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_rmid_forced = 0;
  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         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 105 }
 106 #endif
 107
 108 void __init shm_init (void)
 109 {
 110         shm_init_ns(&init_ipc_ns);
 111         ipc_init_proc_interface("sysvipc/shm",
 112 #if BITS_PER_LONG <= 32
 113                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 114 #else
 115                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 116 #endif
 117                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 118 }
 119
 120 /*
 121  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 122  * is not necessarily held.
 123  */
 124 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 125 {
 126         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 127
 128         if (IS_ERR(ipcp))
 129                 return (struct shmid_kernel *)ipcp;
 130
 131         return container_of(ipcp, struct shmid_kernel, shm_perm);
 132 }
 133
 134 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 135                                                 int id)
 136 {
 137         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 138
 139         if (IS_ERR(ipcp))
 140                 return (struct shmid_kernel *)ipcp;
 141
 142         return container_of(ipcp, struct shmid_kernel, shm_perm);
 143 }
 144
 145 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 146 {
 147         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 148 }
 149
 150
 151 /* This is called by fork, once for every shm attach. */
 152 static void shm_open(struct vm_area_struct *vma)
 153 {
 154         struct file *file = vma->vm_file;
 155         struct shm_file_data *sfd = shm_file_data(file);
 156         struct shmid_kernel *shp;
 157
 158         shp = shm_lock(sfd->ns, sfd->id);
 159         BUG_ON(IS_ERR(shp));
 160         shp->shm_atim = get_seconds();
 161         shp->shm_lprid = task_tgid_vnr(current);
 162         shp->shm_nattch++;
 163         shm_unlock(shp);
 164 }
 165
 166 /*
 167  * shm_destroy - free the struct shmid_kernel
 168  *
 169  * @ns: namespace
 170  * @shp: struct to free
 171  *
 172  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 173  * but returns with shp unlocked and freed.
 174  */
 175 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 176 {
 177         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 178         shm_rmid(ns, shp);
 179         shm_unlock(shp);
 180         if (!is_file_hugepages(shp->shm_file))
 181                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 182         else if (shp->mlock_user)
 183                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 184                                                 shp->mlock_user);
 185         fput (shp->shm_file);
 186         security_shm_free(shp);
 187         ipc_rcu_putref(shp);
 188 }
 189
 190 /*
 191  * shm_may_destroy - identifies whether shm segment should be destroyed now
 192  *
 193  * Returns true if and only if there are no active users of the segment and
 194  * one of the following is true:
 195  *
 196  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 197  *
 198  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 199  */
 200 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 201 {
 202         return (shp->shm_nattch == 0) &&
 203                (ns->shm_rmid_forced ||
 204                 (shp->shm_perm.mode & SHM_DEST));
 205 }
 206
 207 /*
 208  * remove the attach descriptor vma.
 209  * free memory for segment if it is marked destroyed.
 210  * The descriptor has already been removed from the current->mm->mmap list
 211  * and will later be kfree()d.
 212  */
 213 static void shm_close(struct vm_area_struct *vma)
 214 {
 215         struct file * file = vma->vm_file;
 216         struct shm_file_data *sfd = shm_file_data(file);
 217         struct shmid_kernel *shp;
 218         struct ipc_namespace *ns = sfd->ns;
 219
 220         down_write(&shm_ids(ns).rw_mutex);
 221         /* remove from the list of attaches of the shm segment */
 222         shp = shm_lock(ns, sfd->id);
 223         BUG_ON(IS_ERR(shp));
 224         shp->shm_lprid = task_tgid_vnr(current);
 225         shp->shm_dtim = get_seconds();
 226         shp->shm_nattch--;
 227         if (shm_may_destroy(ns, shp))
 228                 shm_destroy(ns, shp);
 229         else
 230                 shm_unlock(shp);
 231         up_write(&shm_ids(ns).rw_mutex);
 232 }
 233
 234 static int shm_try_destroy_current(int id, void *p, void *data)
 235 {
 236         struct ipc_namespace *ns = data;
 237         struct shmid_kernel *shp = shm_lock(ns, id);
 238
 239         if (IS_ERR(shp))
 240                 return 0;
 241
 242         if (shp->shm_cprid != task_tgid_vnr(current)) {
 243                 shm_unlock(shp);
 244                 return 0;
 245         }
 246
 247         if (shm_may_destroy(ns, shp))
 248                 shm_destroy(ns, shp);
 249         else
 250                 shm_unlock(shp);
 251         return 0;
 252 }
 253
 254 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 255 {
 256         struct ipc_namespace *ns = data;
 257         struct shmid_kernel *shp = shm_lock(ns, id);
 258         struct task_struct *task;
 259
 260         if (IS_ERR(shp))
 261                 return 0;
 262
 263         /*
 264          * We want to destroy segments without users and with already
 265          * exit'ed originating process.
 266          *
 267          * XXX: the originating process may exist in another pid namespace.
 268          */
 269         task = find_task_by_vpid(shp->shm_cprid);
 270         if (task != NULL) {
 271                 shm_unlock(shp);
 272                 return 0;
 273         }
 274
 275         if (shm_may_destroy(ns, shp))
 276                 shm_destroy(ns, shp);
 277         else
 278                 shm_unlock(shp);
 279         return 0;
 280 }
 281
 282 void shm_destroy_orphaned(struct ipc_namespace *ns)
 283 {
 284         down_write(&shm_ids(ns).rw_mutex);
 285         idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 286         up_write(&shm_ids(ns).rw_mutex);
 287 }
 288
 289
 290 void exit_shm(struct task_struct *task)
 291 {
 292         struct nsproxy *nsp = task->nsproxy;
 293         struct ipc_namespace *ns;
 294
 295         if (!nsp)
 296                 return;
 297         ns = nsp->ipc_ns;
 298         if (!ns || !ns->shm_rmid_forced)
 299                 return;
 300
 301         /* Destroy all already created segments, but not mapped yet */
 302         down_write(&shm_ids(ns).rw_mutex);
 303         idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
 304         up_write(&shm_ids(ns).rw_mutex);
 305 }
 306
 307 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 308 {
 309         struct file *file = vma->vm_file;
 310         struct shm_file_data *sfd = shm_file_data(file);
 311
 312         return sfd->vm_ops->fault(vma, vmf);
 313 }
 314
 315 #ifdef CONFIG_NUMA
 316 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 317 {
 318         struct file *file = vma->vm_file;
 319         struct shm_file_data *sfd = shm_file_data(file);
 320         int err = 0;
 321         if (sfd->vm_ops->set_policy)
 322                 err = sfd->vm_ops->set_policy(vma, new);
 323         return err;
 324 }
 325
 326 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 327                                         unsigned long addr)
 328 {
 329         struct file *file = vma->vm_file;
 330         struct shm_file_data *sfd = shm_file_data(file);
 331         struct mempolicy *pol = NULL;
 332
 333         if (sfd->vm_ops->get_policy)
 334                 pol = sfd->vm_ops->get_policy(vma, addr);
 335         else if (vma->vm_policy)
 336                 pol = vma->vm_policy;
 337
 338         return pol;
 339 }
 340 #endif
 341
 342 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 343 {
 344         struct shm_file_data *sfd = shm_file_data(file);
 345         int ret;
 346
 347         ret = sfd->file->f_op->mmap(sfd->file, vma);
 348         if (ret != 0)
 349                 return ret;
 350         sfd->vm_ops = vma->vm_ops;
 351 #ifdef CONFIG_MMU
 352         BUG_ON(!sfd->vm_ops->fault);
 353 #endif
 354         vma->vm_ops = &shm_vm_ops;
 355         shm_open(vma);
 356
 357         return ret;
 358 }
 359
 360 static int shm_release(struct inode *ino, struct file *file)
 361 {
 362         struct shm_file_data *sfd = shm_file_data(file);
 363
 364         put_ipc_ns(sfd->ns);
 365         shm_file_data(file) = NULL;
 366         kfree(sfd);
 367         return 0;
 368 }
 369
 370 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 371 {
 372         struct shm_file_data *sfd = shm_file_data(file);
 373
 374         if (!sfd->file->f_op->fsync)
 375                 return -EINVAL;
 376         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 377 }
 378
 379 static unsigned long shm_get_unmapped_area(struct file *file,
 380         unsigned long addr, unsigned long len, unsigned long pgoff,
 381         unsigned long flags)
 382 {
 383         struct shm_file_data *sfd = shm_file_data(file);
 384         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 385                                                 pgoff, flags);
 386 }
 387
 388 static const struct file_operations shm_file_operations = {
 389         .mmap           = shm_mmap,
 390         .fsync          = shm_fsync,
 391         .release        = shm_release,
 392 #ifndef CONFIG_MMU
 393         .get_unmapped_area      = shm_get_unmapped_area,
 394 #endif
 395         .llseek         = noop_llseek,
 396 };
 397
 398 static const struct file_operations shm_file_operations_huge = {
 399         .mmap           = shm_mmap,
 400         .fsync          = shm_fsync,
 401         .release        = shm_release,
 402         .get_unmapped_area      = shm_get_unmapped_area,
 403         .llseek         = noop_llseek,
 404 };
 405
 406 int is_file_shm_hugepages(struct file *file)
 407 {
 408         return file->f_op == &shm_file_operations_huge;
 409 }
 410
 411 static const struct vm_operations_struct shm_vm_ops = {
 412         .open   = shm_open,     /* callback for a new vm-area open */
 413         .close  = shm_close,    /* callback for when the vm-area is released */
 414         .fault  = shm_fault,
 415 #if defined(CONFIG_NUMA)
 416         .set_policy = shm_set_policy,
 417         .get_policy = shm_get_policy,
 418 #endif
 419 };
 420
 421 /**
 422  * newseg - Create a new shared memory segment
 423  * @ns: namespace
 424  * @params: ptr to the structure that contains key, size and shmflg
 425  *
 426  * Called with shm_ids.rw_mutex held as a writer.
 427  */
 428
 429 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 430 {
 431         key_t key = params->key;
 432         int shmflg = params->flg;
 433         size_t size = params->u.size;
 434         int error;
 435         struct shmid_kernel *shp;
 436         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 437         struct file * file;
 438         char name[13];
 439         int id;
 440         vm_flags_t acctflag = 0;
 441
 442         if (size < SHMMIN || size > ns->shm_ctlmax)
 443                 return -EINVAL;
 444
 445         if (ns->shm_tot + numpages > ns->shm_ctlall)
 446                 return -ENOSPC;
 447
 448         shp = ipc_rcu_alloc(sizeof(*shp));
 449         if (!shp)
 450                 return -ENOMEM;
 451
 452         shp->shm_perm.key = key;
 453         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 454         shp->mlock_user = NULL;
 455
 456         shp->shm_perm.security = NULL;
 457         error = security_shm_alloc(shp);
 458         if (error) {
 459                 ipc_rcu_putref(shp);
 460                 return error;
 461         }
 462
 463         sprintf (name, "SYSV%08x", key);
 464         if (shmflg & SHM_HUGETLB) {
 465                 /* hugetlb_file_setup applies strict accounting */
 466                 if (shmflg & SHM_NORESERVE)
 467                         acctflag = VM_NORESERVE;
 468                 file = hugetlb_file_setup(name, size, acctflag,
 469                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
 470         } else {
 471                 /*
 472                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 473                  * if it's asked for.
 474                  */
 475                 if  ((shmflg & SHM_NORESERVE) &&
 476                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 477                         acctflag = VM_NORESERVE;
 478                 file = shmem_file_setup(name, size, acctflag);
 479         }
 480         error = PTR_ERR(file);
 481         if (IS_ERR(file))
 482                 goto no_file;
 483
 484         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 485         if (id < 0) {
 486                 error = id;
 487                 goto no_id;
 488         }
 489
 490         shp->shm_cprid = task_tgid_vnr(current);
 491         shp->shm_lprid = 0;
 492         shp->shm_atim = shp->shm_dtim = 0;
 493         shp->shm_ctim = get_seconds();
 494         shp->shm_segsz = size;
 495         shp->shm_nattch = 0;
 496         shp->shm_file = file;
 497         /*
 498          * shmid gets reported as "inode#" in /proc/pid/maps.
 499          * proc-ps tools use this. Changing this will break them.
 500          */
 501         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 502
 503         ns->shm_tot += numpages;
 504         error = shp->shm_perm.id;
 505         shm_unlock(shp);
 506         return error;
 507
 508 no_id:
 509         if (is_file_hugepages(file) && shp->mlock_user)
 510                 user_shm_unlock(size, shp->mlock_user);
 511         fput(file);
 512 no_file:
 513         security_shm_free(shp);
 514         ipc_rcu_putref(shp);
 515         return error;
 516 }
 517
 518 /*
 519  * Called with shm_ids.rw_mutex and ipcp locked.
 520  */
 521 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 522 {
 523         struct shmid_kernel *shp;
 524
 525         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 526         return security_shm_associate(shp, shmflg);
 527 }
 528
 529 /*
 530  * Called with shm_ids.rw_mutex and ipcp locked.
 531  */
 532 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 533                                 struct ipc_params *params)
 534 {
 535         struct shmid_kernel *shp;
 536
 537         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 538         if (shp->shm_segsz < params->u.size)
 539                 return -EINVAL;
 540
 541         return 0;
 542 }
 543
 544 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 545 {
 546         struct ipc_namespace *ns;
 547         struct ipc_ops shm_ops;
 548         struct ipc_params shm_params;
 549
 550         ns = current->nsproxy->ipc_ns;
 551
 552         shm_ops.getnew = newseg;
 553         shm_ops.associate = shm_security;
 554         shm_ops.more_checks = shm_more_checks;
 555
 556         shm_params.key = key;
 557         shm_params.flg = shmflg;
 558         shm_params.u.size = size;
 559
 560         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 561 }
 562
 563 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 564 {
 565         switch(version) {
 566         case IPC_64:
 567                 return copy_to_user(buf, in, sizeof(*in));
 568         case IPC_OLD:
 569             {
 570                 struct shmid_ds out;
 571
 572                 memset(&out, 0, sizeof(out));
 573                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 574                 out.shm_segsz   = in->shm_segsz;
 575                 out.shm_atime   = in->shm_atime;
 576                 out.shm_dtime   = in->shm_dtime;
 577                 out.shm_ctime   = in->shm_ctime;
 578                 out.shm_cpid    = in->shm_cpid;
 579                 out.shm_lpid    = in->shm_lpid;
 580                 out.shm_nattch  = in->shm_nattch;
 581
 582                 return copy_to_user(buf, &out, sizeof(out));
 583             }
 584         default:
 585                 return -EINVAL;
 586         }
 587 }
 588
 589 static inline unsigned long
 590 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 591 {
 592         switch(version) {
 593         case IPC_64:
 594                 if (copy_from_user(out, buf, sizeof(*out)))
 595                         return -EFAULT;
 596                 return 0;
 597         case IPC_OLD:
 598             {
 599                 struct shmid_ds tbuf_old;
 600
 601                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 602                         return -EFAULT;
 603
 604                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 605                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 606                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 607
 608                 return 0;
 609             }
 610         default:
 611                 return -EINVAL;
 612         }
 613 }
 614
 615 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 616 {
 617         switch(version) {
 618         case IPC_64:
 619                 return copy_to_user(buf, in, sizeof(*in));
 620         case IPC_OLD:
 621             {
 622                 struct shminfo out;
 623
 624                 if(in->shmmax > INT_MAX)
 625                         out.shmmax = INT_MAX;
 626                 else
 627                         out.shmmax = (int)in->shmmax;
 628
 629                 out.shmmin      = in->shmmin;
 630                 out.shmmni      = in->shmmni;
 631                 out.shmseg      = in->shmseg;
 632                 out.shmall      = in->shmall;
 633
 634                 return copy_to_user(buf, &out, sizeof(out));
 635             }
 636         default:
 637                 return -EINVAL;
 638         }
 639 }
 640
 641 /*
 642  * Calculate and add used RSS and swap pages of a shm.
 643  * Called with shm_ids.rw_mutex held as a reader
 644  */
 645 static void shm_add_rss_swap(struct shmid_kernel *shp,
 646         unsigned long *rss_add, unsigned long *swp_add)
 647 {
 648         struct inode *inode;
 649
 650         inode = shp->shm_file->f_path.dentry->d_inode;
 651
 652         if (is_file_hugepages(shp->shm_file)) {
 653                 struct address_space *mapping = inode->i_mapping;
 654                 struct hstate *h = hstate_file(shp->shm_file);
 655                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 656         } else {
 657 #ifdef CONFIG_SHMEM
 658                 struct shmem_inode_info *info = SHMEM_I(inode);
 659                 spin_lock(&info->lock);
 660                 *rss_add += inode->i_mapping->nrpages;
 661                 *swp_add += info->swapped;
 662                 spin_unlock(&info->lock);
 663 #else
 664                 *rss_add += inode->i_mapping->nrpages;
 665 #endif
 666         }
 667 }
 668
 669 /*
 670  * Called with shm_ids.rw_mutex held as a reader
 671  */
 672 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 673                 unsigned long *swp)
 674 {
 675         int next_id;
 676         int total, in_use;
 677
 678         *rss = 0;
 679         *swp = 0;
 680
 681         in_use = shm_ids(ns).in_use;
 682
 683         for (total = 0, next_id = 0; total < in_use; next_id++) {
 684                 struct kern_ipc_perm *ipc;
 685                 struct shmid_kernel *shp;
 686
 687                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 688                 if (ipc == NULL)
 689                         continue;
 690                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 691
 692                 shm_add_rss_swap(shp, rss, swp);
 693
 694                 total++;
 695         }
 696 }
 697
 698 /*
 699  * This function handles some shmctl commands which require the rw_mutex
 700  * to be held in write mode.
 701  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 702  */
 703 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 704                        struct shmid_ds __user *buf, int version)
 705 {
 706         struct kern_ipc_perm *ipcp;
 707         struct shmid64_ds shmid64;
 708         struct shmid_kernel *shp;
 709         int err;
 710
 711         if (cmd == IPC_SET) {
 712                 if (copy_shmid_from_user(&shmid64, buf, version))
 713                         return -EFAULT;
 714         }
 715
 716         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
 717                                &shmid64.shm_perm, 0);
 718         if (IS_ERR(ipcp))
 719                 return PTR_ERR(ipcp);
 720
 721         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 722
 723         err = security_shm_shmctl(shp, cmd);
 724         if (err)
 725                 goto out_unlock;
 726         switch (cmd) {
 727         case IPC_RMID:
 728                 do_shm_rmid(ns, ipcp);
 729                 goto out_up;
 730         case IPC_SET:
 731                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 732                 shp->shm_ctim = get_seconds();
 733                 break;
 734         default:
 735                 err = -EINVAL;
 736         }
 737 out_unlock:
 738         shm_unlock(shp);
 739 out_up:
 740         up_write(&shm_ids(ns).rw_mutex);
 741         return err;
 742 }
 743
 744 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 745 {
 746         struct shmid_kernel *shp;
 747         int err, version;
 748         struct ipc_namespace *ns;
 749
 750         if (cmd < 0 || shmid < 0) {
 751                 err = -EINVAL;
 752                 goto out;
 753         }
 754
 755         version = ipc_parse_version(&cmd);
 756         ns = current->nsproxy->ipc_ns;
 757
 758         switch (cmd) { /* replace with proc interface ? */
 759         case IPC_INFO:
 760         {
 761                 struct shminfo64 shminfo;
 762
 763                 err = security_shm_shmctl(NULL, cmd);
 764                 if (err)
 765                         return err;
 766
 767                 memset(&shminfo, 0, sizeof(shminfo));
 768                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 769                 shminfo.shmmax = ns->shm_ctlmax;
 770                 shminfo.shmall = ns->shm_ctlall;
 771
 772                 shminfo.shmmin = SHMMIN;
 773                 if(copy_shminfo_to_user (buf, &shminfo, version))
 774                         return -EFAULT;
 775
 776                 down_read(&shm_ids(ns).rw_mutex);
 777                 err = ipc_get_maxid(&shm_ids(ns));
 778                 up_read(&shm_ids(ns).rw_mutex);
 779
 780                 if(err<0)
 781                         err = 0;
 782                 goto out;
 783         }
 784         case SHM_INFO:
 785         {
 786                 struct shm_info shm_info;
 787
 788                 err = security_shm_shmctl(NULL, cmd);
 789                 if (err)
 790                         return err;
 791
 792                 memset(&shm_info, 0, sizeof(shm_info));
 793                 down_read(&shm_ids(ns).rw_mutex);
 794                 shm_info.used_ids = shm_ids(ns).in_use;
 795                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 796                 shm_info.shm_tot = ns->shm_tot;
 797                 shm_info.swap_attempts = 0;
 798                 shm_info.swap_successes = 0;
 799                 err = ipc_get_maxid(&shm_ids(ns));
 800                 up_read(&shm_ids(ns).rw_mutex);
 801                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 802                         err = -EFAULT;
 803                         goto out;
 804                 }
 805
 806                 err = err < 0 ? 0 : err;
 807                 goto out;
 808         }
 809         case SHM_STAT:
 810         case IPC_STAT:
 811         {
 812                 struct shmid64_ds tbuf;
 813                 int result;
 814
 815                 if (cmd == SHM_STAT) {
 816                         shp = shm_lock(ns, shmid);
 817                         if (IS_ERR(shp)) {
 818                                 err = PTR_ERR(shp);
 819                                 goto out;
 820                         }
 821                         result = shp->shm_perm.id;
 822                 } else {
 823                         shp = shm_lock_check(ns, shmid);
 824                         if (IS_ERR(shp)) {
 825                                 err = PTR_ERR(shp);
 826                                 goto out;
 827                         }
 828                         result = 0;
 829                 }
 830                 err = -EACCES;
 831                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 832                         goto out_unlock;
 833                 err = security_shm_shmctl(shp, cmd);
 834                 if (err)
 835                         goto out_unlock;
 836                 memset(&tbuf, 0, sizeof(tbuf));
 837                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 838                 tbuf.shm_segsz  = shp->shm_segsz;
 839                 tbuf.shm_atime  = shp->shm_atim;
 840                 tbuf.shm_dtime  = shp->shm_dtim;
 841                 tbuf.shm_ctime  = shp->shm_ctim;
 842                 tbuf.shm_cpid   = shp->shm_cprid;
 843                 tbuf.shm_lpid   = shp->shm_lprid;
 844                 tbuf.shm_nattch = shp->shm_nattch;
 845                 shm_unlock(shp);
 846                 if(copy_shmid_to_user (buf, &tbuf, version))
 847                         err = -EFAULT;
 848                 else
 849                         err = result;
 850                 goto out;
 851         }
 852         case SHM_LOCK:
 853         case SHM_UNLOCK:
 854         {
 855                 struct file *uninitialized_var(shm_file);
 856
 857                 lru_add_drain_all();  /* drain pagevecs to lru lists */
 858
 859                 shp = shm_lock_check(ns, shmid);
 860                 if (IS_ERR(shp)) {
 861                         err = PTR_ERR(shp);
 862                         goto out;
 863                 }
 864
 865                 audit_ipc_obj(&(shp->shm_perm));
 866
 867                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
 868                         uid_t euid = current_euid();
 869                         err = -EPERM;
 870                         if (euid != shp->shm_perm.uid &&
 871                             euid != shp->shm_perm.cuid)
 872                                 goto out_unlock;
 873                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
 874                                 goto out_unlock;
 875                 }
 876
 877                 err = security_shm_shmctl(shp, cmd);
 878                 if (err)
 879                         goto out_unlock;
 880
 881                 if(cmd==SHM_LOCK) {
 882                         struct user_struct *user = current_user();
 883                         if (!is_file_hugepages(shp->shm_file)) {
 884                                 err = shmem_lock(shp->shm_file, 1, user);
 885                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 886                                         shp->shm_perm.mode |= SHM_LOCKED;
 887                                         shp->mlock_user = user;
 888                                 }
 889                         }
 890                 } else if (!is_file_hugepages(shp->shm_file)) {
 891                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 892                         shp->shm_perm.mode &= ~SHM_LOCKED;
 893                         shp->mlock_user = NULL;
 894                 }
 895                 shm_unlock(shp);
 896                 goto out;
 897         }
 898         case IPC_RMID:
 899         case IPC_SET:
 900                 err = shmctl_down(ns, shmid, cmd, buf, version);
 901                 return err;
 902         default:
 903                 return -EINVAL;
 904         }
 905
 906 out_unlock:
 907         shm_unlock(shp);
 908 out:
 909         return err;
 910 }
 911
 912 /*
 913  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 914  *
 915  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 916  * "raddr" thing points to kernel space, and there has to be a wrapper around
 917  * this.
 918  */
 919 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 920 {
 921         struct shmid_kernel *shp;
 922         unsigned long addr;
 923         unsigned long size;
 924         struct file * file;
 925         int    err;
 926         unsigned long flags;
 927         unsigned long prot;
 928         int acc_mode;
 929         unsigned long user_addr;
 930         struct ipc_namespace *ns;
 931         struct shm_file_data *sfd;
 932         struct path path;
 933         fmode_t f_mode;
 934
 935         err = -EINVAL;
 936         if (shmid < 0)
 937                 goto out;
 938         else if ((addr = (ulong)shmaddr)) {
 939                 if (addr & (SHMLBA-1)) {
 940                         if (shmflg & SHM_RND)
 941                                 addr &= ~(SHMLBA-1);       /* round down */
 942                         else
 943 #ifndef __ARCH_FORCE_SHMLBA
 944                                 if (addr & ~PAGE_MASK)
 945 #endif
 946                                         goto out;
 947                 }
 948                 flags = MAP_SHARED | MAP_FIXED;
 949         } else {
 950                 if ((shmflg & SHM_REMAP))
 951                         goto out;
 952
 953                 flags = MAP_SHARED;
 954         }
 955
 956         if (shmflg & SHM_RDONLY) {
 957                 prot = PROT_READ;
 958                 acc_mode = S_IRUGO;
 959                 f_mode = FMODE_READ;
 960         } else {
 961                 prot = PROT_READ | PROT_WRITE;
 962                 acc_mode = S_IRUGO | S_IWUGO;
 963                 f_mode = FMODE_READ | FMODE_WRITE;
 964         }
 965         if (shmflg & SHM_EXEC) {
 966                 prot |= PROT_EXEC;
 967                 acc_mode |= S_IXUGO;
 968         }
 969
 970         /*
 971          * We cannot rely on the fs check since SYSV IPC does have an
 972          * additional creator id...
 973          */
 974         ns = current->nsproxy->ipc_ns;
 975         shp = shm_lock_check(ns, shmid);
 976         if (IS_ERR(shp)) {
 977                 err = PTR_ERR(shp);
 978                 goto out;
 979         }
 980
 981         err = -EACCES;
 982         if (ipcperms(ns, &shp->shm_perm, acc_mode))
 983                 goto out_unlock;
 984
 985         err = security_shm_shmat(shp, shmaddr, shmflg);
 986         if (err)
 987                 goto out_unlock;
 988
 989         path = shp->shm_file->f_path;
 990         path_get(&path);
 991         shp->shm_nattch++;
 992         size = i_size_read(path.dentry->d_inode);
 993         shm_unlock(shp);
 994
 995         err = -ENOMEM;
 996         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 997         if (!sfd)
 998                 goto out_put_dentry;
 999
1000         file = alloc_file(&path, f_mode,
1001                           is_file_hugepages(shp->shm_file) ?
1002                                 &shm_file_operations_huge :
1003                                 &shm_file_operations);
1004         if (!file)
1005                 goto out_free;
1006
1007         file->private_data = sfd;
1008         file->f_mapping = shp->shm_file->f_mapping;
1009         sfd->id = shp->shm_perm.id;
1010         sfd->ns = get_ipc_ns(ns);
1011         sfd->file = shp->shm_file;
1012         sfd->vm_ops = NULL;
1013
1014         down_write(&current->mm->mmap_sem);
1015         if (addr && !(shmflg & SHM_REMAP)) {
1016                 err = -EINVAL;
1017                 if (find_vma_intersection(current->mm, addr, addr + size))
1018                         goto invalid;
1019                 /*
1020                  * If shm segment goes below stack, make sure there is some
1021                  * space left for the stack to grow (at least 4 pages).
1022                  */
1023                 if (addr < current->mm->start_stack &&
1024                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1025                         goto invalid;
1026         }
1027
1028         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1029         *raddr = user_addr;
1030         err = 0;
1031         if (IS_ERR_VALUE(user_addr))
1032                 err = (long)user_addr;
1033 invalid:
1034         up_write(&current->mm->mmap_sem);
1035
1036         fput(file);
1037
1038 out_nattch:
1039         down_write(&shm_ids(ns).rw_mutex);
1040         shp = shm_lock(ns, shmid);
1041         BUG_ON(IS_ERR(shp));
1042         shp->shm_nattch--;
1043         if (shm_may_destroy(ns, shp))
1044                 shm_destroy(ns, shp);
1045         else
1046                 shm_unlock(shp);
1047         up_write(&shm_ids(ns).rw_mutex);
1048
1049 out:
1050         return err;
1051
1052 out_unlock:
1053         shm_unlock(shp);
1054         goto out;
1055
1056 out_free:
1057         kfree(sfd);
1058 out_put_dentry:
1059         path_put(&path);
1060         goto out_nattch;
1061 }
1062
1063 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1064 {
1065         unsigned long ret;
1066         long err;
1067
1068         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1069         if (err)
1070                 return err;
1071         force_successful_syscall_return();
1072         return (long)ret;
1073 }
1074
1075 /*
1076  * detach and kill segment if marked destroyed.
1077  * The work is done in shm_close.
1078  */
1079 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1080 {
1081         struct mm_struct *mm = current->mm;
1082         struct vm_area_struct *vma;
1083         unsigned long addr = (unsigned long)shmaddr;
1084         int retval = -EINVAL;
1085 #ifdef CONFIG_MMU
1086         loff_t size = 0;
1087         struct vm_area_struct *next;
1088 #endif
1089
1090         if (addr & ~PAGE_MASK)
1091                 return retval;
1092
1093         down_write(&mm->mmap_sem);
1094
1095         /*
1096          * This function tries to be smart and unmap shm segments that
1097          * were modified by partial mlock or munmap calls:
1098          * - It first determines the size of the shm segment that should be
1099          *   unmapped: It searches for a vma that is backed by shm and that
1100          *   started at address shmaddr. It records it's size and then unmaps
1101          *   it.
1102          * - Then it unmaps all shm vmas that started at shmaddr and that
1103          *   are within the initially determined size.
1104          * Errors from do_munmap are ignored: the function only fails if
1105          * it's called with invalid parameters or if it's called to unmap
1106          * a part of a vma. Both calls in this function are for full vmas,
1107          * the parameters are directly copied from the vma itself and always
1108          * valid - therefore do_munmap cannot fail. (famous last words?)
1109          */
1110         /*
1111          * If it had been mremap()'d, the starting address would not
1112          * match the usual checks anyway. So assume all vma's are
1113          * above the starting address given.
1114          */
1115         vma = find_vma(mm, addr);
1116
1117 #ifdef CONFIG_MMU
1118         while (vma) {
1119                 next = vma->vm_next;
1120
1121                 /*
1122                  * Check if the starting address would match, i.e. it's
1123                  * a fragment created by mprotect() and/or munmap(), or it
1124                  * otherwise it starts at this address with no hassles.
1125                  */
1126                 if ((vma->vm_ops == &shm_vm_ops) &&
1127                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1128
1129
1130                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1131                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1132                         /*
1133                          * We discovered the size of the shm segment, so
1134                          * break out of here and fall through to the next
1135                          * loop that uses the size information to stop
1136                          * searching for matching vma's.
1137                          */
1138                         retval = 0;
1139                         vma = next;
1140                         break;
1141                 }
1142                 vma = next;
1143         }
1144
1145         /*
1146          * We need look no further than the maximum address a fragment
1147          * could possibly have landed at. Also cast things to loff_t to
1148          * prevent overflows and make comparisons vs. equal-width types.
1149          */
1150         size = PAGE_ALIGN(size);
1151         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1152                 next = vma->vm_next;
1153
1154                 /* finding a matching vma now does not alter retval */
1155                 if ((vma->vm_ops == &shm_vm_ops) &&
1156                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1157
1158                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1159                 vma = next;
1160         }
1161
1162 #else /* CONFIG_MMU */
1163         /* under NOMMU conditions, the exact address to be destroyed must be
1164          * given */
1165         retval = -EINVAL;
1166         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1167                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1168                 retval = 0;
1169         }
1170
1171 #endif
1172
1173         up_write(&mm->mmap_sem);
1174         return retval;
1175 }
1176
1177 #ifdef CONFIG_PROC_FS
1178 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1179 {
1180         struct shmid_kernel *shp = it;
1181         unsigned long rss = 0, swp = 0;
1182
1183         shm_add_rss_swap(shp, &rss, &swp);
1184
1185 #if BITS_PER_LONG <= 32
1186 #define SIZE_SPEC "%10lu"
1187 #else
1188 #define SIZE_SPEC "%21lu"
1189 #endif
1190
1191         return seq_printf(s,
1192                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1193                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1194                           SIZE_SPEC " " SIZE_SPEC "\n",
1195                           shp->shm_perm.key,
1196                           shp->shm_perm.id,
1197                           shp->shm_perm.mode,
1198                           shp->shm_segsz,
1199                           shp->shm_cprid,
1200                           shp->shm_lprid,
1201                           shp->shm_nattch,
1202                           shp->shm_perm.uid,
1203                           shp->shm_perm.gid,
1204                           shp->shm_perm.cuid,
1205                           shp->shm_perm.cgid,
1206                           shp->shm_atim,
1207                           shp->shm_dtim,
1208                           shp->shm_ctim,
1209                           rss * PAGE_SIZE,
1210                           swp * PAGE_SIZE);
1211 }
1212 #endif