 894                 f_mode = FMODE_READ | FMODE_WRITE;
 895         }
 896         if (shmflg & SHM_EXEC) {
 897                 prot |= PROT_EXEC;
 898                 acc_mode |= S_IXUGO;
 899         }
 900
 901         /*
 902          * We cannot rely on the fs check since SYSV IPC does have an
 903          * additional creator id...
 904          */
 905         ns = current->nsproxy->ipc_ns;
 906         shp = shm_lock(ns, shmid);
 907         if(shp == NULL)
 908                 goto out;
 909
 910         err = shm_checkid(ns, shp,shmid);
 911         if (err)
 912                 goto out_unlock;
 913
 914         err = -EACCES;
 915         if (ipcperms(&shp->shm_perm, acc_mode))
 916                 goto out_unlock;
 917
 918         err = security_shm_shmat(shp, shmaddr, shmflg);
 919         if (err)
 920                 goto out_unlock;
 921
 922         path.dentry = dget(shp->shm_file->f_path.dentry);
 923         path.mnt    = shp->shm_file->f_path.mnt;
 924         shp->shm_nattch++;
 925         size = i_size_read(path.dentry->d_inode);
 926         shm_unlock(shp);
 927
 928         err = -ENOMEM;
 929         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 930         if (!sfd)
 931                 goto out_put_dentry;
 932
 933         err = -ENOMEM;
 934
 935         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 936         if (!file)
 937                 goto out_free;
 938
 939         file->private_data = sfd;
 940         file->f_mapping = shp->shm_file->f_mapping;
 941         sfd->id = shp->shm_perm.id;
 942         sfd->ns = get_ipc_ns(ns);
 943         sfd->file = shp->shm_file;
 944         sfd->vm_ops = NULL;
 945
 946         down_write(&current->mm->mmap_sem);
 947         if (addr && !(shmflg & SHM_REMAP)) {
 948                 err = -EINVAL;
 949                 if (find_vma_intersection(current->mm, addr, addr + size))
 950                         goto invalid;
 951                 /*
 952                  * If shm segment goes below stack, make sure there is some
 953                  * space left for the stack to grow (at least 4 pages).
 954                  */
 955                 if (addr < current->mm->start_stack &&
 956                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 957                         goto invalid;
 958         }
 959
 960         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 961         *raddr = user_addr;
 962         err = 0;
 963         if (IS_ERR_VALUE(user_addr))
 964                 err = (long)user_addr;
 965 invalid:
 966         up_write(&current->mm->mmap_sem);
 967
 968         fput(file);
 969
 970 out_nattch:
 971         mutex_lock(&shm_ids(ns).mutex);
 972         shp = shm_lock(ns, shmid);
 973         BUG_ON(!shp);
 974         shp->shm_nattch--;
 975         if(shp->shm_nattch == 0 &&
 976            shp->shm_perm.mode & SHM_DEST)
 977                 shm_destroy(ns, shp);
 978         else
 979                 shm_unlock(shp);
 980         mutex_unlock(&shm_ids(ns).mutex);
 981
 982 out:
 983         return err;
 984
 985 out_unlock:
 986         shm_unlock(shp);
 987         goto out;
 988
 989 out_free:
 990         kfree(sfd);
 991 out_put_dentry:
 992         dput(path.dentry);
 993         goto out_nattch;
 994 }
 995
 996 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 997 {
 998         unsigned long ret;
 999         long err;
1000
1001         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1002         if (err)
1003                 return err;
1004         force_successful_syscall_return();
1005         return (long)ret;
1006 }
1007
1008 /*
1009  * detach and kill segment if marked destroyed.
1010  * The work is done in shm_close.
1011  */
1012 asmlinkage long sys_shmdt(char __user *shmaddr)
1013 {
1014         struct mm_struct *mm = current->mm;
1015         struct vm_area_struct *vma, *next;
1016         unsigned long addr = (unsigned long)shmaddr;
1017         loff_t size = 0;
1018         int retval = -EINVAL;
1019
1020         if (addr & ~PAGE_MASK)
1021                 return retval;
1022
1023         down_write(&mm->mmap_sem);
1024
1025         /*
1026          * This function tries to be smart and unmap shm segments that
1027          * were modified by partial mlock or munmap calls:
1028          * - It first determines the size of the shm segment that should be
1029          *   unmapped: It searches for a vma that is backed by shm and that
1030          *   started at address shmaddr. It records it's size and then unmaps
1031          *   it.
1032          * - Then it unmaps all shm vmas that started at shmaddr and that
1033          *   are within the initially determined size.
1034          * Errors from do_munmap are ignored: the function only fails if
1035          * it's called with invalid parameters or if it's called to unmap
1036          * a part of a vma. Both calls in this function are for full vmas,
1037          * the parameters are directly copied from the vma itself and always
1038          * valid - therefore do_munmap cannot fail. (famous last words?)
1039          */
1040         /*
1041          * If it had been mremap()'d, the starting address would not
1042          * match the usual checks anyway. So assume all vma's are
1043          * above the starting address given.
1044          */
1045         vma = find_vma(mm, addr);
1046
1047         while (vma) {
1048                 next = vma->vm_next;
1049
1050                 /*
1051                  * Check if the starting address would match, i.e. it's
1052                  * a fragment created by mprotect() and/or munmap(), or it
1053                  * otherwise it starts at this address with no hassles.
1054                  */
1055                 if ((vma->vm_ops == &shm_vm_ops) &&
1056                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1057
1058
1059                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1060                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1061                         /*
1062                          * We discovered the size of the shm segment, so
1063                          * break out of here and fall through to the next
1064                          * loop that uses the size information to stop
1065                          * searching for matching vma's.
1066                          */
1067                         retval = 0;
1068                         vma = next;
1069                         break;
1070                 }
1071                 vma = next;
1072         }
1073
1074         /*
1075          * We need look no further than the maximum address a fragment
1076          * could possibly have landed at. Also cast things to loff_t to
1077          * prevent overflows and make comparisions vs. equal-width types.
1078          */
1079         size = PAGE_ALIGN(size);
1080         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1081                 next = vma->vm_next;
1082
1083                 /* finding a matching vma now does not alter retval */
1084                 if ((vma->vm_ops == &shm_vm_ops) &&
1085                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1086
1087                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1088                 vma = next;
1089         }
1090
1091         up_write(&mm->mmap_sem);
1092         return retval;
1093 }
1094
1095 #ifdef CONFIG_PROC_FS
1096 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1097 {
1098         struct shmid_kernel *shp = it;
1099         char *format;
1100
1101 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1102 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1103
1104         if (sizeof(size_t) <= sizeof(int))
1105                 format = SMALL_STRING;
1106         else
1107                 format = BIG_STRING;
1108         return seq_printf(s, format,
1109                           shp->shm_perm.key,
1110                           shp->shm_perm.id,
1111                           shp->shm_perm.mode,
1112                           shp->shm_segsz,
1113                           shp->shm_cprid,
1114                           shp->shm_lprid,
1115                           shp->shm_nattch,
1116                           shp->shm_perm.uid,
1117                           shp->shm_perm.gid,
1118                           shp->shm_perm.cuid,
1119                           shp->shm_perm.cgid,
1120                           shp->shm_atim,
1121                           shp->shm_dtim,
1122                           shp->shm_ctim);
1123 }
1124 #endif