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 <linux/ima.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(&shm_ids(ns));
  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         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 = current_user();
 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         ima_shm_check(file);
 388
 389         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 390         if (id < 0) {
 391                 error = id;
 392                 goto no_id;
 393         }
 394
 395         shp->shm_cprid = task_tgid_vnr(current);
 396         shp->shm_lprid = 0;
 397         shp->shm_atim = shp->shm_dtim = 0;
 398         shp->shm_ctim = get_seconds();
 399         shp->shm_segsz = size;
 400         shp->shm_nattch = 0;
 401         shp->shm_file = file;
 402         /*
 403          * shmid gets reported as "inode#" in /proc/pid/maps.
 404          * proc-ps tools use this. Changing this will break them.
 405          */
 406         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 407
 408         ns->shm_tot += numpages;
 409         error = shp->shm_perm.id;
 410         shm_unlock(shp);
 411         return error;
 412
 413 no_id:
 414         fput(file);
 415 no_file:
 416         security_shm_free(shp);
 417         ipc_rcu_putref(shp);
 418         return error;
 419 }
 420
 421 /*
 422  * Called with shm_ids.rw_mutex and ipcp locked.
 423  */
 424 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 425 {
 426         struct shmid_kernel *shp;
 427
 428         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 429         return security_shm_associate(shp, shmflg);
 430 }
 431
 432 /*
 433  * Called with shm_ids.rw_mutex and ipcp locked.
 434  */
 435 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 436                                 struct ipc_params *params)
 437 {
 438         struct shmid_kernel *shp;
 439
 440         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 441         if (shp->shm_segsz < params->u.size)
 442                 return -EINVAL;
 443
 444         return 0;
 445 }
 446
 447 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 448 {
 449         struct ipc_namespace *ns;
 450         struct ipc_ops shm_ops;
 451         struct ipc_params shm_params;
 452
 453         ns = current->nsproxy->ipc_ns;
 454
 455         shm_ops.getnew = newseg;
 456         shm_ops.associate = shm_security;
 457         shm_ops.more_checks = shm_more_checks;
 458
 459         shm_params.key = key;
 460         shm_params.flg = shmflg;
 461         shm_params.u.size = size;
 462
 463         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 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 #ifdef CONFIG_SHMEM
 573                         struct shmem_inode_info *info = SHMEM_I(inode);
 574                         spin_lock(&info->lock);
 575                         *rss += inode->i_mapping->nrpages;
 576                         *swp += info->swapped;
 577                         spin_unlock(&info->lock);
 578 #else
 579                         *rss += inode->i_mapping->nrpages;
 580 #endif
 581                 }
 582
 583                 total++;
 584         }
 585 }
 586
 587 /*
 588  * This function handles some shmctl commands which require the rw_mutex
 589  * to be held in write mode.
 590  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 591  */
 592 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 593                        struct shmid_ds __user *buf, int version)
 594 {
 595         struct kern_ipc_perm *ipcp;
 596         struct shmid64_ds shmid64;
 597         struct shmid_kernel *shp;
 598         int err;
 599
 600         if (cmd == IPC_SET) {
 601                 if (copy_shmid_from_user(&shmid64, buf, version))
 602                         return -EFAULT;
 603         }
 604
 605         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 606         if (IS_ERR(ipcp))
 607                 return PTR_ERR(ipcp);
 608
 609         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 610
 611         err = security_shm_shmctl(shp, cmd);
 612         if (err)
 613                 goto out_unlock;
 614         switch (cmd) {
 615         case IPC_RMID:
 616                 do_shm_rmid(ns, ipcp);
 617                 goto out_up;
 618         case IPC_SET:
 619                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 620                 shp->shm_ctim = get_seconds();
 621                 break;
 622         default:
 623                 err = -EINVAL;
 624         }
 625 out_unlock:
 626         shm_unlock(shp);
 627 out_up:
 628         up_write(&shm_ids(ns).rw_mutex);
 629         return err;
 630 }
 631
 632 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 633 {
 634         struct shmid_kernel *shp;
 635         int err, version;
 636         struct ipc_namespace *ns;
 637
 638         if (cmd < 0 || shmid < 0) {
 639                 err = -EINVAL;
 640                 goto out;
 641         }
 642
 643         version = ipc_parse_version(&cmd);
 644         ns = current->nsproxy->ipc_ns;
 645
 646         switch (cmd) { /* replace with proc interface ? */
 647         case IPC_INFO:
 648         {
 649                 struct shminfo64 shminfo;
 650
 651                 err = security_shm_shmctl(NULL, cmd);
 652                 if (err)
 653                         return err;
 654
 655                 memset(&shminfo, 0, sizeof(shminfo));
 656                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 657                 shminfo.shmmax = ns->shm_ctlmax;
 658                 shminfo.shmall = ns->shm_ctlall;
 659
 660                 shminfo.shmmin = SHMMIN;
 661                 if(copy_shminfo_to_user (buf, &shminfo, version))
 662                         return -EFAULT;
 663
 664                 down_read(&shm_ids(ns).rw_mutex);
 665                 err = ipc_get_maxid(&shm_ids(ns));
 666                 up_read(&shm_ids(ns).rw_mutex);
 667
 668                 if(err<0)
 669                         err = 0;
 670                 goto out;
 671         }
 672         case SHM_INFO:
 673         {
 674                 struct shm_info shm_info;
 675
 676                 err = security_shm_shmctl(NULL, cmd);
 677                 if (err)
 678                         return err;
 679
 680                 memset(&shm_info, 0, sizeof(shm_info));
 681                 down_read(&shm_ids(ns).rw_mutex);
 682                 shm_info.used_ids = shm_ids(ns).in_use;
 683                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 684                 shm_info.shm_tot = ns->shm_tot;
 685                 shm_info.swap_attempts = 0;
 686                 shm_info.swap_successes = 0;
 687                 err = ipc_get_maxid(&shm_ids(ns));
 688                 up_read(&shm_ids(ns).rw_mutex);
 689                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 690                         err = -EFAULT;
 691                         goto out;
 692                 }
 693
 694                 err = err < 0 ? 0 : err;
 695                 goto out;
 696         }
 697         case SHM_STAT:
 698         case IPC_STAT:
 699         {
 700                 struct shmid64_ds tbuf;
 701                 int result;
 702
 703                 if (cmd == SHM_STAT) {
 704                         shp = shm_lock(ns, shmid);
 705                         if (IS_ERR(shp)) {
 706                                 err = PTR_ERR(shp);
 707                                 goto out;
 708                         }
 709                         result = shp->shm_perm.id;
 710                 } else {
 711                         shp = shm_lock_check(ns, shmid);
 712                         if (IS_ERR(shp)) {
 713                                 err = PTR_ERR(shp);
 714                                 goto out;
 715                         }
 716                         result = 0;
 717                 }
 718                 err = -EACCES;
 719                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 720                         goto out_unlock;
 721                 err = security_shm_shmctl(shp, cmd);
 722                 if (err)
 723                         goto out_unlock;
 724                 memset(&tbuf, 0, sizeof(tbuf));
 725                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 726                 tbuf.shm_segsz  = shp->shm_segsz;
 727                 tbuf.shm_atime  = shp->shm_atim;
 728                 tbuf.shm_dtime  = shp->shm_dtim;
 729                 tbuf.shm_ctime  = shp->shm_ctim;
 730                 tbuf.shm_cpid   = shp->shm_cprid;
 731                 tbuf.shm_lpid   = shp->shm_lprid;
 732                 tbuf.shm_nattch = shp->shm_nattch;
 733                 shm_unlock(shp);
 734                 if(copy_shmid_to_user (buf, &tbuf, version))
 735                         err = -EFAULT;
 736                 else
 737                         err = result;
 738                 goto out;
 739         }
 740         case SHM_LOCK:
 741         case SHM_UNLOCK:
 742         {
 743                 struct file *uninitialized_var(shm_file);
 744
 745                 lru_add_drain_all();  /* drain pagevecs to lru lists */
 746
 747                 shp = shm_lock_check(ns, shmid);
 748                 if (IS_ERR(shp)) {
 749                         err = PTR_ERR(shp);
 750                         goto out;
 751                 }
 752
 753                 audit_ipc_obj(&(shp->shm_perm));
 754
 755                 if (!capable(CAP_IPC_LOCK)) {
 756                         uid_t euid = current_euid();
 757                         err = -EPERM;
 758                         if (euid != shp->shm_perm.uid &&
 759                             euid != shp->shm_perm.cuid)
 760                                 goto out_unlock;
 761                         if (cmd == SHM_LOCK &&
 762                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 763                                 goto out_unlock;
 764                 }
 765
 766                 err = security_shm_shmctl(shp, cmd);
 767                 if (err)
 768                         goto out_unlock;
 769
 770                 if(cmd==SHM_LOCK) {
 771                         struct user_struct *user = current_user();
 772                         if (!is_file_hugepages(shp->shm_file)) {
 773                                 err = shmem_lock(shp->shm_file, 1, user);
 774                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 775                                         shp->shm_perm.mode |= SHM_LOCKED;
 776                                         shp->mlock_user = user;
 777                                 }
 778                         }
 779                 } else if (!is_file_hugepages(shp->shm_file)) {
 780                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 781                         shp->shm_perm.mode &= ~SHM_LOCKED;
 782                         shp->mlock_user = NULL;
 783                 }
 784                 shm_unlock(shp);
 785                 goto out;
 786         }
 787         case IPC_RMID:
 788         case IPC_SET:
 789                 err = shmctl_down(ns, shmid, cmd, buf, version);
 790                 return err;
 791         default:
 792                 return -EINVAL;
 793         }
 794
 795 out_unlock:
 796         shm_unlock(shp);
 797 out:
 798         return err;
 799 }
 800
 801 /*
 802  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 803  *
 804  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 805  * "raddr" thing points to kernel space, and there has to be a wrapper around
 806  * this.
 807  */
 808 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 809 {
 810         struct shmid_kernel *shp;
 811         unsigned long addr;
 812         unsigned long size;
 813         struct file * file;
 814         int    err;
 815         unsigned long flags;
 816         unsigned long prot;
 817         int acc_mode;
 818         unsigned long user_addr;
 819         struct ipc_namespace *ns;
 820         struct shm_file_data *sfd;
 821         struct path path;
 822         fmode_t f_mode;
 823
 824         err = -EINVAL;
 825         if (shmid < 0)
 826                 goto out;
 827         else if ((addr = (ulong)shmaddr)) {
 828                 if (addr & (SHMLBA-1)) {
 829                         if (shmflg & SHM_RND)
 830                                 addr &= ~(SHMLBA-1);       /* round down */
 831                         else
 832 #ifndef __ARCH_FORCE_SHMLBA
 833                                 if (addr & ~PAGE_MASK)
 834 #endif
 835                                         goto out;
 836                 }
 837                 flags = MAP_SHARED | MAP_FIXED;
 838         } else {
 839                 if ((shmflg & SHM_REMAP))
 840                         goto out;
 841
 842                 flags = MAP_SHARED;
 843         }
 844
 845         if (shmflg & SHM_RDONLY) {
 846                 prot = PROT_READ;
 847                 acc_mode = S_IRUGO;
 848                 f_mode = FMODE_READ;
 849         } else {
 850                 prot = PROT_READ | PROT_WRITE;
 851                 acc_mode = S_IRUGO | S_IWUGO;
 852                 f_mode = FMODE_READ | FMODE_WRITE;
 853         }
 854         if (shmflg & SHM_EXEC) {
 855                 prot |= PROT_EXEC;
 856                 acc_mode |= S_IXUGO;
 857         }
 858
 859         /*
 860          * We cannot rely on the fs check since SYSV IPC does have an
 861          * additional creator id...
 862          */
 863         ns = current->nsproxy->ipc_ns;
 864         shp = shm_lock_check(ns, shmid);
 865         if (IS_ERR(shp)) {
 866                 err = PTR_ERR(shp);
 867                 goto out;
 868         }
 869
 870         err = -EACCES;
 871         if (ipcperms(&shp->shm_perm, acc_mode))
 872                 goto out_unlock;
 873
 874         err = security_shm_shmat(shp, shmaddr, shmflg);
 875         if (err)
 876                 goto out_unlock;
 877
 878         path.dentry = dget(shp->shm_file->f_path.dentry);
 879         path.mnt    = shp->shm_file->f_path.mnt;
 880         shp->shm_nattch++;
 881         size = i_size_read(path.dentry->d_inode);
 882         shm_unlock(shp);
 883
 884         err = -ENOMEM;
 885         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 886         if (!sfd)
 887                 goto out_put_dentry;
 888
 889         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 890         if (!file)
 891                 goto out_free;
 892         ima_shm_check(file);
 893
 894         file->private_data = sfd;
 895         file->f_mapping = shp->shm_file->f_mapping;
 896         sfd->id = shp->shm_perm.id;
 897         sfd->ns = get_ipc_ns(ns);
 898         sfd->file = shp->shm_file;
 899         sfd->vm_ops = NULL;
 900
 901         down_write(&current->mm->mmap_sem);
 902         if (addr && !(shmflg & SHM_REMAP)) {
 903                 err = -EINVAL;
 904                 if (find_vma_intersection(current->mm, addr, addr + size))
 905                         goto invalid;
 906                 /*
 907                  * If shm segment goes below stack, make sure there is some
 908                  * space left for the stack to grow (at least 4 pages).
 909                  */
 910                 if (addr < current->mm->start_stack &&
 911                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 912                         goto invalid;
 913         }
 914
 915         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 916         *raddr = user_addr;
 917         err = 0;
 918         if (IS_ERR_VALUE(user_addr))
 919                 err = (long)user_addr;
 920 invalid:
 921         up_write(&current->mm->mmap_sem);
 922
 923         fput(file);
 924
 925 out_nattch:
 926         down_write(&shm_ids(ns).rw_mutex);
 927         shp = shm_lock(ns, shmid);
 928         BUG_ON(IS_ERR(shp));
 929         shp->shm_nattch--;
 930         if(shp->shm_nattch == 0 &&
 931            shp->shm_perm.mode & SHM_DEST)
 932                 shm_destroy(ns, shp);
 933         else
 934                 shm_unlock(shp);
 935         up_write(&shm_ids(ns).rw_mutex);
 936
 937 out:
 938         return err;
 939
 940 out_unlock:
 941         shm_unlock(shp);
 942         goto out;
 943
 944 out_free:
 945         kfree(sfd);
 946 out_put_dentry:
 947         dput(path.dentry);
 948         goto out_nattch;
 949 }
 950
 951 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 952 {
 953         unsigned long ret;
 954         long err;
 955
 956         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 957         if (err)
 958                 return err;
 959         force_successful_syscall_return();
 960         return (long)ret;
 961 }
 962
 963 /*
 964  * detach and kill segment if marked destroyed.
 965  * The work is done in shm_close.
 966  */
 967 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 968 {
 969         struct mm_struct *mm = current->mm;
 970         struct vm_area_struct *vma, *next;
 971         unsigned long addr = (unsigned long)shmaddr;
 972         loff_t size = 0;
 973         int retval = -EINVAL;
 974
 975         if (addr & ~PAGE_MASK)
 976                 return retval;
 977
 978         down_write(&mm->mmap_sem);
 979
 980         /*
 981          * This function tries to be smart and unmap shm segments that
 982          * were modified by partial mlock or munmap calls:
 983          * - It first determines the size of the shm segment that should be
 984          *   unmapped: It searches for a vma that is backed by shm and that
 985          *   started at address shmaddr. It records it's size and then unmaps
 986          *   it.
 987          * - Then it unmaps all shm vmas that started at shmaddr and that
 988          *   are within the initially determined size.
 989          * Errors from do_munmap are ignored: the function only fails if
 990          * it's called with invalid parameters or if it's called to unmap
 991          * a part of a vma. Both calls in this function are for full vmas,
 992          * the parameters are directly copied from the vma itself and always
 993          * valid - therefore do_munmap cannot fail. (famous last words?)
 994          */
 995         /*
 996          * If it had been mremap()'d, the starting address would not
 997          * match the usual checks anyway. So assume all vma's are
 998          * above the starting address given.
 999          */
1000         vma = find_vma(mm, addr);
1001
1002 #ifdef CONFIG_MMU
1003         while (vma) {
1004                 next = vma->vm_next;
1005
1006                 /*
1007                  * Check if the starting address would match, i.e. it's
1008                  * a fragment created by mprotect() and/or munmap(), or it
1009                  * otherwise it starts at this address with no hassles.
1010                  */
1011                 if ((vma->vm_ops == &shm_vm_ops) &&
1012                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1013
1014
1015                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1016                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1017                         /*
1018                          * We discovered the size of the shm segment, so
1019                          * break out of here and fall through to the next
1020                          * loop that uses the size information to stop
1021                          * searching for matching vma's.
1022                          */
1023                         retval = 0;
1024                         vma = next;
1025                         break;
1026                 }
1027                 vma = next;
1028         }
1029
1030         /*
1031          * We need look no further than the maximum address a fragment
1032          * could possibly have landed at. Also cast things to loff_t to
1033          * prevent overflows and make comparisions vs. equal-width types.
1034          */
1035         size = PAGE_ALIGN(size);
1036         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1037                 next = vma->vm_next;
1038
1039                 /* finding a matching vma now does not alter retval */
1040                 if ((vma->vm_ops == &shm_vm_ops) &&
1041                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1042
1043                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1044                 vma = next;
1045         }
1046
1047 #else /* CONFIG_MMU */
1048         /* under NOMMU conditions, the exact address to be destroyed must be
1049          * given */
1050         retval = -EINVAL;
1051         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1052                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1053                 retval = 0;
1054         }
1055
1056 #endif
1057
1058         up_write(&mm->mmap_sem);
1059         return retval;
1060 }
1061
1062 #ifdef CONFIG_PROC_FS
1063 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1064 {
1065         struct shmid_kernel *shp = it;
1066
1067 #if BITS_PER_LONG <= 32
1068 #define SIZE_SPEC "%10lu"
1069 #else
1070 #define SIZE_SPEC "%21lu"
1071 #endif
1072
1073         return seq_printf(s,
1074                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1075                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1076                           shp->shm_perm.key,
1077                           shp->shm_perm.id,
1078                           shp->shm_perm.mode,
1079                           shp->shm_segsz,
1080                           shp->shm_cprid,
1081                           shp->shm_lprid,
1082                           shp->shm_nattch,
1083                           shp->shm_perm.uid,
1084                           shp->shm_perm.gid,
1085                           shp->shm_perm.cuid,
1086                           shp->shm_perm.cgid,
1087                           shp->shm_atim,
1088                           shp->shm_dtim,
1089                           shp->shm_ctim);
1090 }
1091 #endif