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  * Better ipc lock (kern_ipc_perm.lock) handling
  24  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
  25  */
  26
  27 #include <linux/slab.h>
  28 #include <linux/mm.h>
  29 #include <linux/hugetlb.h>
  30 #include <linux/shm.h>
  31 #include <linux/init.h>
  32 #include <linux/file.h>
  33 #include <linux/mman.h>
  34 #include <linux/shmem_fs.h>
  35 #include <linux/security.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/audit.h>
  38 #include <linux/capability.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/rwsem.h>
  42 #include <linux/nsproxy.h>
  43 #include <linux/mount.h>
  44 #include <linux/ipc_namespace.h>
  45
  46 #include <linux/uaccess.h>
  47
  48 #include "util.h"
  49
  50 struct shm_file_data {
  51         int id;
  52         struct ipc_namespace *ns;
  53         struct file *file;
  54         const struct vm_operations_struct *vm_ops;
  55 };
  56
  57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  58
  59 static const struct file_operations shm_file_operations;
  60 static const struct vm_operations_struct shm_vm_ops;
  61
  62 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  63
  64 #define shm_unlock(shp)                 \
  65         ipc_unlock(&(shp)->shm_perm)
  66
  67 static int newseg(struct ipc_namespace *, struct ipc_params *);
  68 static void shm_open(struct vm_area_struct *vma);
  69 static void shm_close(struct vm_area_struct *vma);
  70 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
  71 #ifdef CONFIG_PROC_FS
  72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  73 #endif
  74
  75 void shm_init_ns(struct ipc_namespace *ns)
  76 {
  77         ns->shm_ctlmax = SHMMAX;
  78         ns->shm_ctlall = SHMALL;
  79         ns->shm_ctlmni = SHMMNI;
  80         ns->shm_rmid_forced = 0;
  81         ns->shm_tot = 0;
  82         ipc_init_ids(&shm_ids(ns));
  83 }
  84
  85 /*
  86  * Called with shm_ids.rwsem (writer) and the shp structure locked.
  87  * Only shm_ids.rwsem remains locked on exit.
  88  */
  89 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  90 {
  91         struct shmid_kernel *shp;
  92         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  93
  94         if (shp->shm_nattch) {
  95                 shp->shm_perm.mode |= SHM_DEST;
  96                 /* Do not find it any more */
  97                 shp->shm_perm.key = IPC_PRIVATE;
  98                 shm_unlock(shp);
  99         } else
 100                 shm_destroy(ns, shp);
 101 }
 102
 103 #ifdef CONFIG_IPC_NS
 104 void shm_exit_ns(struct ipc_namespace *ns)
 105 {
 106         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 107         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 108 }
 109 #endif
 110
 111 static int __init ipc_ns_init(void)
 112 {
 113         shm_init_ns(&init_ipc_ns);
 114         return 0;
 115 }
 116
 117 pure_initcall(ipc_ns_init);
 118
 119 void __init shm_init(void)
 120 {
 121         ipc_init_proc_interface("sysvipc/shm",
 122 #if BITS_PER_LONG <= 32
 123                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 124 #else
 125                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 126 #endif
 127                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 128 }
 129
 130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 131 {
 132         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 133
 134         if (IS_ERR(ipcp))
 135                 return ERR_CAST(ipcp);
 136
 137         return container_of(ipcp, struct shmid_kernel, shm_perm);
 138 }
 139
 140 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 141 {
 142         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 143
 144         if (IS_ERR(ipcp))
 145                 return ERR_CAST(ipcp);
 146
 147         return container_of(ipcp, struct shmid_kernel, shm_perm);
 148 }
 149
 150 /*
 151  * shm_lock_(check_) routines are called in the paths where the rwsem
 152  * is not necessarily held.
 153  */
 154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 155 {
 156         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 157
 158         /*
 159          * Callers of shm_lock() must validate the status of the returned ipc
 160          * object pointer (as returned by ipc_lock()), and error out as
 161          * appropriate.
 162          */
 163         if (IS_ERR(ipcp))
 164                 return (void *)ipcp;
 165         return container_of(ipcp, struct shmid_kernel, shm_perm);
 166 }
 167
 168 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 169 {
 170         rcu_read_lock();
 171         ipc_lock_object(&ipcp->shm_perm);
 172 }
 173
 174 static void shm_rcu_free(struct rcu_head *head)
 175 {
 176         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
 177         struct shmid_kernel *shp = ipc_rcu_to_struct(p);
 178
 179         security_shm_free(shp);
 180         ipc_rcu_free(head);
 181 }
 182
 183 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 184 {
 185         list_del(&s->shm_clist);
 186         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 187 }
 188
 189
 190 static int __shm_open(struct vm_area_struct *vma)
 191 {
 192         struct file *file = vma->vm_file;
 193         struct shm_file_data *sfd = shm_file_data(file);
 194         struct shmid_kernel *shp;
 195
 196         shp = shm_lock(sfd->ns, sfd->id);
 197
 198         if (IS_ERR(shp))
 199                 return PTR_ERR(shp);
 200
 201         if (shp->shm_file != sfd->file) {
 202                 /* ID was reused */
 203                 shm_unlock(shp);
 204                 return -EINVAL;
 205         }
 206
 207         shp->shm_atim = get_seconds();
 208         shp->shm_lprid = task_tgid_vnr(current);
 209         shp->shm_nattch++;
 210         shm_unlock(shp);
 211         return 0;
 212 }
 213
 214 /* This is called by fork, once for every shm attach. */
 215 static void shm_open(struct vm_area_struct *vma)
 216 {
 217         int err = __shm_open(vma);
 218         /*
 219          * We raced in the idr lookup or with shm_destroy().
 220          * Either way, the ID is busted.
 221          */
 222         WARN_ON_ONCE(err);
 223 }
 224
 225 /*
 226  * shm_destroy - free the struct shmid_kernel
 227  *
 228  * @ns: namespace
 229  * @shp: struct to free
 230  *
 231  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 232  * but returns with shp unlocked and freed.
 233  */
 234 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 235 {
 236         struct file *shm_file;
 237
 238         shm_file = shp->shm_file;
 239         shp->shm_file = NULL;
 240         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 241         shm_rmid(ns, shp);
 242         shm_unlock(shp);
 243         if (!is_file_hugepages(shm_file))
 244                 shmem_lock(shm_file, 0, shp->mlock_user);
 245         else if (shp->mlock_user)
 246                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 247                                 shp->mlock_user);
 248         fput(shm_file);
 249         ipc_rcu_putref(shp, shm_rcu_free);
 250 }
 251
 252 /*
 253  * shm_may_destroy - identifies whether shm segment should be destroyed now
 254  *
 255  * Returns true if and only if there are no active users of the segment and
 256  * one of the following is true:
 257  *
 258  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 259  *
 260  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 261  */
 262 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 263 {
 264         return (shp->shm_nattch == 0) &&
 265                (ns->shm_rmid_forced ||
 266                 (shp->shm_perm.mode & SHM_DEST));
 267 }
 268
 269 /*
 270  * remove the attach descriptor vma.
 271  * free memory for segment if it is marked destroyed.
 272  * The descriptor has already been removed from the current->mm->mmap list
 273  * and will later be kfree()d.
 274  */
 275 static void shm_close(struct vm_area_struct *vma)
 276 {
 277         struct file *file = vma->vm_file;
 278         struct shm_file_data *sfd = shm_file_data(file);
 279         struct shmid_kernel *shp;
 280         struct ipc_namespace *ns = sfd->ns;
 281
 282         down_write(&shm_ids(ns).rwsem);
 283         /* remove from the list of attaches of the shm segment */
 284         shp = shm_lock(ns, sfd->id);
 285
 286         /*
 287          * We raced in the idr lookup or with shm_destroy().
 288          * Either way, the ID is busted.
 289          */
 290         if (WARN_ON_ONCE(IS_ERR(shp)))
 291                 goto done; /* no-op */
 292
 293         shp->shm_lprid = task_tgid_vnr(current);
 294         shp->shm_dtim = get_seconds();
 295         shp->shm_nattch--;
 296         if (shm_may_destroy(ns, shp))
 297                 shm_destroy(ns, shp);
 298         else
 299                 shm_unlock(shp);
 300 done:
 301         up_write(&shm_ids(ns).rwsem);
 302 }
 303
 304 /* Called with ns->shm_ids(ns).rwsem locked */
 305 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 306 {
 307         struct ipc_namespace *ns = data;
 308         struct kern_ipc_perm *ipcp = p;
 309         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 310
 311         /*
 312          * We want to destroy segments without users and with already
 313          * exit'ed originating process.
 314          *
 315          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 316          */
 317         if (shp->shm_creator != NULL)
 318                 return 0;
 319
 320         if (shm_may_destroy(ns, shp)) {
 321                 shm_lock_by_ptr(shp);
 322                 shm_destroy(ns, shp);
 323         }
 324         return 0;
 325 }
 326
 327 void shm_destroy_orphaned(struct ipc_namespace *ns)
 328 {
 329         down_write(&shm_ids(ns).rwsem);
 330         if (shm_ids(ns).in_use)
 331                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 332         up_write(&shm_ids(ns).rwsem);
 333 }
 334
 335 /* Locking assumes this will only be called with task == current */
 336 void exit_shm(struct task_struct *task)
 337 {
 338         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 339         struct shmid_kernel *shp, *n;
 340
 341         if (list_empty(&task->sysvshm.shm_clist))
 342                 return;
 343
 344         /*
 345          * If kernel.shm_rmid_forced is not set then only keep track of
 346          * which shmids are orphaned, so that a later set of the sysctl
 347          * can clean them up.
 348          */
 349         if (!ns->shm_rmid_forced) {
 350                 down_read(&shm_ids(ns).rwsem);
 351                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
 352                         shp->shm_creator = NULL;
 353                 /*
 354                  * Only under read lock but we are only called on current
 355                  * so no entry on the list will be shared.
 356                  */
 357                 list_del(&task->sysvshm.shm_clist);
 358                 up_read(&shm_ids(ns).rwsem);
 359                 return;
 360         }
 361
 362         /*
 363          * Destroy all already created segments, that were not yet mapped,
 364          * and mark any mapped as orphan to cover the sysctl toggling.
 365          * Destroy is skipped if shm_may_destroy() returns false.
 366          */
 367         down_write(&shm_ids(ns).rwsem);
 368         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
 369                 shp->shm_creator = NULL;
 370
 371                 if (shm_may_destroy(ns, shp)) {
 372                         shm_lock_by_ptr(shp);
 373                         shm_destroy(ns, shp);
 374                 }
 375         }
 376
 377         /* Remove the list head from any segments still attached. */
 378         list_del(&task->sysvshm.shm_clist);
 379         up_write(&shm_ids(ns).rwsem);
 380 }
 381
 382 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 383 {
 384         struct file *file = vma->vm_file;
 385         struct shm_file_data *sfd = shm_file_data(file);
 386
 387         return sfd->vm_ops->fault(vma, vmf);
 388 }
 389
 390 #ifdef CONFIG_NUMA
 391 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 392 {
 393         struct file *file = vma->vm_file;
 394         struct shm_file_data *sfd = shm_file_data(file);
 395         int err = 0;
 396         if (sfd->vm_ops->set_policy)
 397                 err = sfd->vm_ops->set_policy(vma, new);
 398         return err;
 399 }
 400
 401 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 402                                         unsigned long addr)
 403 {
 404         struct file *file = vma->vm_file;
 405         struct shm_file_data *sfd = shm_file_data(file);
 406         struct mempolicy *pol = NULL;
 407
 408         if (sfd->vm_ops->get_policy)
 409                 pol = sfd->vm_ops->get_policy(vma, addr);
 410         else if (vma->vm_policy)
 411                 pol = vma->vm_policy;
 412
 413         return pol;
 414 }
 415 #endif
 416
 417 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 418 {
 419         struct shm_file_data *sfd = shm_file_data(file);
 420         int ret;
 421
 422         /*
 423          * In case of remap_file_pages() emulation, the file can represent an
 424          * IPC ID that was removed, and possibly even reused by another shm
 425          * segment already.  Propagate this case as an error to caller.
 426          */
 427         ret =__shm_open(vma);
 428         if (ret)
 429                 return ret;
 430
 431         ret = sfd->file->f_op->mmap(sfd->file, vma);
 432         if (ret) {
 433                 shm_close(vma);
 434                 return ret;
 435         }
 436         sfd->vm_ops = vma->vm_ops;
 437 #ifdef CONFIG_MMU
 438         WARN_ON(!sfd->vm_ops->fault);
 439 #endif
 440         vma->vm_ops = &shm_vm_ops;
 441         return 0;
 442 }
 443
 444 static int shm_release(struct inode *ino, struct file *file)
 445 {
 446         struct shm_file_data *sfd = shm_file_data(file);
 447
 448         put_ipc_ns(sfd->ns);
 449         fput(sfd->file);
 450         shm_file_data(file) = NULL;
 451         kfree(sfd);
 452         return 0;
 453 }
 454
 455 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 456 {
 457         struct shm_file_data *sfd = shm_file_data(file);
 458
 459         if (!sfd->file->f_op->fsync)
 460                 return -EINVAL;
 461         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 462 }
 463
 464 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 465                           loff_t len)
 466 {
 467         struct shm_file_data *sfd = shm_file_data(file);
 468
 469         if (!sfd->file->f_op->fallocate)
 470                 return -EOPNOTSUPP;
 471         return sfd->file->f_op->fallocate(file, mode, offset, len);
 472 }
 473
 474 static unsigned long shm_get_unmapped_area(struct file *file,
 475         unsigned long addr, unsigned long len, unsigned long pgoff,
 476         unsigned long flags)
 477 {
 478         struct shm_file_data *sfd = shm_file_data(file);
 479         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 480                                                 pgoff, flags);
 481 }
 482
 483 static const struct file_operations shm_file_operations = {
 484         .mmap           = shm_mmap,
 485         .fsync          = shm_fsync,
 486         .release        = shm_release,
 487 #ifndef CONFIG_MMU
 488         .get_unmapped_area      = shm_get_unmapped_area,
 489 #endif
 490         .llseek         = noop_llseek,
 491         .fallocate      = shm_fallocate,
 492 };
 493
 494 static const struct file_operations shm_file_operations_huge = {
 495         .mmap           = shm_mmap,
 496         .fsync          = shm_fsync,
 497         .release        = shm_release,
 498         .get_unmapped_area      = shm_get_unmapped_area,
 499         .llseek         = noop_llseek,
 500         .fallocate      = shm_fallocate,
 501 };
 502
 503 int is_file_shm_hugepages(struct file *file)
 504 {
 505         return file->f_op == &shm_file_operations_huge;
 506 }
 507
 508 static const struct vm_operations_struct shm_vm_ops = {
 509         .open   = shm_open,     /* callback for a new vm-area open */
 510         .close  = shm_close,    /* callback for when the vm-area is released */
 511         .fault  = shm_fault,
 512 #if defined(CONFIG_NUMA)
 513         .set_policy = shm_set_policy,
 514         .get_policy = shm_get_policy,
 515 #endif
 516 };
 517
 518 /**
 519  * newseg - Create a new shared memory segment
 520  * @ns: namespace
 521  * @params: ptr to the structure that contains key, size and shmflg
 522  *
 523  * Called with shm_ids.rwsem held as a writer.
 524  */
 525 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 526 {
 527         key_t key = params->key;
 528         int shmflg = params->flg;
 529         size_t size = params->u.size;
 530         int error;
 531         struct shmid_kernel *shp;
 532         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 533         struct file *file;
 534         char name[13];
 535         int id;
 536         vm_flags_t acctflag = 0;
 537
 538         if (size < SHMMIN || size > ns->shm_ctlmax)
 539                 return -EINVAL;
 540
 541         if (numpages << PAGE_SHIFT < size)
 542                 return -ENOSPC;
 543
 544         if (ns->shm_tot + numpages < ns->shm_tot ||
 545                         ns->shm_tot + numpages > ns->shm_ctlall)
 546                 return -ENOSPC;
 547
 548         shp = ipc_rcu_alloc(sizeof(*shp));
 549         if (!shp)
 550                 return -ENOMEM;
 551
 552         shp->shm_perm.key = key;
 553         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 554         shp->mlock_user = NULL;
 555
 556         shp->shm_perm.security = NULL;
 557         error = security_shm_alloc(shp);
 558         if (error) {
 559                 ipc_rcu_putref(shp, ipc_rcu_free);
 560                 return error;
 561         }
 562
 563         sprintf(name, "SYSV%08x", key);
 564         if (shmflg & SHM_HUGETLB) {
 565                 struct hstate *hs;
 566                 size_t hugesize;
 567
 568                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 569                 if (!hs) {
 570                         error = -EINVAL;
 571                         goto no_file;
 572                 }
 573                 hugesize = ALIGN(size, huge_page_size(hs));
 574
 575                 /* hugetlb_file_setup applies strict accounting */
 576                 if (shmflg & SHM_NORESERVE)
 577                         acctflag = VM_NORESERVE;
 578                 file = hugetlb_file_setup(name, hugesize, acctflag,
 579                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 580                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 581         } else {
 582                 /*
 583                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 584                  * if it's asked for.
 585                  */
 586                 if  ((shmflg & SHM_NORESERVE) &&
 587                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 588                         acctflag = VM_NORESERVE;
 589                 file = shmem_kernel_file_setup(name, size, acctflag);
 590         }
 591         error = PTR_ERR(file);
 592         if (IS_ERR(file))
 593                 goto no_file;
 594
 595         shp->shm_cprid = task_tgid_vnr(current);
 596         shp->shm_lprid = 0;
 597         shp->shm_atim = shp->shm_dtim = 0;
 598         shp->shm_ctim = get_seconds();
 599         shp->shm_segsz = size;
 600         shp->shm_nattch = 0;
 601         shp->shm_file = file;
 602         shp->shm_creator = current;
 603
 604         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 605         if (id < 0) {
 606                 error = id;
 607                 goto no_id;
 608         }
 609
 610         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 611
 612         /*
 613          * shmid gets reported as "inode#" in /proc/pid/maps.
 614          * proc-ps tools use this. Changing this will break them.
 615          */
 616         file_inode(file)->i_ino = shp->shm_perm.id;
 617
 618         ns->shm_tot += numpages;
 619         error = shp->shm_perm.id;
 620
 621         ipc_unlock_object(&shp->shm_perm);
 622         rcu_read_unlock();
 623         return error;
 624
 625 no_id:
 626         if (is_file_hugepages(file) && shp->mlock_user)
 627                 user_shm_unlock(size, shp->mlock_user);
 628         fput(file);
 629 no_file:
 630         ipc_rcu_putref(shp, shm_rcu_free);
 631         return error;
 632 }
 633
 634 /*
 635  * Called with shm_ids.rwsem and ipcp locked.
 636  */
 637 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 638 {
 639         struct shmid_kernel *shp;
 640
 641         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 642         return security_shm_associate(shp, shmflg);
 643 }
 644
 645 /*
 646  * Called with shm_ids.rwsem and ipcp locked.
 647  */
 648 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 649                                 struct ipc_params *params)
 650 {
 651         struct shmid_kernel *shp;
 652
 653         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 654         if (shp->shm_segsz < params->u.size)
 655                 return -EINVAL;
 656
 657         return 0;
 658 }
 659
 660 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 661 {
 662         struct ipc_namespace *ns;
 663         static const struct ipc_ops shm_ops = {
 664                 .getnew = newseg,
 665                 .associate = shm_security,
 666                 .more_checks = shm_more_checks,
 667         };
 668         struct ipc_params shm_params;
 669
 670         ns = current->nsproxy->ipc_ns;
 671
 672         shm_params.key = key;
 673         shm_params.flg = shmflg;
 674         shm_params.u.size = size;
 675
 676         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 677 }
 678
 679 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 680 {
 681         switch (version) {
 682         case IPC_64:
 683                 return copy_to_user(buf, in, sizeof(*in));
 684         case IPC_OLD:
 685             {
 686                 struct shmid_ds out;
 687
 688                 memset(&out, 0, sizeof(out));
 689                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 690                 out.shm_segsz   = in->shm_segsz;
 691                 out.shm_atime   = in->shm_atime;
 692                 out.shm_dtime   = in->shm_dtime;
 693                 out.shm_ctime   = in->shm_ctime;
 694                 out.shm_cpid    = in->shm_cpid;
 695                 out.shm_lpid    = in->shm_lpid;
 696                 out.shm_nattch  = in->shm_nattch;
 697
 698                 return copy_to_user(buf, &out, sizeof(out));
 699             }
 700         default:
 701                 return -EINVAL;
 702         }
 703 }
 704
 705 static inline unsigned long
 706 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 707 {
 708         switch (version) {
 709         case IPC_64:
 710                 if (copy_from_user(out, buf, sizeof(*out)))
 711                         return -EFAULT;
 712                 return 0;
 713         case IPC_OLD:
 714             {
 715                 struct shmid_ds tbuf_old;
 716
 717                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 718                         return -EFAULT;
 719
 720                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 721                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 722                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 723
 724                 return 0;
 725             }
 726         default:
 727                 return -EINVAL;
 728         }
 729 }
 730
 731 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 732 {
 733         switch (version) {
 734         case IPC_64:
 735                 return copy_to_user(buf, in, sizeof(*in));
 736         case IPC_OLD:
 737             {
 738                 struct shminfo out;
 739
 740                 if (in->shmmax > INT_MAX)
 741                         out.shmmax = INT_MAX;
 742                 else
 743                         out.shmmax = (int)in->shmmax;
 744
 745                 out.shmmin      = in->shmmin;
 746                 out.shmmni      = in->shmmni;
 747                 out.shmseg      = in->shmseg;
 748                 out.shmall      = in->shmall;
 749
 750                 return copy_to_user(buf, &out, sizeof(out));
 751             }
 752         default:
 753                 return -EINVAL;
 754         }
 755 }
 756
 757 /*
 758  * Calculate and add used RSS and swap pages of a shm.
 759  * Called with shm_ids.rwsem held as a reader
 760  */
 761 static void shm_add_rss_swap(struct shmid_kernel *shp,
 762         unsigned long *rss_add, unsigned long *swp_add)
 763 {
 764         struct inode *inode;
 765
 766         inode = file_inode(shp->shm_file);
 767
 768         if (is_file_hugepages(shp->shm_file)) {
 769                 struct address_space *mapping = inode->i_mapping;
 770                 struct hstate *h = hstate_file(shp->shm_file);
 771                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 772         } else {
 773 #ifdef CONFIG_SHMEM
 774                 struct shmem_inode_info *info = SHMEM_I(inode);
 775                 spin_lock(&info->lock);
 776                 *rss_add += inode->i_mapping->nrpages;
 777                 *swp_add += info->swapped;
 778                 spin_unlock(&info->lock);
 779 #else
 780                 *rss_add += inode->i_mapping->nrpages;
 781 #endif
 782         }
 783 }
 784
 785 /*
 786  * Called with shm_ids.rwsem held as a reader
 787  */
 788 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 789                 unsigned long *swp)
 790 {
 791         int next_id;
 792         int total, in_use;
 793
 794         *rss = 0;
 795         *swp = 0;
 796
 797         in_use = shm_ids(ns).in_use;
 798
 799         for (total = 0, next_id = 0; total < in_use; next_id++) {
 800                 struct kern_ipc_perm *ipc;
 801                 struct shmid_kernel *shp;
 802
 803                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 804                 if (ipc == NULL)
 805                         continue;
 806                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 807
 808                 shm_add_rss_swap(shp, rss, swp);
 809
 810                 total++;
 811         }
 812 }
 813
 814 /*
 815  * This function handles some shmctl commands which require the rwsem
 816  * to be held in write mode.
 817  * NOTE: no locks must be held, the rwsem is taken inside this function.
 818  */
 819 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 820                        struct shmid_ds __user *buf, int version)
 821 {
 822         struct kern_ipc_perm *ipcp;
 823         struct shmid64_ds shmid64;
 824         struct shmid_kernel *shp;
 825         int err;
 826
 827         if (cmd == IPC_SET) {
 828                 if (copy_shmid_from_user(&shmid64, buf, version))
 829                         return -EFAULT;
 830         }
 831
 832         down_write(&shm_ids(ns).rwsem);
 833         rcu_read_lock();
 834
 835         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 836                                       &shmid64.shm_perm, 0);
 837         if (IS_ERR(ipcp)) {
 838                 err = PTR_ERR(ipcp);
 839                 goto out_unlock1;
 840         }
 841
 842         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 843
 844         err = security_shm_shmctl(shp, cmd);
 845         if (err)
 846                 goto out_unlock1;
 847
 848         switch (cmd) {
 849         case IPC_RMID:
 850                 ipc_lock_object(&shp->shm_perm);
 851                 /* do_shm_rmid unlocks the ipc object and rcu */
 852                 do_shm_rmid(ns, ipcp);
 853                 goto out_up;
 854         case IPC_SET:
 855                 ipc_lock_object(&shp->shm_perm);
 856                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
 857                 if (err)
 858                         goto out_unlock0;
 859                 shp->shm_ctim = get_seconds();
 860                 break;
 861         default:
 862                 err = -EINVAL;
 863                 goto out_unlock1;
 864         }
 865
 866 out_unlock0:
 867         ipc_unlock_object(&shp->shm_perm);
 868 out_unlock1:
 869         rcu_read_unlock();
 870 out_up:
 871         up_write(&shm_ids(ns).rwsem);
 872         return err;
 873 }
 874
 875 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
 876                          int cmd, int version, void __user *buf)
 877 {
 878         int err;
 879         struct shmid_kernel *shp;
 880
 881         /* preliminary security checks for *_INFO */
 882         if (cmd == IPC_INFO || cmd == SHM_INFO) {
 883                 err = security_shm_shmctl(NULL, cmd);
 884                 if (err)
 885                         return err;
 886         }
 887
 888         switch (cmd) {
 889         case IPC_INFO:
 890         {
 891                 struct shminfo64 shminfo;
 892
 893                 memset(&shminfo, 0, sizeof(shminfo));
 894                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 895                 shminfo.shmmax = ns->shm_ctlmax;
 896                 shminfo.shmall = ns->shm_ctlall;
 897
 898                 shminfo.shmmin = SHMMIN;
 899                 if (copy_shminfo_to_user(buf, &shminfo, version))
 900                         return -EFAULT;
 901
 902                 down_read(&shm_ids(ns).rwsem);
 903                 err = ipc_get_maxid(&shm_ids(ns));
 904                 up_read(&shm_ids(ns).rwsem);
 905
 906                 if (err < 0)
 907                         err = 0;
 908                 goto out;
 909         }
 910         case SHM_INFO:
 911         {
 912                 struct shm_info shm_info;
 913
 914                 memset(&shm_info, 0, sizeof(shm_info));
 915                 down_read(&shm_ids(ns).rwsem);
 916                 shm_info.used_ids = shm_ids(ns).in_use;
 917                 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
 918                 shm_info.shm_tot = ns->shm_tot;
 919                 shm_info.swap_attempts = 0;
 920                 shm_info.swap_successes = 0;
 921                 err = ipc_get_maxid(&shm_ids(ns));
 922                 up_read(&shm_ids(ns).rwsem);
 923                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 924                         err = -EFAULT;
 925                         goto out;
 926                 }
 927
 928                 err = err < 0 ? 0 : err;
 929                 goto out;
 930         }
 931         case SHM_STAT:
 932         case IPC_STAT:
 933         {
 934                 struct shmid64_ds tbuf;
 935                 int result;
 936
 937                 rcu_read_lock();
 938                 if (cmd == SHM_STAT) {
 939                         shp = shm_obtain_object(ns, shmid);
 940                         if (IS_ERR(shp)) {
 941                                 err = PTR_ERR(shp);
 942                                 goto out_unlock;
 943                         }
 944                         result = shp->shm_perm.id;
 945                 } else {
 946                         shp = shm_obtain_object_check(ns, shmid);
 947                         if (IS_ERR(shp)) {
 948                                 err = PTR_ERR(shp);
 949                                 goto out_unlock;
 950                         }
 951                         result = 0;
 952                 }
 953
 954                 err = -EACCES;
 955                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 956                         goto out_unlock;
 957
 958                 err = security_shm_shmctl(shp, cmd);
 959                 if (err)
 960                         goto out_unlock;
 961
 962                 memset(&tbuf, 0, sizeof(tbuf));
 963                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 964                 tbuf.shm_segsz  = shp->shm_segsz;
 965                 tbuf.shm_atime  = shp->shm_atim;
 966                 tbuf.shm_dtime  = shp->shm_dtim;
 967                 tbuf.shm_ctime  = shp->shm_ctim;
 968                 tbuf.shm_cpid   = shp->shm_cprid;
 969                 tbuf.shm_lpid   = shp->shm_lprid;
 970                 tbuf.shm_nattch = shp->shm_nattch;
 971                 rcu_read_unlock();
 972
 973                 if (copy_shmid_to_user(buf, &tbuf, version))
 974                         err = -EFAULT;
 975                 else
 976                         err = result;
 977                 goto out;
 978         }
 979         default:
 980                 return -EINVAL;
 981         }
 982
 983 out_unlock:
 984         rcu_read_unlock();
 985 out:
 986         return err;
 987 }
 988
 989 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 990 {
 991         struct shmid_kernel *shp;
 992         int err, version;
 993         struct ipc_namespace *ns;
 994
 995         if (cmd < 0 || shmid < 0)
 996                 return -EINVAL;
 997
 998         version = ipc_parse_version(&cmd);
 999         ns = current->nsproxy->ipc_ns;
1000
1001         switch (cmd) {
1002         case IPC_INFO:
1003         case SHM_INFO:
1004         case SHM_STAT:
1005         case IPC_STAT:
1006                 return shmctl_nolock(ns, shmid, cmd, version, buf);
1007         case IPC_RMID:
1008         case IPC_SET:
1009                 return shmctl_down(ns, shmid, cmd, buf, version);
1010         case SHM_LOCK:
1011         case SHM_UNLOCK:
1012         {
1013                 struct file *shm_file;
1014
1015                 rcu_read_lock();
1016                 shp = shm_obtain_object_check(ns, shmid);
1017                 if (IS_ERR(shp)) {
1018                         err = PTR_ERR(shp);
1019                         goto out_unlock1;
1020                 }
1021
1022                 audit_ipc_obj(&(shp->shm_perm));
1023                 err = security_shm_shmctl(shp, cmd);
1024                 if (err)
1025                         goto out_unlock1;
1026
1027                 ipc_lock_object(&shp->shm_perm);
1028
1029                 /* check if shm_destroy() is tearing down shp */
1030                 if (!ipc_valid_object(&shp->shm_perm)) {
1031                         err = -EIDRM;
1032                         goto out_unlock0;
1033                 }
1034
1035                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1036                         kuid_t euid = current_euid();
1037                         if (!uid_eq(euid, shp->shm_perm.uid) &&
1038                             !uid_eq(euid, shp->shm_perm.cuid)) {
1039                                 err = -EPERM;
1040                                 goto out_unlock0;
1041                         }
1042                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1043                                 err = -EPERM;
1044                                 goto out_unlock0;
1045                         }
1046                 }
1047
1048                 shm_file = shp->shm_file;
1049                 if (is_file_hugepages(shm_file))
1050                         goto out_unlock0;
1051
1052                 if (cmd == SHM_LOCK) {
1053                         struct user_struct *user = current_user();
1054                         err = shmem_lock(shm_file, 1, user);
1055                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1056                                 shp->shm_perm.mode |= SHM_LOCKED;
1057                                 shp->mlock_user = user;
1058                         }
1059                         goto out_unlock0;
1060                 }
1061
1062                 /* SHM_UNLOCK */
1063                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1064                         goto out_unlock0;
1065                 shmem_lock(shm_file, 0, shp->mlock_user);
1066                 shp->shm_perm.mode &= ~SHM_LOCKED;
1067                 shp->mlock_user = NULL;
1068                 get_file(shm_file);
1069                 ipc_unlock_object(&shp->shm_perm);
1070                 rcu_read_unlock();
1071                 shmem_unlock_mapping(shm_file->f_mapping);
1072
1073                 fput(shm_file);
1074                 return err;
1075         }
1076         default:
1077                 return -EINVAL;
1078         }
1079
1080 out_unlock0:
1081         ipc_unlock_object(&shp->shm_perm);
1082 out_unlock1:
1083         rcu_read_unlock();
1084         return err;
1085 }
1086
1087 /*
1088  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1089  *
1090  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1091  * "raddr" thing points to kernel space, and there has to be a wrapper around
1092  * this.
1093  */
1094 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1095               ulong *raddr, unsigned long shmlba)
1096 {
1097         struct shmid_kernel *shp;
1098         unsigned long addr;
1099         unsigned long size;
1100         struct file *file;
1101         int    err;
1102         unsigned long flags;
1103         unsigned long prot;
1104         int acc_mode;
1105         struct ipc_namespace *ns;
1106         struct shm_file_data *sfd;
1107         struct path path;
1108         fmode_t f_mode;
1109         unsigned long populate = 0;
1110
1111         err = -EINVAL;
1112         if (shmid < 0)
1113                 goto out;
1114         else if ((addr = (ulong)shmaddr)) {
1115                 if (addr & (shmlba - 1)) {
1116                         if (shmflg & SHM_RND) {
1117                                 addr &= ~(shmlba - 1);  /* round down */
1118
1119                                 /*
1120                                  * Ensure that the round-down is non-nil
1121                                  * when remapping. This can happen for
1122                                  * cases when addr < shmlba.
1123                                  */
1124                                 if (!addr && (shmflg & SHM_REMAP))
1125                                         goto out;
1126                         } else
1127 #ifndef __ARCH_FORCE_SHMLBA
1128                                 if (addr & ~PAGE_MASK)
1129 #endif
1130                                         goto out;
1131                 }
1132                 flags = MAP_SHARED | MAP_FIXED;
1133         } else {
1134                 if ((shmflg & SHM_REMAP))
1135                         goto out;
1136
1137                 flags = MAP_SHARED;
1138         }
1139
1140         if (shmflg & SHM_RDONLY) {
1141                 prot = PROT_READ;
1142                 acc_mode = S_IRUGO;
1143                 f_mode = FMODE_READ;
1144         } else {
1145                 prot = PROT_READ | PROT_WRITE;
1146                 acc_mode = S_IRUGO | S_IWUGO;
1147                 f_mode = FMODE_READ | FMODE_WRITE;
1148         }
1149         if (shmflg & SHM_EXEC) {
1150                 prot |= PROT_EXEC;
1151                 acc_mode |= S_IXUGO;
1152         }
1153
1154         /*
1155          * We cannot rely on the fs check since SYSV IPC does have an
1156          * additional creator id...
1157          */
1158         ns = current->nsproxy->ipc_ns;
1159         rcu_read_lock();
1160         shp = shm_obtain_object_check(ns, shmid);
1161         if (IS_ERR(shp)) {
1162                 err = PTR_ERR(shp);
1163                 goto out_unlock;
1164         }
1165
1166         err = -EACCES;
1167         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1168                 goto out_unlock;
1169
1170         err = security_shm_shmat(shp, shmaddr, shmflg);
1171         if (err)
1172                 goto out_unlock;
1173
1174         ipc_lock_object(&shp->shm_perm);
1175
1176         /* check if shm_destroy() is tearing down shp */
1177         if (!ipc_valid_object(&shp->shm_perm)) {
1178                 ipc_unlock_object(&shp->shm_perm);
1179                 err = -EIDRM;
1180                 goto out_unlock;
1181         }
1182
1183         path = shp->shm_file->f_path;
1184         path_get(&path);
1185         shp->shm_nattch++;
1186         size = i_size_read(d_inode(path.dentry));
1187         ipc_unlock_object(&shp->shm_perm);
1188         rcu_read_unlock();
1189
1190         err = -ENOMEM;
1191         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1192         if (!sfd) {
1193                 path_put(&path);
1194                 goto out_nattch;
1195         }
1196
1197         file = alloc_file(&path, f_mode,
1198                           is_file_hugepages(shp->shm_file) ?
1199                                 &shm_file_operations_huge :
1200                                 &shm_file_operations);
1201         err = PTR_ERR(file);
1202         if (IS_ERR(file)) {
1203                 kfree(sfd);
1204                 path_put(&path);
1205                 goto out_nattch;
1206         }
1207
1208         file->private_data = sfd;
1209         file->f_mapping = shp->shm_file->f_mapping;
1210         sfd->id = shp->shm_perm.id;
1211         sfd->ns = get_ipc_ns(ns);
1212         /*
1213          * We need to take a reference to the real shm file to prevent the
1214          * pointer from becoming stale in cases where the lifetime of the outer
1215          * file extends beyond that of the shm segment.  It's not usually
1216          * possible, but it can happen during remap_file_pages() emulation as
1217          * that unmaps the memory, then does ->mmap() via file reference only.
1218          * We'll deny the ->mmap() if the shm segment was since removed, but to
1219          * detect shm ID reuse we need to compare the file pointers.
1220          */
1221         sfd->file = get_file(shp->shm_file);
1222         sfd->vm_ops = NULL;
1223
1224         err = security_mmap_file(file, prot, flags);
1225         if (err)
1226                 goto out_fput;
1227
1228         down_write(&current->mm->mmap_sem);
1229         if (addr && !(shmflg & SHM_REMAP)) {
1230                 err = -EINVAL;
1231                 if (addr + size < addr)
1232                         goto invalid;
1233
1234                 if (find_vma_intersection(current->mm, addr, addr + size))
1235                         goto invalid;
1236         }
1237
1238         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1239         *raddr = addr;
1240         err = 0;
1241         if (IS_ERR_VALUE(addr))
1242                 err = (long)addr;
1243 invalid:
1244         up_write(&current->mm->mmap_sem);
1245         if (populate)
1246                 mm_populate(addr, populate);
1247
1248 out_fput:
1249         fput(file);
1250
1251 out_nattch:
1252         down_write(&shm_ids(ns).rwsem);
1253         shp = shm_lock(ns, shmid);
1254         shp->shm_nattch--;
1255         if (shm_may_destroy(ns, shp))
1256                 shm_destroy(ns, shp);
1257         else
1258                 shm_unlock(shp);
1259         up_write(&shm_ids(ns).rwsem);
1260         return err;
1261
1262 out_unlock:
1263         rcu_read_unlock();
1264 out:
1265         return err;
1266 }
1267
1268 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1269 {
1270         unsigned long ret;
1271         long err;
1272
1273         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1274         if (err)
1275                 return err;
1276         force_successful_syscall_return();
1277         return (long)ret;
1278 }
1279
1280 /*
1281  * detach and kill segment if marked destroyed.
1282  * The work is done in shm_close.
1283  */
1284 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1285 {
1286         struct mm_struct *mm = current->mm;
1287         struct vm_area_struct *vma;
1288         unsigned long addr = (unsigned long)shmaddr;
1289         int retval = -EINVAL;
1290 #ifdef CONFIG_MMU
1291         loff_t size = 0;
1292         struct file *file;
1293         struct vm_area_struct *next;
1294 #endif
1295
1296         if (addr & ~PAGE_MASK)
1297                 return retval;
1298
1299         down_write(&mm->mmap_sem);
1300
1301         /*
1302          * This function tries to be smart and unmap shm segments that
1303          * were modified by partial mlock or munmap calls:
1304          * - It first determines the size of the shm segment that should be
1305          *   unmapped: It searches for a vma that is backed by shm and that
1306          *   started at address shmaddr. It records it's size and then unmaps
1307          *   it.
1308          * - Then it unmaps all shm vmas that started at shmaddr and that
1309          *   are within the initially determined size and that are from the
1310          *   same shm segment from which we determined the size.
1311          * Errors from do_munmap are ignored: the function only fails if
1312          * it's called with invalid parameters or if it's called to unmap
1313          * a part of a vma. Both calls in this function are for full vmas,
1314          * the parameters are directly copied from the vma itself and always
1315          * valid - therefore do_munmap cannot fail. (famous last words?)
1316          */
1317         /*
1318          * If it had been mremap()'d, the starting address would not
1319          * match the usual checks anyway. So assume all vma's are
1320          * above the starting address given.
1321          */
1322         vma = find_vma(mm, addr);
1323
1324 #ifdef CONFIG_MMU
1325         while (vma) {
1326                 next = vma->vm_next;
1327
1328                 /*
1329                  * Check if the starting address would match, i.e. it's
1330                  * a fragment created by mprotect() and/or munmap(), or it
1331                  * otherwise it starts at this address with no hassles.
1332                  */
1333                 if ((vma->vm_ops == &shm_vm_ops) &&
1334                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1335
1336                         /*
1337                          * Record the file of the shm segment being
1338                          * unmapped.  With mremap(), someone could place
1339                          * page from another segment but with equal offsets
1340                          * in the range we are unmapping.
1341                          */
1342                         file = vma->vm_file;
1343                         size = i_size_read(file_inode(vma->vm_file));
1344                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1345                         /*
1346                          * We discovered the size of the shm segment, so
1347                          * break out of here and fall through to the next
1348                          * loop that uses the size information to stop
1349                          * searching for matching vma's.
1350                          */
1351                         retval = 0;
1352                         vma = next;
1353                         break;
1354                 }
1355                 vma = next;
1356         }
1357
1358         /*
1359          * We need look no further than the maximum address a fragment
1360          * could possibly have landed at. Also cast things to loff_t to
1361          * prevent overflows and make comparisons vs. equal-width types.
1362          */
1363         size = PAGE_ALIGN(size);
1364         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1365                 next = vma->vm_next;
1366
1367                 /* finding a matching vma now does not alter retval */
1368                 if ((vma->vm_ops == &shm_vm_ops) &&
1369                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1370                     (vma->vm_file == file))
1371                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1372                 vma = next;
1373         }
1374
1375 #else /* CONFIG_MMU */
1376         /* under NOMMU conditions, the exact address to be destroyed must be
1377          * given */
1378         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1379                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1380                 retval = 0;
1381         }
1382
1383 #endif
1384
1385         up_write(&mm->mmap_sem);
1386         return retval;
1387 }
1388
1389 #ifdef CONFIG_PROC_FS
1390 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1391 {
1392         struct user_namespace *user_ns = seq_user_ns(s);
1393         struct shmid_kernel *shp = it;
1394         unsigned long rss = 0, swp = 0;
1395
1396         shm_add_rss_swap(shp, &rss, &swp);
1397
1398 #if BITS_PER_LONG <= 32
1399 #define SIZE_SPEC "%10lu"
1400 #else
1401 #define SIZE_SPEC "%21lu"
1402 #endif
1403
1404         seq_printf(s,
1405                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1406                    "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1407                    SIZE_SPEC " " SIZE_SPEC "\n",
1408                    shp->shm_perm.key,
1409                    shp->shm_perm.id,
1410                    shp->shm_perm.mode,
1411                    shp->shm_segsz,
1412                    shp->shm_cprid,
1413                    shp->shm_lprid,
1414                    shp->shm_nattch,
1415                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1416                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1417                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1418                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1419                    shp->shm_atim,
1420                    shp->shm_dtim,
1421                    shp->shm_ctim,
1422                    rss * PAGE_SIZE,
1423                    swp * PAGE_SIZE);
1424
1425         return 0;
1426 }
1427 #endif