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
  43 #include <asm/uaccess.h>
  44
  45 #include "util.h"
  46
  47 struct shm_file_data {
  48         int id;
  49         struct ipc_namespace *ns;
  50         struct file *file;
  51         const struct vm_operations_struct *vm_ops;
  52 };
  53
  54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  55
  56 static const struct file_operations shm_file_operations;
  57 static const struct vm_operations_struct shm_vm_ops;
  58
  59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  60
  61 #define shm_unlock(shp)                 \
  62         ipc_unlock(&(shp)->shm_perm)
  63
  64 static int newseg(struct ipc_namespace *, struct ipc_params *);
  65 static void shm_open(struct vm_area_struct *vma);
  66 static void shm_close(struct vm_area_struct *vma);
  67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  68 #ifdef CONFIG_PROC_FS
  69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  70 #endif
  71
  72 void shm_init_ns(struct ipc_namespace *ns)
  73 {
  74         ns->shm_ctlmax = SHMMAX;
  75         ns->shm_ctlall = SHMALL;
  76         ns->shm_ctlmni = SHMMNI;
  77         ns->shm_tot = 0;
  78         ipc_init_ids(&shm_ids(ns));
  79 }
  80
  81 /*
  82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  83  * Only shm_ids.rw_mutex remains locked on exit.
  84  */
  85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  86 {
  87         struct shmid_kernel *shp;
  88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  89
  90         if (shp->shm_nattch){
  91                 shp->shm_perm.mode |= SHM_DEST;
  92                 /* Do not find it any more */
  93                 shp->shm_perm.key = IPC_PRIVATE;
  94                 shm_unlock(shp);
  95         } else
  96                 shm_destroy(ns, shp);
  97 }
  98
  99 #ifdef CONFIG_IPC_NS
 100 void shm_exit_ns(struct ipc_namespace *ns)
 101 {
 102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 103         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 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 if (shp->mlock_user)
 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, int datasync)
 277 {
 278         struct shm_file_data *sfd = shm_file_data(file);
 279
 280         if (!sfd->file->f_op->fsync)
 281                 return -EINVAL;
 282         return sfd->file->f_op->fsync(sfd->file, datasync);
 283 }
 284
 285 static unsigned long shm_get_unmapped_area(struct file *file,
 286         unsigned long addr, unsigned long len, unsigned long pgoff,
 287         unsigned long flags)
 288 {
 289         struct shm_file_data *sfd = shm_file_data(file);
 290         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 291                                                 pgoff, flags);
 292 }
 293
 294 static const struct file_operations shm_file_operations = {
 295         .mmap           = shm_mmap,
 296         .fsync          = shm_fsync,
 297         .release        = shm_release,
 298 #ifndef CONFIG_MMU
 299         .get_unmapped_area      = shm_get_unmapped_area,
 300 #endif
 301 };
 302
 303 static const struct file_operations shm_file_operations_huge = {
 304         .mmap           = shm_mmap,
 305         .fsync          = shm_fsync,
 306         .release        = shm_release,
 307         .get_unmapped_area      = shm_get_unmapped_area,
 308 };
 309
 310 int is_file_shm_hugepages(struct file *file)
 311 {
 312         return file->f_op == &shm_file_operations_huge;
 313 }
 314
 315 static const 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         int acctflag = 0;
 345
 346         if (size < SHMMIN || size > ns->shm_ctlmax)
 347                 return -EINVAL;
 348
 349         if (ns->shm_tot + numpages > ns->shm_ctlall)
 350                 return -ENOSPC;
 351
 352         shp = ipc_rcu_alloc(sizeof(*shp));
 353         if (!shp)
 354                 return -ENOMEM;
 355
 356         shp->shm_perm.key = key;
 357         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 358         shp->mlock_user = NULL;
 359
 360         shp->shm_perm.security = NULL;
 361         error = security_shm_alloc(shp);
 362         if (error) {
 363                 ipc_rcu_putref(shp);
 364                 return error;
 365         }
 366
 367         sprintf (name, "SYSV%08x", key);
 368         if (shmflg & SHM_HUGETLB) {
 369                 /* hugetlb_file_setup applies strict accounting */
 370                 if (shmflg & SHM_NORESERVE)
 371                         acctflag = VM_NORESERVE;
 372                 file = hugetlb_file_setup(name, size, acctflag,
 373                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
 374         } else {
 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 = VM_NORESERVE;
 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         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 389         if (id < 0) {
 390                 error = id;
 391                 goto no_id;
 392         }
 393
 394         shp->shm_cprid = task_tgid_vnr(current);
 395         shp->shm_lprid = 0;
 396         shp->shm_atim = shp->shm_dtim = 0;
 397         shp->shm_ctim = get_seconds();
 398         shp->shm_segsz = size;
 399         shp->shm_nattch = 0;
 400         shp->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->shm_perm.id;
 406
 407         ns->shm_tot += numpages;
 408         error = shp->shm_perm.id;
 409         shm_unlock(shp);
 410         return error;
 411
 412 no_id:
 413         if (is_file_hugepages(file) && shp->mlock_user)
 414                 user_shm_unlock(size, shp->mlock_user);
 415         fput(file);
 416 no_file:
 417         security_shm_free(shp);
 418         ipc_rcu_putref(shp);
 419         return error;
 420 }
 421
 422 /*
 423  * Called with shm_ids.rw_mutex and ipcp locked.
 424  */
 425 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 426 {
 427         struct shmid_kernel *shp;
 428
 429         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 430         return security_shm_associate(shp, shmflg);
 431 }
 432
 433 /*
 434  * Called with shm_ids.rw_mutex and ipcp locked.
 435  */
 436 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 437                                 struct ipc_params *params)
 438 {
 439         struct shmid_kernel *shp;
 440
 441         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 442         if (shp->shm_segsz < params->u.size)
 443                 return -EINVAL;
 444
 445         return 0;
 446 }
 447
 448 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 449 {
 450         struct ipc_namespace *ns;
 451         struct ipc_ops shm_ops;
 452         struct ipc_params shm_params;
 453
 454         ns = current->nsproxy->ipc_ns;
 455
 456         shm_ops.getnew = newseg;
 457         shm_ops.associate = shm_security;
 458         shm_ops.more_checks = shm_more_checks;
 459
 460         shm_params.key = key;
 461         shm_params.flg = shmflg;
 462         shm_params.u.size = size;
 463
 464         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 465 }
 466
 467 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 468 {
 469         switch(version) {
 470         case IPC_64:
 471                 return copy_to_user(buf, in, sizeof(*in));
 472         case IPC_OLD:
 473             {
 474                 struct shmid_ds out;
 475
 476                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 477                 out.shm_segsz   = in->shm_segsz;
 478                 out.shm_atime   = in->shm_atime;
 479                 out.shm_dtime   = in->shm_dtime;
 480                 out.shm_ctime   = in->shm_ctime;
 481                 out.shm_cpid    = in->shm_cpid;
 482                 out.shm_lpid    = in->shm_lpid;
 483                 out.shm_nattch  = in->shm_nattch;
 484
 485                 return copy_to_user(buf, &out, sizeof(out));
 486             }
 487         default:
 488                 return -EINVAL;
 489         }
 490 }
 491
 492 static inline unsigned long
 493 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 494 {
 495         switch(version) {
 496         case IPC_64:
 497                 if (copy_from_user(out, buf, sizeof(*out)))
 498                         return -EFAULT;
 499                 return 0;
 500         case IPC_OLD:
 501             {
 502                 struct shmid_ds tbuf_old;
 503
 504                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 505                         return -EFAULT;
 506
 507                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 508                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 509                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 510
 511                 return 0;
 512             }
 513         default:
 514                 return -EINVAL;
 515         }
 516 }
 517
 518 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 519 {
 520         switch(version) {
 521         case IPC_64:
 522                 return copy_to_user(buf, in, sizeof(*in));
 523         case IPC_OLD:
 524             {
 525                 struct shminfo out;
 526
 527                 if(in->shmmax > INT_MAX)
 528                         out.shmmax = INT_MAX;
 529                 else
 530                         out.shmmax = (int)in->shmmax;
 531
 532                 out.shmmin      = in->shmmin;
 533                 out.shmmni      = in->shmmni;
 534                 out.shmseg      = in->shmseg;
 535                 out.shmall      = in->shmall;
 536
 537                 return copy_to_user(buf, &out, sizeof(out));
 538             }
 539         default:
 540                 return -EINVAL;
 541         }
 542 }
 543
 544 /*
 545  * Called with shm_ids.rw_mutex held as a reader
 546  */
 547 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 548                 unsigned long *swp)
 549 {
 550         int next_id;
 551         int total, in_use;
 552
 553         *rss = 0;
 554         *swp = 0;
 555
 556         in_use = shm_ids(ns).in_use;
 557
 558         for (total = 0, next_id = 0; total < in_use; next_id++) {
 559                 struct kern_ipc_perm *ipc;
 560                 struct shmid_kernel *shp;
 561                 struct inode *inode;
 562
 563                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 564                 if (ipc == NULL)
 565                         continue;
 566                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 567
 568                 inode = shp->shm_file->f_path.dentry->d_inode;
 569
 570                 if (is_file_hugepages(shp->shm_file)) {
 571                         struct address_space *mapping = inode->i_mapping;
 572                         struct hstate *h = hstate_file(shp->shm_file);
 573                         *rss += pages_per_huge_page(h) * mapping->nrpages;
 574                 } else {
 575 #ifdef CONFIG_SHMEM
 576                         struct shmem_inode_info *info = SHMEM_I(inode);
 577                         spin_lock(&info->lock);
 578                         *rss += inode->i_mapping->nrpages;
 579                         *swp += info->swapped;
 580                         spin_unlock(&info->lock);
 581 #else
 582                         *rss += inode->i_mapping->nrpages;
 583 #endif
 584                 }
 585
 586                 total++;
 587         }
 588 }
 589
 590 /*
 591  * This function handles some shmctl commands which require the rw_mutex
 592  * to be held in write mode.
 593  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 594  */
 595 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 596                        struct shmid_ds __user *buf, int version)
 597 {
 598         struct kern_ipc_perm *ipcp;
 599         struct shmid64_ds shmid64;
 600         struct shmid_kernel *shp;
 601         int err;
 602
 603         if (cmd == IPC_SET) {
 604                 if (copy_shmid_from_user(&shmid64, buf, version))
 605                         return -EFAULT;
 606         }
 607
 608         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 609         if (IS_ERR(ipcp))
 610                 return PTR_ERR(ipcp);
 611
 612         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 613
 614         err = security_shm_shmctl(shp, cmd);
 615         if (err)
 616                 goto out_unlock;
 617         switch (cmd) {
 618         case IPC_RMID:
 619                 do_shm_rmid(ns, ipcp);
 620                 goto out_up;
 621         case IPC_SET:
 622                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 623                 shp->shm_ctim = get_seconds();
 624                 break;
 625         default:
 626                 err = -EINVAL;
 627         }
 628 out_unlock:
 629         shm_unlock(shp);
 630 out_up:
 631         up_write(&shm_ids(ns).rw_mutex);
 632         return err;
 633 }
 634
 635 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 636 {
 637         struct shmid_kernel *shp;
 638         int err, version;
 639         struct ipc_namespace *ns;
 640
 641         if (cmd < 0 || shmid < 0) {
 642                 err = -EINVAL;
 643                 goto out;
 644         }
 645
 646         version = ipc_parse_version(&cmd);
 647         ns = current->nsproxy->ipc_ns;
 648
 649         switch (cmd) { /* replace with proc interface ? */
 650         case IPC_INFO:
 651         {
 652                 struct shminfo64 shminfo;
 653
 654                 err = security_shm_shmctl(NULL, cmd);
 655                 if (err)
 656                         return err;
 657
 658                 memset(&shminfo, 0, sizeof(shminfo));
 659                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 660                 shminfo.shmmax = ns->shm_ctlmax;
 661                 shminfo.shmall = ns->shm_ctlall;
 662
 663                 shminfo.shmmin = SHMMIN;
 664                 if(copy_shminfo_to_user (buf, &shminfo, version))
 665                         return -EFAULT;
 666
 667                 down_read(&shm_ids(ns).rw_mutex);
 668                 err = ipc_get_maxid(&shm_ids(ns));
 669                 up_read(&shm_ids(ns).rw_mutex);
 670
 671                 if(err<0)
 672                         err = 0;
 673                 goto out;
 674         }
 675         case SHM_INFO:
 676         {
 677                 struct shm_info shm_info;
 678
 679                 err = security_shm_shmctl(NULL, cmd);
 680                 if (err)
 681                         return err;
 682
 683                 memset(&shm_info, 0, sizeof(shm_info));
 684                 down_read(&shm_ids(ns).rw_mutex);
 685                 shm_info.used_ids = shm_ids(ns).in_use;
 686                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 687                 shm_info.shm_tot = ns->shm_tot;
 688                 shm_info.swap_attempts = 0;
 689                 shm_info.swap_successes = 0;
 690                 err = ipc_get_maxid(&shm_ids(ns));
 691                 up_read(&shm_ids(ns).rw_mutex);
 692                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 693                         err = -EFAULT;
 694                         goto out;
 695                 }
 696
 697                 err = err < 0 ? 0 : err;
 698                 goto out;
 699         }
 700         case SHM_STAT:
 701         case IPC_STAT:
 702         {
 703                 struct shmid64_ds tbuf;
 704                 int result;
 705
 706                 if (cmd == SHM_STAT) {
 707                         shp = shm_lock(ns, shmid);
 708                         if (IS_ERR(shp)) {
 709                                 err = PTR_ERR(shp);
 710                                 goto out;
 711                         }
 712                         result = shp->shm_perm.id;
 713                 } else {
 714                         shp = shm_lock_check(ns, shmid);
 715                         if (IS_ERR(shp)) {
 716                                 err = PTR_ERR(shp);
 717                                 goto out;
 718                         }
 719                         result = 0;
 720                 }
 721                 err = -EACCES;
 722                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 723                         goto out_unlock;
 724                 err = security_shm_shmctl(shp, cmd);
 725                 if (err)
 726                         goto out_unlock;
 727                 memset(&tbuf, 0, sizeof(tbuf));
 728                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 729                 tbuf.shm_segsz  = shp->shm_segsz;
 730                 tbuf.shm_atime  = shp->shm_atim;
 731                 tbuf.shm_dtime  = shp->shm_dtim;
 732                 tbuf.shm_ctime  = shp->shm_ctim;
 733                 tbuf.shm_cpid   = shp->shm_cprid;
 734                 tbuf.shm_lpid   = shp->shm_lprid;
 735                 tbuf.shm_nattch = shp->shm_nattch;
 736                 shm_unlock(shp);
 737                 if(copy_shmid_to_user (buf, &tbuf, version))
 738                         err = -EFAULT;
 739                 else
 740                         err = result;
 741                 goto out;
 742         }
 743         case SHM_LOCK:
 744         case SHM_UNLOCK:
 745         {
 746                 struct file *uninitialized_var(shm_file);
 747
 748                 lru_add_drain_all();  /* drain pagevecs to lru lists */
 749
 750                 shp = shm_lock_check(ns, shmid);
 751                 if (IS_ERR(shp)) {
 752                         err = PTR_ERR(shp);
 753                         goto out;
 754                 }
 755
 756                 audit_ipc_obj(&(shp->shm_perm));
 757
 758                 if (!capable(CAP_IPC_LOCK)) {
 759                         uid_t euid = current_euid();
 760                         err = -EPERM;
 761                         if (euid != shp->shm_perm.uid &&
 762                             euid != shp->shm_perm.cuid)
 763                                 goto out_unlock;
 764                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
 765                                 goto out_unlock;
 766                 }
 767
 768                 err = security_shm_shmctl(shp, cmd);
 769                 if (err)
 770                         goto out_unlock;
 771
 772                 if(cmd==SHM_LOCK) {
 773                         struct user_struct *user = current_user();
 774                         if (!is_file_hugepages(shp->shm_file)) {
 775                                 err = shmem_lock(shp->shm_file, 1, user);
 776                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 777                                         shp->shm_perm.mode |= SHM_LOCKED;
 778                                         shp->mlock_user = user;
 779                                 }
 780                         }
 781                 } else if (!is_file_hugepages(shp->shm_file)) {
 782                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 783                         shp->shm_perm.mode &= ~SHM_LOCKED;
 784                         shp->mlock_user = NULL;
 785                 }
 786                 shm_unlock(shp);
 787                 goto out;
 788         }
 789         case IPC_RMID:
 790         case IPC_SET:
 791                 err = shmctl_down(ns, shmid, cmd, buf, version);
 792                 return err;
 793         default:
 794                 return -EINVAL;
 795         }
 796
 797 out_unlock:
 798         shm_unlock(shp);
 799 out:
 800         return err;
 801 }
 802
 803 /*
 804  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 805  *
 806  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 807  * "raddr" thing points to kernel space, and there has to be a wrapper around
 808  * this.
 809  */
 810 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 811 {
 812         struct shmid_kernel *shp;
 813         unsigned long addr;
 814         unsigned long size;
 815         struct file * file;
 816         int    err;
 817         unsigned long flags;
 818         unsigned long prot;
 819         int acc_mode;
 820         unsigned long user_addr;
 821         struct ipc_namespace *ns;
 822         struct shm_file_data *sfd;
 823         struct path path;
 824         fmode_t f_mode;
 825
 826         err = -EINVAL;
 827         if (shmid < 0)
 828                 goto out;
 829         else if ((addr = (ulong)shmaddr)) {
 830                 if (addr & (SHMLBA-1)) {
 831                         if (shmflg & SHM_RND)
 832                                 addr &= ~(SHMLBA-1);       /* round down */
 833                         else
 834 #ifndef __ARCH_FORCE_SHMLBA
 835                                 if (addr & ~PAGE_MASK)
 836 #endif
 837                                         goto out;
 838                 }
 839                 flags = MAP_SHARED | MAP_FIXED;
 840         } else {
 841                 if ((shmflg & SHM_REMAP))
 842                         goto out;
 843
 844                 flags = MAP_SHARED;
 845         }
 846
 847         if (shmflg & SHM_RDONLY) {
 848                 prot = PROT_READ;
 849                 acc_mode = S_IRUGO;
 850                 f_mode = FMODE_READ;
 851         } else {
 852                 prot = PROT_READ | PROT_WRITE;
 853                 acc_mode = S_IRUGO | S_IWUGO;
 854                 f_mode = FMODE_READ | FMODE_WRITE;
 855         }
 856         if (shmflg & SHM_EXEC) {
 857                 prot |= PROT_EXEC;
 858                 acc_mode |= S_IXUGO;
 859         }
 860
 861         /*
 862          * We cannot rely on the fs check since SYSV IPC does have an
 863          * additional creator id...
 864          */
 865         ns = current->nsproxy->ipc_ns;
 866         shp = shm_lock_check(ns, shmid);
 867         if (IS_ERR(shp)) {
 868                 err = PTR_ERR(shp);
 869                 goto out;
 870         }
 871
 872         err = -EACCES;
 873         if (ipcperms(&shp->shm_perm, acc_mode))
 874                 goto out_unlock;
 875
 876         err = security_shm_shmat(shp, shmaddr, shmflg);
 877         if (err)
 878                 goto out_unlock;
 879
 880         path = shp->shm_file->f_path;
 881         path_get(&path);
 882         shp->shm_nattch++;
 883         size = i_size_read(path.dentry->d_inode);
 884         shm_unlock(shp);
 885
 886         err = -ENOMEM;
 887         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 888         if (!sfd)
 889                 goto out_put_dentry;
 890
 891         file = alloc_file(&path, f_mode,
 892                           is_file_hugepages(shp->shm_file) ?
 893                                 &shm_file_operations_huge :
 894                                 &shm_file_operations);
 895         if (!file)
 896                 goto out_free;
 897
 898         file->private_data = sfd;
 899         file->f_mapping = shp->shm_file->f_mapping;
 900         sfd->id = shp->shm_perm.id;
 901         sfd->ns = get_ipc_ns(ns);
 902         sfd->file = shp->shm_file;
 903         sfd->vm_ops = NULL;
 904
 905         down_write(&current->mm->mmap_sem);
 906         if (addr && !(shmflg & SHM_REMAP)) {
 907                 err = -EINVAL;
 908                 if (find_vma_intersection(current->mm, addr, addr + size))
 909                         goto invalid;
 910                 /*
 911                  * If shm segment goes below stack, make sure there is some
 912                  * space left for the stack to grow (at least 4 pages).
 913                  */
 914                 if (addr < current->mm->start_stack &&
 915                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 916                         goto invalid;
 917         }
 918
 919         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 920         *raddr = user_addr;
 921         err = 0;
 922         if (IS_ERR_VALUE(user_addr))
 923                 err = (long)user_addr;
 924 invalid:
 925         up_write(&current->mm->mmap_sem);
 926
 927         fput(file);
 928
 929 out_nattch:
 930         down_write(&shm_ids(ns).rw_mutex);
 931         shp = shm_lock(ns, shmid);
 932         BUG_ON(IS_ERR(shp));
 933         shp->shm_nattch--;
 934         if(shp->shm_nattch == 0 &&
 935            shp->shm_perm.mode & SHM_DEST)
 936                 shm_destroy(ns, shp);
 937         else
 938                 shm_unlock(shp);
 939         up_write(&shm_ids(ns).rw_mutex);
 940
 941 out:
 942         return err;
 943
 944 out_unlock:
 945         shm_unlock(shp);
 946         goto out;
 947
 948 out_free:
 949         kfree(sfd);
 950 out_put_dentry:
 951         path_put(&path);
 952         goto out_nattch;
 953 }
 954
 955 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 956 {
 957         unsigned long ret;
 958         long err;
 959
 960         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 961         if (err)
 962                 return err;
 963         force_successful_syscall_return();
 964         return (long)ret;
 965 }
 966
 967 /*
 968  * detach and kill segment if marked destroyed.
 969  * The work is done in shm_close.
 970  */
 971 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 972 {
 973         struct mm_struct *mm = current->mm;
 974         struct vm_area_struct *vma;
 975         unsigned long addr = (unsigned long)shmaddr;
 976         int retval = -EINVAL;
 977 #ifdef CONFIG_MMU
 978         loff_t size = 0;
 979         struct vm_area_struct *next;
 980 #endif
 981
 982         if (addr & ~PAGE_MASK)
 983                 return retval;
 984
 985         down_write(&mm->mmap_sem);
 986
 987         /*
 988          * This function tries to be smart and unmap shm segments that
 989          * were modified by partial mlock or munmap calls:
 990          * - It first determines the size of the shm segment that should be
 991          *   unmapped: It searches for a vma that is backed by shm and that
 992          *   started at address shmaddr. It records it's size and then unmaps
 993          *   it.
 994          * - Then it unmaps all shm vmas that started at shmaddr and that
 995          *   are within the initially determined size.
 996          * Errors from do_munmap are ignored: the function only fails if
 997          * it's called with invalid parameters or if it's called to unmap
 998          * a part of a vma. Both calls in this function are for full vmas,
 999          * the parameters are directly copied from the vma itself and always
1000          * valid - therefore do_munmap cannot fail. (famous last words?)
1001          */
1002         /*
1003          * If it had been mremap()'d, the starting address would not
1004          * match the usual checks anyway. So assume all vma's are
1005          * above the starting address given.
1006          */
1007         vma = find_vma(mm, addr);
1008
1009 #ifdef CONFIG_MMU
1010         while (vma) {
1011                 next = vma->vm_next;
1012
1013                 /*
1014                  * Check if the starting address would match, i.e. it's
1015                  * a fragment created by mprotect() and/or munmap(), or it
1016                  * otherwise it starts at this address with no hassles.
1017                  */
1018                 if ((vma->vm_ops == &shm_vm_ops) &&
1019                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1020
1021
1022                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1023                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1024                         /*
1025                          * We discovered the size of the shm segment, so
1026                          * break out of here and fall through to the next
1027                          * loop that uses the size information to stop
1028                          * searching for matching vma's.
1029                          */
1030                         retval = 0;
1031                         vma = next;
1032                         break;
1033                 }
1034                 vma = next;
1035         }
1036
1037         /*
1038          * We need look no further than the maximum address a fragment
1039          * could possibly have landed at. Also cast things to loff_t to
1040          * prevent overflows and make comparisions vs. equal-width types.
1041          */
1042         size = PAGE_ALIGN(size);
1043         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1044                 next = vma->vm_next;
1045
1046                 /* finding a matching vma now does not alter retval */
1047                 if ((vma->vm_ops == &shm_vm_ops) &&
1048                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1049
1050                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051                 vma = next;
1052         }
1053
1054 #else /* CONFIG_MMU */
1055         /* under NOMMU conditions, the exact address to be destroyed must be
1056          * given */
1057         retval = -EINVAL;
1058         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1059                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1060                 retval = 0;
1061         }
1062
1063 #endif
1064
1065         up_write(&mm->mmap_sem);
1066         return retval;
1067 }
1068
1069 #ifdef CONFIG_PROC_FS
1070 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1071 {
1072         struct shmid_kernel *shp = it;
1073
1074 #if BITS_PER_LONG <= 32
1075 #define SIZE_SPEC "%10lu"
1076 #else
1077 #define SIZE_SPEC "%21lu"
1078 #endif
1079
1080         return seq_printf(s,
1081                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1082                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1083                           shp->shm_perm.key,
1084                           shp->shm_perm.id,
1085                           shp->shm_perm.mode,
1086                           shp->shm_segsz,
1087                           shp->shm_cprid,
1088                           shp->shm_lprid,
1089                           shp->shm_nattch,
1090                           shp->shm_perm.uid,
1091                           shp->shm_perm.gid,
1092                           shp->shm_perm.cuid,
1093                           shp->shm_perm.cgid,
1094                           shp->shm_atim,
1095                           shp->shm_dtim,
1096                           shp->shm_ctim);
1097 }
1098 #endif