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/rwsem.h>
  39 #include <linux/nsproxy.h>
  40 #include <linux/mount.h>
  41 #include <linux/ipc_namespace.h>
  42 #include <trace/ipc.h>
  43
  44 #include <asm/uaccess.h>
  45
  46 #include "util.h"
  47
  48 struct shm_file_data {
  49         int id;
  50         struct ipc_namespace *ns;
  51         struct file *file;
  52         const struct vm_operations_struct *vm_ops;
  53 };
  54
  55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  56
  57 static const struct file_operations shm_file_operations;
  58 static struct vm_operations_struct shm_vm_ops;
  59
  60 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  61
  62 #define shm_unlock(shp)                 \
  63         ipc_unlock(&(shp)->shm_perm)
  64
  65 static int newseg(struct ipc_namespace *, struct ipc_params *);
  66 static void shm_open(struct vm_area_struct *vma);
  67 static void shm_close(struct vm_area_struct *vma);
  68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  69 #ifdef CONFIG_PROC_FS
  70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  71 #endif
  72
  73 void shm_init_ns(struct ipc_namespace *ns)
  74 {
  75         ns->shm_ctlmax = SHMMAX;
  76         ns->shm_ctlall = SHMALL;
  77         ns->shm_ctlmni = SHMMNI;
  78         ns->shm_tot = 0;
  79         ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
  80 }
  81
  82 /*
  83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  84  * Only shm_ids.rw_mutex remains locked on exit.
  85  */
  86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  87 {
  88         struct shmid_kernel *shp;
  89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  90
  91         if (shp->shm_nattch){
  92                 shp->shm_perm.mode |= SHM_DEST;
  93                 /* Do not find it any more */
  94                 shp->shm_perm.key = IPC_PRIVATE;
  95                 shm_unlock(shp);
  96         } else
  97                 shm_destroy(ns, shp);
  98 }
  99
 100 #ifdef CONFIG_IPC_NS
 101 void shm_exit_ns(struct ipc_namespace *ns)
 102 {
 103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 104 }
 105 #endif
 106
 107 void __init shm_init (void)
 108 {
 109         shm_init_ns(&init_ipc_ns);
 110         ipc_init_proc_interface("sysvipc/shm",
 111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 113 }
 114
 115 /*
 116  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 117  * is not necessarily held.
 118  */
 119 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 120 {
 121         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 122
 123         if (IS_ERR(ipcp))
 124                 return (struct shmid_kernel *)ipcp;
 125
 126         return container_of(ipcp, struct shmid_kernel, shm_perm);
 127 }
 128
 129 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 130                                                 int id)
 131 {
 132         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 133
 134         if (IS_ERR(ipcp))
 135                 return (struct shmid_kernel *)ipcp;
 136
 137         return container_of(ipcp, struct shmid_kernel, shm_perm);
 138 }
 139
 140 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 141 {
 142         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 143 }
 144
 145
 146 /* This is called by fork, once for every shm attach. */
 147 static void shm_open(struct vm_area_struct *vma)
 148 {
 149         struct file *file = vma->vm_file;
 150         struct shm_file_data *sfd = shm_file_data(file);
 151         struct shmid_kernel *shp;
 152
 153         shp = shm_lock(sfd->ns, sfd->id);
 154         BUG_ON(IS_ERR(shp));
 155         shp->shm_atim = get_seconds();
 156         shp->shm_lprid = task_tgid_vnr(current);
 157         shp->shm_nattch++;
 158         shm_unlock(shp);
 159 }
 160
 161 /*
 162  * shm_destroy - free the struct shmid_kernel
 163  *
 164  * @ns: namespace
 165  * @shp: struct to free
 166  *
 167  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 168  * but returns with shp unlocked and freed.
 169  */
 170 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 171 {
 172         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 173         shm_rmid(ns, shp);
 174         shm_unlock(shp);
 175         if (!is_file_hugepages(shp->shm_file))
 176                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 177         else
 178                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 179                                                 shp->mlock_user);
 180         fput (shp->shm_file);
 181         security_shm_free(shp);
 182         ipc_rcu_putref(shp);
 183 }
 184
 185 /*
 186  * remove the attach descriptor vma.
 187  * free memory for segment if it is marked destroyed.
 188  * The descriptor has already been removed from the current->mm->mmap list
 189  * and will later be kfree()d.
 190  */
 191 static void shm_close(struct vm_area_struct *vma)
 192 {
 193         struct file * file = vma->vm_file;
 194         struct shm_file_data *sfd = shm_file_data(file);
 195         struct shmid_kernel *shp;
 196         struct ipc_namespace *ns = sfd->ns;
 197
 198         down_write(&shm_ids(ns).rw_mutex);
 199         /* remove from the list of attaches of the shm segment */
 200         shp = shm_lock(ns, sfd->id);
 201         BUG_ON(IS_ERR(shp));
 202         shp->shm_lprid = task_tgid_vnr(current);
 203         shp->shm_dtim = get_seconds();
 204         shp->shm_nattch--;
 205         if(shp->shm_nattch == 0 &&
 206            shp->shm_perm.mode & SHM_DEST)
 207                 shm_destroy(ns, shp);
 208         else
 209                 shm_unlock(shp);
 210         up_write(&shm_ids(ns).rw_mutex);
 211 }
 212
 213 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 214 {
 215         struct file *file = vma->vm_file;
 216         struct shm_file_data *sfd = shm_file_data(file);
 217
 218         return sfd->vm_ops->fault(vma, vmf);
 219 }
 220
 221 #ifdef CONFIG_NUMA
 222 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 223 {
 224         struct file *file = vma->vm_file;
 225         struct shm_file_data *sfd = shm_file_data(file);
 226         int err = 0;
 227         if (sfd->vm_ops->set_policy)
 228                 err = sfd->vm_ops->set_policy(vma, new);
 229         return err;
 230 }
 231
 232 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 233                                         unsigned long addr)
 234 {
 235         struct file *file = vma->vm_file;
 236         struct shm_file_data *sfd = shm_file_data(file);
 237         struct mempolicy *pol = NULL;
 238
 239         if (sfd->vm_ops->get_policy)
 240                 pol = sfd->vm_ops->get_policy(vma, addr);
 241         else if (vma->vm_policy)
 242                 pol = vma->vm_policy;
 243
 244         return pol;
 245 }
 246 #endif
 247
 248 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 249 {
 250         struct shm_file_data *sfd = shm_file_data(file);
 251         int ret;
 252
 253         ret = sfd->file->f_op->mmap(sfd->file, vma);
 254         if (ret != 0)
 255                 return ret;
 256         sfd->vm_ops = vma->vm_ops;
 257 #ifdef CONFIG_MMU
 258         BUG_ON(!sfd->vm_ops->fault);
 259 #endif
 260         vma->vm_ops = &shm_vm_ops;
 261         shm_open(vma);
 262
 263         return ret;
 264 }
 265
 266 static int shm_release(struct inode *ino, struct file *file)
 267 {
 268         struct shm_file_data *sfd = shm_file_data(file);
 269
 270         put_ipc_ns(sfd->ns);
 271         shm_file_data(file) = NULL;
 272         kfree(sfd);
 273         return 0;
 274 }
 275
 276 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 277 {
 278         int (*fsync) (struct file *, struct dentry *, int datasync);
 279         struct shm_file_data *sfd = shm_file_data(file);
 280         int ret = -EINVAL;
 281
 282         fsync = sfd->file->f_op->fsync;
 283         if (fsync)
 284                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 285         return ret;
 286 }
 287
 288 static unsigned long shm_get_unmapped_area(struct file *file,
 289         unsigned long addr, unsigned long len, unsigned long pgoff,
 290         unsigned long flags)
 291 {
 292         struct shm_file_data *sfd = shm_file_data(file);
 293         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 294 }
 295
 296 int is_file_shm_hugepages(struct file *file)
 297 {
 298         int ret = 0;
 299
 300         if (file->f_op == &shm_file_operations) {
 301                 struct shm_file_data *sfd;
 302                 sfd = shm_file_data(file);
 303                 ret = is_file_hugepages(sfd->file);
 304         }
 305         return ret;
 306 }
 307
 308 static const struct file_operations shm_file_operations = {
 309         .mmap           = shm_mmap,
 310         .fsync          = shm_fsync,
 311         .release        = shm_release,
 312         .get_unmapped_area      = shm_get_unmapped_area,
 313 };
 314
 315 static struct vm_operations_struct shm_vm_ops = {
 316         .open   = shm_open,     /* callback for a new vm-area open */
 317         .close  = shm_close,    /* callback for when the vm-area is released */
 318         .fault  = shm_fault,
 319 #if defined(CONFIG_NUMA)
 320         .set_policy = shm_set_policy,
 321         .get_policy = shm_get_policy,
 322 #endif
 323 };
 324
 325 /**
 326  * newseg - Create a new shared memory segment
 327  * @ns: namespace
 328  * @params: ptr to the structure that contains key, size and shmflg
 329  *
 330  * Called with shm_ids.rw_mutex held as a writer.
 331  */
 332
 333 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 334 {
 335         key_t key = params->key;
 336         int shmflg = params->flg;
 337         size_t size = params->u.size;
 338         int error;
 339         struct shmid_kernel *shp;
 340         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 341         struct file * file;
 342         char name[13];
 343         int id;
 344
 345         if (size < SHMMIN || size > ns->shm_ctlmax)
 346                 return -EINVAL;
 347
 348         if (ns->shm_tot + numpages > ns->shm_ctlall)
 349                 return -ENOSPC;
 350
 351         shp = ipc_rcu_alloc(sizeof(*shp));
 352         if (!shp)
 353                 return -ENOMEM;
 354
 355         shp->shm_perm.key = key;
 356         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 357         shp->mlock_user = NULL;
 358
 359         shp->shm_perm.security = NULL;
 360         error = security_shm_alloc(shp);
 361         if (error) {
 362                 ipc_rcu_putref(shp);
 363                 return error;
 364         }
 365
 366         sprintf (name, "SYSV%08x", key);
 367         if (shmflg & SHM_HUGETLB) {
 368                 /* hugetlb_file_setup takes care of mlock user accounting */
 369                 file = hugetlb_file_setup(name, size);
 370                 shp->mlock_user = current->user;
 371         } else {
 372                 int acctflag = VM_ACCOUNT;
 373                 /*
 374                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 375                  * if it's asked for.
 376                  */
 377                 if  ((shmflg & SHM_NORESERVE) &&
 378                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 379                         acctflag = 0;
 380                 file = shmem_file_setup(name, size, acctflag);
 381         }
 382         error = PTR_ERR(file);
 383         if (IS_ERR(file))
 384                 goto no_file;
 385
 386         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 387         if (id < 0) {
 388                 error = id;
 389                 goto no_id;
 390         }
 391
 392         shp->shm_cprid = task_tgid_vnr(current);
 393         shp->shm_lprid = 0;
 394         shp->shm_atim = shp->shm_dtim = 0;
 395         shp->shm_ctim = get_seconds();
 396         shp->shm_segsz = size;
 397         shp->shm_nattch = 0;
 398         shp->shm_file = file;
 399         /*
 400          * shmid gets reported as "inode#" in /proc/pid/maps.
 401          * proc-ps tools use this. Changing this will break them.
 402          */
 403         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 404
 405         ns->shm_tot += numpages;
 406         error = shp->shm_perm.id;
 407         shm_unlock(shp);
 408         return error;
 409
 410 no_id:
 411         fput(file);
 412 no_file:
 413         security_shm_free(shp);
 414         ipc_rcu_putref(shp);
 415         return error;
 416 }
 417
 418 /*
 419  * Called with shm_ids.rw_mutex and ipcp locked.
 420  */
 421 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 422 {
 423         struct shmid_kernel *shp;
 424
 425         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 426         return security_shm_associate(shp, shmflg);
 427 }
 428
 429 /*
 430  * Called with shm_ids.rw_mutex and ipcp locked.
 431  */
 432 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 433                                 struct ipc_params *params)
 434 {
 435         struct shmid_kernel *shp;
 436
 437         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 438         if (shp->shm_segsz < params->u.size)
 439                 return -EINVAL;
 440
 441         return 0;
 442 }
 443
 444 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 445 {
 446         struct ipc_namespace *ns;
 447         struct ipc_ops shm_ops;
 448         struct ipc_params shm_params;
 449         long err;
 450
 451         ns = current->nsproxy->ipc_ns;
 452
 453         shm_ops.getnew = newseg;
 454         shm_ops.associate = shm_security;
 455         shm_ops.more_checks = shm_more_checks;
 456
 457         shm_params.key = key;
 458         shm_params.flg = shmflg;
 459         shm_params.u.size = size;
 460
 461         err = ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 462         trace_ipc_shm_create(err, shmflg);
 463         return err;
 464 }
 465
 466 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 467 {
 468         switch(version) {
 469         case IPC_64:
 470                 return copy_to_user(buf, in, sizeof(*in));
 471         case IPC_OLD:
 472             {
 473                 struct shmid_ds out;
 474
 475                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 476                 out.shm_segsz   = in->shm_segsz;
 477                 out.shm_atime   = in->shm_atime;
 478                 out.shm_dtime   = in->shm_dtime;
 479                 out.shm_ctime   = in->shm_ctime;
 480                 out.shm_cpid    = in->shm_cpid;
 481                 out.shm_lpid    = in->shm_lpid;
 482                 out.shm_nattch  = in->shm_nattch;
 483
 484                 return copy_to_user(buf, &out, sizeof(out));
 485             }
 486         default:
 487                 return -EINVAL;
 488         }
 489 }
 490
 491 static inline unsigned long
 492 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 493 {
 494         switch(version) {
 495         case IPC_64:
 496                 if (copy_from_user(out, buf, sizeof(*out)))
 497                         return -EFAULT;
 498                 return 0;
 499         case IPC_OLD:
 500             {
 501                 struct shmid_ds tbuf_old;
 502
 503                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 504                         return -EFAULT;
 505
 506                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 507                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 508                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 509
 510                 return 0;
 511             }
 512         default:
 513                 return -EINVAL;
 514         }
 515 }
 516
 517 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 518 {
 519         switch(version) {
 520         case IPC_64:
 521                 return copy_to_user(buf, in, sizeof(*in));
 522         case IPC_OLD:
 523             {
 524                 struct shminfo out;
 525
 526                 if(in->shmmax > INT_MAX)
 527                         out.shmmax = INT_MAX;
 528                 else
 529                         out.shmmax = (int)in->shmmax;
 530
 531                 out.shmmin      = in->shmmin;
 532                 out.shmmni      = in->shmmni;
 533                 out.shmseg      = in->shmseg;
 534                 out.shmall      = in->shmall;
 535
 536                 return copy_to_user(buf, &out, sizeof(out));
 537             }
 538         default:
 539                 return -EINVAL;
 540         }
 541 }
 542
 543 /*
 544  * Called with shm_ids.rw_mutex held as a reader
 545  */
 546 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 547                 unsigned long *swp)
 548 {
 549         int next_id;
 550         int total, in_use;
 551
 552         *rss = 0;
 553         *swp = 0;
 554
 555         in_use = shm_ids(ns).in_use;
 556
 557         for (total = 0, next_id = 0; total < in_use; next_id++) {
 558                 struct shmid_kernel *shp;
 559                 struct inode *inode;
 560
 561                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 562                 if (shp == NULL)
 563                         continue;
 564
 565                 inode = shp->shm_file->f_path.dentry->d_inode;
 566
 567                 if (is_file_hugepages(shp->shm_file)) {
 568                         struct address_space *mapping = inode->i_mapping;
 569                         struct hstate *h = hstate_file(shp->shm_file);
 570                         *rss += pages_per_huge_page(h) * 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                 total++;
 580         }
 581 }
 582
 583 /*
 584  * This function handles some shmctl commands which require the rw_mutex
 585  * to be held in write mode.
 586  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 587  */
 588 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 589                        struct shmid_ds __user *buf, int version)
 590 {
 591         struct kern_ipc_perm *ipcp;
 592         struct shmid64_ds shmid64;
 593         struct shmid_kernel *shp;
 594         int err;
 595
 596         if (cmd == IPC_SET) {
 597                 if (copy_shmid_from_user(&shmid64, buf, version))
 598                         return -EFAULT;
 599         }
 600
 601         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 602         if (IS_ERR(ipcp))
 603                 return PTR_ERR(ipcp);
 604
 605         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 606
 607         err = security_shm_shmctl(shp, cmd);
 608         if (err)
 609                 goto out_unlock;
 610         switch (cmd) {
 611         case IPC_RMID:
 612                 do_shm_rmid(ns, ipcp);
 613                 goto out_up;
 614         case IPC_SET:
 615                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 616                 shp->shm_ctim = get_seconds();
 617                 break;
 618         default:
 619                 err = -EINVAL;
 620         }
 621 out_unlock:
 622         shm_unlock(shp);
 623 out_up:
 624         up_write(&shm_ids(ns).rw_mutex);
 625         return err;
 626 }
 627
 628 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 629 {
 630         struct shmid_kernel *shp;
 631         int err, version;
 632         struct ipc_namespace *ns;
 633
 634         if (cmd < 0 || shmid < 0) {
 635                 err = -EINVAL;
 636                 goto out;
 637         }
 638
 639         version = ipc_parse_version(&cmd);
 640         ns = current->nsproxy->ipc_ns;
 641
 642         switch (cmd) { /* replace with proc interface ? */
 643         case IPC_INFO:
 644         {
 645                 struct shminfo64 shminfo;
 646
 647                 err = security_shm_shmctl(NULL, cmd);
 648                 if (err)
 649                         return err;
 650
 651                 memset(&shminfo,0,sizeof(shminfo));
 652                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 653                 shminfo.shmmax = ns->shm_ctlmax;
 654                 shminfo.shmall = ns->shm_ctlall;
 655
 656                 shminfo.shmmin = SHMMIN;
 657                 if(copy_shminfo_to_user (buf, &shminfo, version))
 658                         return -EFAULT;
 659
 660                 down_read(&shm_ids(ns).rw_mutex);
 661                 err = ipc_get_maxid(&shm_ids(ns));
 662                 up_read(&shm_ids(ns).rw_mutex);
 663
 664                 if(err<0)
 665                         err = 0;
 666                 goto out;
 667         }
 668         case SHM_INFO:
 669         {
 670                 struct shm_info shm_info;
 671
 672                 err = security_shm_shmctl(NULL, cmd);
 673                 if (err)
 674                         return err;
 675
 676                 memset(&shm_info,0,sizeof(shm_info));
 677                 down_read(&shm_ids(ns).rw_mutex);
 678                 shm_info.used_ids = shm_ids(ns).in_use;
 679                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 680                 shm_info.shm_tot = ns->shm_tot;
 681                 shm_info.swap_attempts = 0;
 682                 shm_info.swap_successes = 0;
 683                 err = ipc_get_maxid(&shm_ids(ns));
 684                 up_read(&shm_ids(ns).rw_mutex);
 685                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 686                         err = -EFAULT;
 687                         goto out;
 688                 }
 689
 690                 err = err < 0 ? 0 : err;
 691                 goto out;
 692         }
 693         case SHM_STAT:
 694         case IPC_STAT:
 695         {
 696                 struct shmid64_ds tbuf;
 697                 int result;
 698
 699                 if (!buf) {
 700                         err = -EFAULT;
 701                         goto out;
 702                 }
 703
 704                 if (cmd == SHM_STAT) {
 705                         shp = shm_lock(ns, shmid);
 706                         if (IS_ERR(shp)) {
 707                                 err = PTR_ERR(shp);
 708                                 goto out;
 709                         }
 710                         result = shp->shm_perm.id;
 711                 } else {
 712                         shp = shm_lock_check(ns, shmid);
 713                         if (IS_ERR(shp)) {
 714                                 err = PTR_ERR(shp);
 715                                 goto out;
 716                         }
 717                         result = 0;
 718                 }
 719                 err=-EACCES;
 720                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 721                         goto out_unlock;
 722                 err = security_shm_shmctl(shp, cmd);
 723                 if (err)
 724                         goto out_unlock;
 725                 memset(&tbuf, 0, sizeof(tbuf));
 726                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 727                 tbuf.shm_segsz  = shp->shm_segsz;
 728                 tbuf.shm_atime  = shp->shm_atim;
 729                 tbuf.shm_dtime  = shp->shm_dtim;
 730                 tbuf.shm_ctime  = shp->shm_ctim;
 731                 tbuf.shm_cpid   = shp->shm_cprid;
 732                 tbuf.shm_lpid   = shp->shm_lprid;
 733                 tbuf.shm_nattch = shp->shm_nattch;
 734                 shm_unlock(shp);
 735                 if(copy_shmid_to_user (buf, &tbuf, version))
 736                         err = -EFAULT;
 737                 else
 738                         err = result;
 739                 goto out;
 740         }
 741         case SHM_LOCK:
 742         case SHM_UNLOCK:
 743         {
 744                 shp = shm_lock_check(ns, shmid);
 745                 if (IS_ERR(shp)) {
 746                         err = PTR_ERR(shp);
 747                         goto out;
 748                 }
 749
 750                 err = audit_ipc_obj(&(shp->shm_perm));
 751                 if (err)
 752                         goto out_unlock;
 753
 754                 if (!capable(CAP_IPC_LOCK)) {
 755                         err = -EPERM;
 756                         if (current->euid != shp->shm_perm.uid &&
 757                             current->euid != shp->shm_perm.cuid)
 758                                 goto out_unlock;
 759                         if (cmd == SHM_LOCK &&
 760                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 761                                 goto out_unlock;
 762                 }
 763
 764                 err = security_shm_shmctl(shp, cmd);
 765                 if (err)
 766                         goto out_unlock;
 767
 768                 if(cmd==SHM_LOCK) {
 769                         struct user_struct * user = current->user;
 770                         if (!is_file_hugepages(shp->shm_file)) {
 771                                 err = shmem_lock(shp->shm_file, 1, user);
 772                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 773                                         shp->shm_perm.mode |= SHM_LOCKED;
 774                                         shp->mlock_user = user;
 775                                 }
 776                         }
 777                 } else if (!is_file_hugepages(shp->shm_file)) {
 778                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 779                         shp->shm_perm.mode &= ~SHM_LOCKED;
 780                         shp->mlock_user = NULL;
 781                 }
 782                 shm_unlock(shp);
 783                 goto out;
 784         }
 785         case IPC_RMID:
 786         case IPC_SET:
 787                 err = shmctl_down(ns, shmid, cmd, buf, version);
 788                 return err;
 789         default:
 790                 return -EINVAL;
 791         }
 792
 793 out_unlock:
 794         shm_unlock(shp);
 795 out:
 796         return err;
 797 }
 798
 799 /*
 800  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 801  *
 802  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 803  * "raddr" thing points to kernel space, and there has to be a wrapper around
 804  * this.
 805  */
 806 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 807 {
 808         struct shmid_kernel *shp;
 809         unsigned long addr;
 810         unsigned long size;
 811         struct file * file;
 812         int    err;
 813         unsigned long flags;
 814         unsigned long prot;
 815         int acc_mode;
 816         unsigned long user_addr;
 817         struct ipc_namespace *ns;
 818         struct shm_file_data *sfd;
 819         struct path path;
 820         mode_t f_mode;
 821
 822         err = -EINVAL;
 823         if (shmid < 0)
 824                 goto out;
 825         else if ((addr = (ulong)shmaddr)) {
 826                 if (addr & (SHMLBA-1)) {
 827                         if (shmflg & SHM_RND)
 828                                 addr &= ~(SHMLBA-1);       /* round down */
 829                         else
 830 #ifndef __ARCH_FORCE_SHMLBA
 831                                 if (addr & ~PAGE_MASK)
 832 #endif
 833                                         goto out;
 834                 }
 835                 flags = MAP_SHARED | MAP_FIXED;
 836         } else {
 837                 if ((shmflg & SHM_REMAP))
 838                         goto out;
 839
 840                 flags = MAP_SHARED;
 841         }
 842
 843         if (shmflg & SHM_RDONLY) {
 844                 prot = PROT_READ;
 845                 acc_mode = S_IRUGO;
 846                 f_mode = FMODE_READ;
 847         } else {
 848                 prot = PROT_READ | PROT_WRITE;
 849                 acc_mode = S_IRUGO | S_IWUGO;
 850                 f_mode = FMODE_READ | FMODE_WRITE;
 851         }
 852         if (shmflg & SHM_EXEC) {
 853                 prot |= PROT_EXEC;
 854                 acc_mode |= S_IXUGO;
 855         }
 856
 857         /*
 858          * We cannot rely on the fs check since SYSV IPC does have an
 859          * additional creator id...
 860          */
 861         ns = current->nsproxy->ipc_ns;
 862         shp = shm_lock_check(ns, shmid);
 863         if (IS_ERR(shp)) {
 864                 err = PTR_ERR(shp);
 865                 goto out;
 866         }
 867
 868         err = -EACCES;
 869         if (ipcperms(&shp->shm_perm, acc_mode))
 870                 goto out_unlock;
 871
 872         err = security_shm_shmat(shp, shmaddr, shmflg);
 873         if (err)
 874                 goto out_unlock;
 875
 876         path.dentry = dget(shp->shm_file->f_path.dentry);
 877         path.mnt    = shp->shm_file->f_path.mnt;
 878         shp->shm_nattch++;
 879         size = i_size_read(path.dentry->d_inode);
 880         shm_unlock(shp);
 881
 882         err = -ENOMEM;
 883         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 884         if (!sfd)
 885                 goto out_put_dentry;
 886
 887         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 888         if (!file)
 889                 goto out_free;
 890
 891         file->private_data = sfd;
 892         file->f_mapping = shp->shm_file->f_mapping;
 893         sfd->id = shp->shm_perm.id;
 894         sfd->ns = get_ipc_ns(ns);
 895         sfd->file = shp->shm_file;
 896         sfd->vm_ops = NULL;
 897
 898         down_write(&current->mm->mmap_sem);
 899         if (addr && !(shmflg & SHM_REMAP)) {
 900                 err = -EINVAL;
 901                 if (find_vma_intersection(current->mm, addr, addr + size))
 902                         goto invalid;
 903                 /*
 904                  * If shm segment goes below stack, make sure there is some
 905                  * space left for the stack to grow (at least 4 pages).
 906                  */
 907                 if (addr < current->mm->start_stack &&
 908                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 909                         goto invalid;
 910         }
 911
 912         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 913         *raddr = user_addr;
 914         err = 0;
 915         if (IS_ERR_VALUE(user_addr))
 916                 err = (long)user_addr;
 917 invalid:
 918         up_write(&current->mm->mmap_sem);
 919
 920         fput(file);
 921
 922 out_nattch:
 923         down_write(&shm_ids(ns).rw_mutex);
 924         shp = shm_lock(ns, shmid);
 925         BUG_ON(IS_ERR(shp));
 926         shp->shm_nattch--;
 927         if(shp->shm_nattch == 0 &&
 928            shp->shm_perm.mode & SHM_DEST)
 929                 shm_destroy(ns, shp);
 930         else
 931                 shm_unlock(shp);
 932         up_write(&shm_ids(ns).rw_mutex);
 933
 934 out:
 935         return err;
 936
 937 out_unlock:
 938         shm_unlock(shp);
 939         goto out;
 940
 941 out_free:
 942         kfree(sfd);
 943 out_put_dentry:
 944         dput(path.dentry);
 945         goto out_nattch;
 946 }
 947
 948 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 949 {
 950         unsigned long ret;
 951         long err;
 952
 953         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 954         if (err)
 955                 return err;
 956         force_successful_syscall_return();
 957         return (long)ret;
 958 }
 959
 960 /*
 961  * detach and kill segment if marked destroyed.
 962  * The work is done in shm_close.
 963  */
 964 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 965 {
 966         struct mm_struct *mm = current->mm;
 967         struct vm_area_struct *vma, *next;
 968         unsigned long addr = (unsigned long)shmaddr;
 969         loff_t size = 0;
 970         int retval = -EINVAL;
 971
 972         if (addr & ~PAGE_MASK)
 973                 return retval;
 974
 975         down_write(&mm->mmap_sem);
 976
 977         /*
 978          * This function tries to be smart and unmap shm segments that
 979          * were modified by partial mlock or munmap calls:
 980          * - It first determines the size of the shm segment that should be
 981          *   unmapped: It searches for a vma that is backed by shm and that
 982          *   started at address shmaddr. It records it's size and then unmaps
 983          *   it.
 984          * - Then it unmaps all shm vmas that started at shmaddr and that
 985          *   are within the initially determined size.
 986          * Errors from do_munmap are ignored: the function only fails if
 987          * it's called with invalid parameters or if it's called to unmap
 988          * a part of a vma. Both calls in this function are for full vmas,
 989          * the parameters are directly copied from the vma itself and always
 990          * valid - therefore do_munmap cannot fail. (famous last words?)
 991          */
 992         /*
 993          * If it had been mremap()'d, the starting address would not
 994          * match the usual checks anyway. So assume all vma's are
 995          * above the starting address given.
 996          */
 997         vma = find_vma(mm, addr);
 998
 999         while (vma) {
1000                 next = vma->vm_next;
1001
1002                 /*
1003                  * Check if the starting address would match, i.e. it's
1004                  * a fragment created by mprotect() and/or munmap(), or it
1005                  * otherwise it starts at this address with no hassles.
1006                  */
1007                 if ((vma->vm_ops == &shm_vm_ops) &&
1008                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1009
1010
1011                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1012                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1013                         /*
1014                          * We discovered the size of the shm segment, so
1015                          * break out of here and fall through to the next
1016                          * loop that uses the size information to stop
1017                          * searching for matching vma's.
1018                          */
1019                         retval = 0;
1020                         vma = next;
1021                         break;
1022                 }
1023                 vma = next;
1024         }
1025
1026         /*
1027          * We need look no further than the maximum address a fragment
1028          * could possibly have landed at. Also cast things to loff_t to
1029          * prevent overflows and make comparisions vs. equal-width types.
1030          */
1031         size = PAGE_ALIGN(size);
1032         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1033                 next = vma->vm_next;
1034
1035                 /* finding a matching vma now does not alter retval */
1036                 if ((vma->vm_ops == &shm_vm_ops) &&
1037                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1038
1039                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1040                 vma = next;
1041         }
1042
1043         up_write(&mm->mmap_sem);
1044         return retval;
1045 }
1046
1047 #ifdef CONFIG_PROC_FS
1048 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1049 {
1050         struct shmid_kernel *shp = it;
1051
1052 #if BITS_PER_LONG <= 32
1053 #define SIZE_SPEC "%10lu"
1054 #else
1055 #define SIZE_SPEC "%21lu"
1056 #endif
1057
1058         return seq_printf(s,
1059                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1060                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1061                           shp->shm_perm.key,
1062                           shp->shm_perm.id,
1063                           shp->shm_perm.mode,
1064                           shp->shm_segsz,
1065                           shp->shm_cprid,
1066                           shp->shm_lprid,
1067                           shp->shm_nattch,
1068                           shp->shm_perm.uid,
1069                           shp->shm_perm.gid,
1070                           shp->shm_perm.cuid,
1071                           shp->shm_perm.cgid,
1072                           shp->shm_atim,
1073                           shp->shm_dtim,
1074                           shp->shm_ctim);
1075 }
1076 #endif