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
  93         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  94
  95         if (shp->shm_nattch) {
  96                 shp->shm_perm.mode |= SHM_DEST;
  97                 /* Do not find it any more */
  98                 shp->shm_perm.key = IPC_PRIVATE;
  99                 shm_unlock(shp);
 100         } else
 101                 shm_destroy(ns, shp);
 102 }
 103
 104 #ifdef CONFIG_IPC_NS
 105 void shm_exit_ns(struct ipc_namespace *ns)
 106 {
 107         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 108         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 109 }
 110 #endif
 111
 112 static int __init ipc_ns_init(void)
 113 {
 114         shm_init_ns(&init_ipc_ns);
 115         return 0;
 116 }
 117
 118 pure_initcall(ipc_ns_init);
 119
 120 void __init shm_init(void)
 121 {
 122         ipc_init_proc_interface("sysvipc/shm",
 123 #if BITS_PER_LONG <= 32
 124                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 125 #else
 126                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 127 #endif
 128                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 129 }
 130
 131 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 132 {
 133         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 134
 135         if (IS_ERR(ipcp))
 136                 return ERR_CAST(ipcp);
 137
 138         return container_of(ipcp, struct shmid_kernel, shm_perm);
 139 }
 140
 141 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 142 {
 143         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 144
 145         if (IS_ERR(ipcp))
 146                 return ERR_CAST(ipcp);
 147
 148         return container_of(ipcp, struct shmid_kernel, shm_perm);
 149 }
 150
 151 /*
 152  * shm_lock_(check_) routines are called in the paths where the rwsem
 153  * is not necessarily held.
 154  */
 155 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 156 {
 157         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 158
 159         /*
 160          * Callers of shm_lock() must validate the status of the returned ipc
 161          * object pointer (as returned by ipc_lock()), and error out as
 162          * appropriate.
 163          */
 164         if (IS_ERR(ipcp))
 165                 return (void *)ipcp;
 166         return container_of(ipcp, struct shmid_kernel, shm_perm);
 167 }
 168
 169 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 170 {
 171         rcu_read_lock();
 172         ipc_lock_object(&ipcp->shm_perm);
 173 }
 174
 175 static void __shm_free(struct shmid_kernel *shp)
 176 {
 177         kvfree(shp);
 178 }
 179
 180 static void shm_rcu_free(struct rcu_head *head)
 181 {
 182         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 183                                                         rcu);
 184         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 185                                                         shm_perm);
 186         security_shm_free(shp);
 187         __shm_free(shp);
 188 }
 189
 190 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 191 {
 192         list_del(&s->shm_clist);
 193         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 194 }
 195
 196
 197 static int __shm_open(struct vm_area_struct *vma)
 198 {
 199         struct file *file = vma->vm_file;
 200         struct shm_file_data *sfd = shm_file_data(file);
 201         struct shmid_kernel *shp;
 202
 203         shp = shm_lock(sfd->ns, sfd->id);
 204
 205         if (IS_ERR(shp))
 206                 return PTR_ERR(shp);
 207
 208         shp->shm_atim = get_seconds();
 209         shp->shm_lprid = task_tgid_vnr(current);
 210         shp->shm_nattch++;
 211         shm_unlock(shp);
 212         return 0;
 213 }
 214
 215 /* This is called by fork, once for every shm attach. */
 216 static void shm_open(struct vm_area_struct *vma)
 217 {
 218         int err = __shm_open(vma);
 219         /*
 220          * We raced in the idr lookup or with shm_destroy().
 221          * Either way, the ID is busted.
 222          */
 223         WARN_ON_ONCE(err);
 224 }
 225
 226 /*
 227  * shm_destroy - free the struct shmid_kernel
 228  *
 229  * @ns: namespace
 230  * @shp: struct to free
 231  *
 232  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 233  * but returns with shp unlocked and freed.
 234  */
 235 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 236 {
 237         struct file *shm_file;
 238
 239         shm_file = shp->shm_file;
 240         shp->shm_file = NULL;
 241         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 242         shm_rmid(ns, shp);
 243         shm_unlock(shp);
 244         if (!is_file_hugepages(shm_file))
 245                 shmem_lock(shm_file, 0, shp->mlock_user);
 246         else if (shp->mlock_user)
 247                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 248                                 shp->mlock_user);
 249         fput(shm_file);
 250         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 251 }
 252
 253 /*
 254  * shm_may_destroy - identifies whether shm segment should be destroyed now
 255  *
 256  * Returns true if and only if there are no active users of the segment and
 257  * one of the following is true:
 258  *
 259  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 260  *
 261  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 262  */
 263 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 264 {
 265         return (shp->shm_nattch == 0) &&
 266                (ns->shm_rmid_forced ||
 267                 (shp->shm_perm.mode & SHM_DEST));
 268 }
 269
 270 /*
 271  * remove the attach descriptor vma.
 272  * free memory for segment if it is marked destroyed.
 273  * The descriptor has already been removed from the current->mm->mmap list
 274  * and will later be kfree()d.
 275  */
 276 static void shm_close(struct vm_area_struct *vma)
 277 {
 278         struct file *file = vma->vm_file;
 279         struct shm_file_data *sfd = shm_file_data(file);
 280         struct shmid_kernel *shp;
 281         struct ipc_namespace *ns = sfd->ns;
 282
 283         down_write(&shm_ids(ns).rwsem);
 284         /* remove from the list of attaches of the shm segment */
 285         shp = shm_lock(ns, sfd->id);
 286
 287         /*
 288          * We raced in the idr lookup or with shm_destroy().
 289          * Either way, the ID is busted.
 290          */
 291         if (WARN_ON_ONCE(IS_ERR(shp)))
 292                 goto done; /* no-op */
 293
 294         shp->shm_lprid = task_tgid_vnr(current);
 295         shp->shm_dtim = get_seconds();
 296         shp->shm_nattch--;
 297         if (shm_may_destroy(ns, shp))
 298                 shm_destroy(ns, shp);
 299         else
 300                 shm_unlock(shp);
 301 done:
 302         up_write(&shm_ids(ns).rwsem);
 303 }
 304
 305 /* Called with ns->shm_ids(ns).rwsem locked */
 306 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 307 {
 308         struct ipc_namespace *ns = data;
 309         struct kern_ipc_perm *ipcp = p;
 310         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 311
 312         /*
 313          * We want to destroy segments without users and with already
 314          * exit'ed originating process.
 315          *
 316          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 317          */
 318         if (shp->shm_creator != NULL)
 319                 return 0;
 320
 321         if (shm_may_destroy(ns, shp)) {
 322                 shm_lock_by_ptr(shp);
 323                 shm_destroy(ns, shp);
 324         }
 325         return 0;
 326 }
 327
 328 void shm_destroy_orphaned(struct ipc_namespace *ns)
 329 {
 330         down_write(&shm_ids(ns).rwsem);
 331         if (shm_ids(ns).in_use)
 332                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 333         up_write(&shm_ids(ns).rwsem);
 334 }
 335
 336 /* Locking assumes this will only be called with task == current */
 337 void exit_shm(struct task_struct *task)
 338 {
 339         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 340         struct shmid_kernel *shp, *n;
 341
 342         if (list_empty(&task->sysvshm.shm_clist))
 343                 return;
 344
 345         /*
 346          * If kernel.shm_rmid_forced is not set then only keep track of
 347          * which shmids are orphaned, so that a later set of the sysctl
 348          * can clean them up.
 349          */
 350         if (!ns->shm_rmid_forced) {
 351                 down_read(&shm_ids(ns).rwsem);
 352                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
 353                         shp->shm_creator = NULL;
 354                 /*
 355                  * Only under read lock but we are only called on current
 356                  * so no entry on the list will be shared.
 357                  */
 358                 list_del(&task->sysvshm.shm_clist);
 359                 up_read(&shm_ids(ns).rwsem);
 360                 return;
 361         }
 362
 363         /*
 364          * Destroy all already created segments, that were not yet mapped,
 365          * and mark any mapped as orphan to cover the sysctl toggling.
 366          * Destroy is skipped if shm_may_destroy() returns false.
 367          */
 368         down_write(&shm_ids(ns).rwsem);
 369         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
 370                 shp->shm_creator = NULL;
 371
 372                 if (shm_may_destroy(ns, shp)) {
 373                         shm_lock_by_ptr(shp);
 374                         shm_destroy(ns, shp);
 375                 }
 376         }
 377
 378         /* Remove the list head from any segments still attached. */
 379         list_del(&task->sysvshm.shm_clist);
 380         up_write(&shm_ids(ns).rwsem);
 381 }
 382
 383 static int shm_fault(struct vm_fault *vmf)
 384 {
 385         struct file *file = vmf->vma->vm_file;
 386         struct shm_file_data *sfd = shm_file_data(file);
 387
 388         return sfd->vm_ops->fault(vmf);
 389 }
 390
 391 #ifdef CONFIG_NUMA
 392 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 393 {
 394         struct file *file = vma->vm_file;
 395         struct shm_file_data *sfd = shm_file_data(file);
 396         int err = 0;
 397
 398         if (sfd->vm_ops->set_policy)
 399                 err = sfd->vm_ops->set_policy(vma, new);
 400         return err;
 401 }
 402
 403 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 404                                         unsigned long addr)
 405 {
 406         struct file *file = vma->vm_file;
 407         struct shm_file_data *sfd = shm_file_data(file);
 408         struct mempolicy *pol = NULL;
 409
 410         if (sfd->vm_ops->get_policy)
 411                 pol = sfd->vm_ops->get_policy(vma, addr);
 412         else if (vma->vm_policy)
 413                 pol = vma->vm_policy;
 414
 415         return pol;
 416 }
 417 #endif
 418
 419 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 420 {
 421         struct shm_file_data *sfd = shm_file_data(file);
 422         int ret;
 423
 424         /*
 425          * In case of remap_file_pages() emulation, the file can represent
 426          * removed IPC ID: propogate shm_lock() error to caller.
 427          */
 428         ret = __shm_open(vma);
 429         if (ret)
 430                 return ret;
 431
 432         ret = call_mmap(sfd->file, vma);
 433         if (ret) {
 434                 shm_close(vma);
 435                 return ret;
 436         }
 437         sfd->vm_ops = vma->vm_ops;
 438 #ifdef CONFIG_MMU
 439         WARN_ON(!sfd->vm_ops->fault);
 440 #endif
 441         vma->vm_ops = &shm_vm_ops;
 442         return 0;
 443 }
 444
 445 static int shm_release(struct inode *ino, struct file *file)
 446 {
 447         struct shm_file_data *sfd = shm_file_data(file);
 448
 449         put_ipc_ns(sfd->ns);
 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
 480         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 481                                                 pgoff, flags);
 482 }
 483
 484 static const struct file_operations shm_file_operations = {
 485         .mmap           = shm_mmap,
 486         .fsync          = shm_fsync,
 487         .release        = shm_release,
 488         .get_unmapped_area      = shm_get_unmapped_area,
 489         .llseek         = noop_llseek,
 490         .fallocate      = shm_fallocate,
 491 };
 492
 493 /*
 494  * shm_file_operations_huge is now identical to shm_file_operations,
 495  * but we keep it distinct for the sake of is_file_shm_hugepages().
 496  */
 497 static const struct file_operations shm_file_operations_huge = {
 498         .mmap           = shm_mmap,
 499         .fsync          = shm_fsync,
 500         .release        = shm_release,
 501         .get_unmapped_area      = shm_get_unmapped_area,
 502         .llseek         = noop_llseek,
 503         .fallocate      = shm_fallocate,
 504 };
 505
 506 bool is_file_shm_hugepages(struct file *file)
 507 {
 508         return file->f_op == &shm_file_operations_huge;
 509 }
 510
 511 static const struct vm_operations_struct shm_vm_ops = {
 512         .open   = shm_open,     /* callback for a new vm-area open */
 513         .close  = shm_close,    /* callback for when the vm-area is released */
 514         .fault  = shm_fault,
 515 #if defined(CONFIG_NUMA)
 516         .set_policy = shm_set_policy,
 517         .get_policy = shm_get_policy,
 518 #endif
 519 };
 520
 521 static struct shmid_kernel *shm_alloc(void)
 522 {
 523         struct shmid_kernel *shp;
 524
 525         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
 526         if (unlikely(!shp))
 527                 return NULL;
 528
 529         return shp;
 530 }
 531
 532 /**
 533  * newseg - Create a new shared memory segment
 534  * @ns: namespace
 535  * @params: ptr to the structure that contains key, size and shmflg
 536  *
 537  * Called with shm_ids.rwsem held as a writer.
 538  */
 539 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 540 {
 541         key_t key = params->key;
 542         int shmflg = params->flg;
 543         size_t size = params->u.size;
 544         int error;
 545         struct shmid_kernel *shp;
 546         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 547         struct file *file;
 548         char name[13];
 549         vm_flags_t acctflag = 0;
 550
 551         if (size < SHMMIN || size > ns->shm_ctlmax)
 552                 return -EINVAL;
 553
 554         if (numpages << PAGE_SHIFT < size)
 555                 return -ENOSPC;
 556
 557         if (ns->shm_tot + numpages < ns->shm_tot ||
 558                         ns->shm_tot + numpages > ns->shm_ctlall)
 559                 return -ENOSPC;
 560
 561         shp = shm_alloc();
 562         if (!shp)
 563                 return -ENOMEM;
 564
 565         shp->shm_perm.key = key;
 566         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 567         shp->mlock_user = NULL;
 568
 569         shp->shm_perm.security = NULL;
 570         error = security_shm_alloc(shp);
 571         if (error) {
 572                 __shm_free(shp);
 573                 return error;
 574         }
 575
 576         sprintf(name, "SYSV%08x", key);
 577         if (shmflg & SHM_HUGETLB) {
 578                 struct hstate *hs;
 579                 size_t hugesize;
 580
 581                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 582                 if (!hs) {
 583                         error = -EINVAL;
 584                         goto no_file;
 585                 }
 586                 hugesize = ALIGN(size, huge_page_size(hs));
 587
 588                 /* hugetlb_file_setup applies strict accounting */
 589                 if (shmflg & SHM_NORESERVE)
 590                         acctflag = VM_NORESERVE;
 591                 file = hugetlb_file_setup(name, hugesize, acctflag,
 592                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 593                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 594         } else {
 595                 /*
 596                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 597                  * if it's asked for.
 598                  */
 599                 if  ((shmflg & SHM_NORESERVE) &&
 600                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 601                         acctflag = VM_NORESERVE;
 602                 file = shmem_kernel_file_setup(name, size, acctflag);
 603         }
 604         error = PTR_ERR(file);
 605         if (IS_ERR(file))
 606                 goto no_file;
 607
 608         shp->shm_cprid = task_tgid_vnr(current);
 609         shp->shm_lprid = 0;
 610         shp->shm_atim = shp->shm_dtim = 0;
 611         shp->shm_ctim = get_seconds();
 612         shp->shm_segsz = size;
 613         shp->shm_nattch = 0;
 614         shp->shm_file = file;
 615         shp->shm_creator = current;
 616
 617         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 618         if (error < 0)
 619                 goto no_id;
 620
 621         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 622
 623         /*
 624          * shmid gets reported as "inode#" in /proc/pid/maps.
 625          * proc-ps tools use this. Changing this will break them.
 626          */
 627         file_inode(file)->i_ino = shp->shm_perm.id;
 628
 629         ns->shm_tot += numpages;
 630         error = shp->shm_perm.id;
 631
 632         ipc_unlock_object(&shp->shm_perm);
 633         rcu_read_unlock();
 634         return error;
 635
 636 no_id:
 637         if (is_file_hugepages(file) && shp->mlock_user)
 638                 user_shm_unlock(size, shp->mlock_user);
 639         fput(file);
 640 no_file:
 641         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 642         return error;
 643 }
 644
 645 /*
 646  * Called with shm_ids.rwsem and ipcp locked.
 647  */
 648 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 649 {
 650         struct shmid_kernel *shp;
 651
 652         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 653         return security_shm_associate(shp, shmflg);
 654 }
 655
 656 /*
 657  * Called with shm_ids.rwsem and ipcp locked.
 658  */
 659 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 660                                 struct ipc_params *params)
 661 {
 662         struct shmid_kernel *shp;
 663
 664         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 665         if (shp->shm_segsz < params->u.size)
 666                 return -EINVAL;
 667
 668         return 0;
 669 }
 670
 671 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 672 {
 673         struct ipc_namespace *ns;
 674         static const struct ipc_ops shm_ops = {
 675                 .getnew = newseg,
 676                 .associate = shm_security,
 677                 .more_checks = shm_more_checks,
 678         };
 679         struct ipc_params shm_params;
 680
 681         ns = current->nsproxy->ipc_ns;
 682
 683         shm_params.key = key;
 684         shm_params.flg = shmflg;
 685         shm_params.u.size = size;
 686
 687         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 688 }
 689
 690 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 691 {
 692         switch (version) {
 693         case IPC_64:
 694                 return copy_to_user(buf, in, sizeof(*in));
 695         case IPC_OLD:
 696             {
 697                 struct shmid_ds out;
 698
 699                 memset(&out, 0, sizeof(out));
 700                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 701                 out.shm_segsz   = in->shm_segsz;
 702                 out.shm_atime   = in->shm_atime;
 703                 out.shm_dtime   = in->shm_dtime;
 704                 out.shm_ctime   = in->shm_ctime;
 705                 out.shm_cpid    = in->shm_cpid;
 706                 out.shm_lpid    = in->shm_lpid;
 707                 out.shm_nattch  = in->shm_nattch;
 708
 709                 return copy_to_user(buf, &out, sizeof(out));
 710             }
 711         default:
 712                 return -EINVAL;
 713         }
 714 }
 715
 716 static inline unsigned long
 717 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 718 {
 719         switch (version) {
 720         case IPC_64:
 721                 if (copy_from_user(out, buf, sizeof(*out)))
 722                         return -EFAULT;
 723                 return 0;
 724         case IPC_OLD:
 725             {
 726                 struct shmid_ds tbuf_old;
 727
 728                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 729                         return -EFAULT;
 730
 731                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 732                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 733                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 734
 735                 return 0;
 736             }
 737         default:
 738                 return -EINVAL;
 739         }
 740 }
 741
 742 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 743 {
 744         switch (version) {
 745         case IPC_64:
 746                 return copy_to_user(buf, in, sizeof(*in));
 747         case IPC_OLD:
 748             {
 749                 struct shminfo out;
 750
 751                 if (in->shmmax > INT_MAX)
 752                         out.shmmax = INT_MAX;
 753                 else
 754                         out.shmmax = (int)in->shmmax;
 755
 756                 out.shmmin      = in->shmmin;
 757                 out.shmmni      = in->shmmni;
 758                 out.shmseg      = in->shmseg;
 759                 out.shmall      = in->shmall;
 760
 761                 return copy_to_user(buf, &out, sizeof(out));
 762             }
 763         default:
 764                 return -EINVAL;
 765         }
 766 }
 767
 768 /*
 769  * Calculate and add used RSS and swap pages of a shm.
 770  * Called with shm_ids.rwsem held as a reader
 771  */
 772 static void shm_add_rss_swap(struct shmid_kernel *shp,
 773         unsigned long *rss_add, unsigned long *swp_add)
 774 {
 775         struct inode *inode;
 776
 777         inode = file_inode(shp->shm_file);
 778
 779         if (is_file_hugepages(shp->shm_file)) {
 780                 struct address_space *mapping = inode->i_mapping;
 781                 struct hstate *h = hstate_file(shp->shm_file);
 782                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 783         } else {
 784 #ifdef CONFIG_SHMEM
 785                 struct shmem_inode_info *info = SHMEM_I(inode);
 786
 787                 spin_lock_irq(&info->lock);
 788                 *rss_add += inode->i_mapping->nrpages;
 789                 *swp_add += info->swapped;
 790                 spin_unlock_irq(&info->lock);
 791 #else
 792                 *rss_add += inode->i_mapping->nrpages;
 793 #endif
 794         }
 795 }
 796
 797 /*
 798  * Called with shm_ids.rwsem held as a reader
 799  */
 800 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 801                 unsigned long *swp)
 802 {
 803         int next_id;
 804         int total, in_use;
 805
 806         *rss = 0;
 807         *swp = 0;
 808
 809         in_use = shm_ids(ns).in_use;
 810
 811         for (total = 0, next_id = 0; total < in_use; next_id++) {
 812                 struct kern_ipc_perm *ipc;
 813                 struct shmid_kernel *shp;
 814
 815                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 816                 if (ipc == NULL)
 817                         continue;
 818                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 819
 820                 shm_add_rss_swap(shp, rss, swp);
 821
 822                 total++;
 823         }
 824 }
 825
 826 /*
 827  * This function handles some shmctl commands which require the rwsem
 828  * to be held in write mode.
 829  * NOTE: no locks must be held, the rwsem is taken inside this function.
 830  */
 831 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 832                        struct shmid_ds __user *buf, int version)
 833 {
 834         struct kern_ipc_perm *ipcp;
 835         struct shmid64_ds shmid64;
 836         struct shmid_kernel *shp;
 837         int err;
 838
 839         if (cmd == IPC_SET) {
 840                 if (copy_shmid_from_user(&shmid64, buf, version))
 841                         return -EFAULT;
 842         }
 843
 844         down_write(&shm_ids(ns).rwsem);
 845         rcu_read_lock();
 846
 847         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 848                                       &shmid64.shm_perm, 0);
 849         if (IS_ERR(ipcp)) {
 850                 err = PTR_ERR(ipcp);
 851                 goto out_unlock1;
 852         }
 853
 854         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 855
 856         err = security_shm_shmctl(shp, cmd);
 857         if (err)
 858                 goto out_unlock1;
 859
 860         switch (cmd) {
 861         case IPC_RMID:
 862                 ipc_lock_object(&shp->shm_perm);
 863                 /* do_shm_rmid unlocks the ipc object and rcu */
 864                 do_shm_rmid(ns, ipcp);
 865                 goto out_up;
 866         case IPC_SET:
 867                 ipc_lock_object(&shp->shm_perm);
 868                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
 869                 if (err)
 870                         goto out_unlock0;
 871                 shp->shm_ctim = get_seconds();
 872                 break;
 873         default:
 874                 err = -EINVAL;
 875                 goto out_unlock1;
 876         }
 877
 878 out_unlock0:
 879         ipc_unlock_object(&shp->shm_perm);
 880 out_unlock1:
 881         rcu_read_unlock();
 882 out_up:
 883         up_write(&shm_ids(ns).rwsem);
 884         return err;
 885 }
 886
 887 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
 888                          int cmd, int version, void __user *buf)
 889 {
 890         int err;
 891         struct shmid_kernel *shp;
 892
 893         /* preliminary security checks for *_INFO */
 894         if (cmd == IPC_INFO || cmd == SHM_INFO) {
 895                 err = security_shm_shmctl(NULL, cmd);
 896                 if (err)
 897                         return err;
 898         }
 899
 900         switch (cmd) {
 901         case IPC_INFO:
 902         {
 903                 struct shminfo64 shminfo;
 904
 905                 memset(&shminfo, 0, sizeof(shminfo));
 906                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 907                 shminfo.shmmax = ns->shm_ctlmax;
 908                 shminfo.shmall = ns->shm_ctlall;
 909
 910                 shminfo.shmmin = SHMMIN;
 911                 if (copy_shminfo_to_user(buf, &shminfo, version))
 912                         return -EFAULT;
 913
 914                 down_read(&shm_ids(ns).rwsem);
 915                 err = ipc_get_maxid(&shm_ids(ns));
 916                 up_read(&shm_ids(ns).rwsem);
 917
 918                 if (err < 0)
 919                         err = 0;
 920                 goto out;
 921         }
 922         case SHM_INFO:
 923         {
 924                 struct shm_info shm_info;
 925
 926                 memset(&shm_info, 0, sizeof(shm_info));
 927                 down_read(&shm_ids(ns).rwsem);
 928                 shm_info.used_ids = shm_ids(ns).in_use;
 929                 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
 930                 shm_info.shm_tot = ns->shm_tot;
 931                 shm_info.swap_attempts = 0;
 932                 shm_info.swap_successes = 0;
 933                 err = ipc_get_maxid(&shm_ids(ns));
 934                 up_read(&shm_ids(ns).rwsem);
 935                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 936                         err = -EFAULT;
 937                         goto out;
 938                 }
 939
 940                 err = err < 0 ? 0 : err;
 941                 goto out;
 942         }
 943         case SHM_STAT:
 944         case IPC_STAT:
 945         {
 946                 struct shmid64_ds tbuf;
 947                 int result;
 948
 949                 rcu_read_lock();
 950                 if (cmd == SHM_STAT) {
 951                         shp = shm_obtain_object(ns, shmid);
 952                         if (IS_ERR(shp)) {
 953                                 err = PTR_ERR(shp);
 954                                 goto out_unlock;
 955                         }
 956                         result = shp->shm_perm.id;
 957                 } else {
 958                         shp = shm_obtain_object_check(ns, shmid);
 959                         if (IS_ERR(shp)) {
 960                                 err = PTR_ERR(shp);
 961                                 goto out_unlock;
 962                         }
 963                         result = 0;
 964                 }
 965
 966                 err = -EACCES;
 967                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 968                         goto out_unlock;
 969
 970                 err = security_shm_shmctl(shp, cmd);
 971                 if (err)
 972                         goto out_unlock;
 973
 974                 memset(&tbuf, 0, sizeof(tbuf));
 975                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 976                 tbuf.shm_segsz  = shp->shm_segsz;
 977                 tbuf.shm_atime  = shp->shm_atim;
 978                 tbuf.shm_dtime  = shp->shm_dtim;
 979                 tbuf.shm_ctime  = shp->shm_ctim;
 980                 tbuf.shm_cpid   = shp->shm_cprid;
 981                 tbuf.shm_lpid   = shp->shm_lprid;
 982                 tbuf.shm_nattch = shp->shm_nattch;
 983                 rcu_read_unlock();
 984
 985                 if (copy_shmid_to_user(buf, &tbuf, version))
 986                         err = -EFAULT;
 987                 else
 988                         err = result;
 989                 goto out;
 990         }
 991         default:
 992                 return -EINVAL;
 993         }
 994
 995 out_unlock:
 996         rcu_read_unlock();
 997 out:
 998         return err;
 999 }
1000
1001 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1002 {
1003         struct shmid_kernel *shp;
1004         int err, version;
1005         struct ipc_namespace *ns;
1006
1007         if (cmd < 0 || shmid < 0)
1008                 return -EINVAL;
1009
1010         version = ipc_parse_version(&cmd);
1011         ns = current->nsproxy->ipc_ns;
1012
1013         switch (cmd) {
1014         case IPC_INFO:
1015         case SHM_INFO:
1016         case SHM_STAT:
1017         case IPC_STAT:
1018                 return shmctl_nolock(ns, shmid, cmd, version, buf);
1019         case IPC_RMID:
1020         case IPC_SET:
1021                 return shmctl_down(ns, shmid, cmd, buf, version);
1022         case SHM_LOCK:
1023         case SHM_UNLOCK:
1024         {
1025                 struct file *shm_file;
1026
1027                 rcu_read_lock();
1028                 shp = shm_obtain_object_check(ns, shmid);
1029                 if (IS_ERR(shp)) {
1030                         err = PTR_ERR(shp);
1031                         goto out_unlock1;
1032                 }
1033
1034                 audit_ipc_obj(&(shp->shm_perm));
1035                 err = security_shm_shmctl(shp, cmd);
1036                 if (err)
1037                         goto out_unlock1;
1038
1039                 ipc_lock_object(&shp->shm_perm);
1040
1041                 /* check if shm_destroy() is tearing down shp */
1042                 if (!ipc_valid_object(&shp->shm_perm)) {
1043                         err = -EIDRM;
1044                         goto out_unlock0;
1045                 }
1046
1047                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1048                         kuid_t euid = current_euid();
1049
1050                         if (!uid_eq(euid, shp->shm_perm.uid) &&
1051                             !uid_eq(euid, shp->shm_perm.cuid)) {
1052                                 err = -EPERM;
1053                                 goto out_unlock0;
1054                         }
1055                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1056                                 err = -EPERM;
1057                                 goto out_unlock0;
1058                         }
1059                 }
1060
1061                 shm_file = shp->shm_file;
1062                 if (is_file_hugepages(shm_file))
1063                         goto out_unlock0;
1064
1065                 if (cmd == SHM_LOCK) {
1066                         struct user_struct *user = current_user();
1067
1068                         err = shmem_lock(shm_file, 1, user);
1069                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1070                                 shp->shm_perm.mode |= SHM_LOCKED;
1071                                 shp->mlock_user = user;
1072                         }
1073                         goto out_unlock0;
1074                 }
1075
1076                 /* SHM_UNLOCK */
1077                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1078                         goto out_unlock0;
1079                 shmem_lock(shm_file, 0, shp->mlock_user);
1080                 shp->shm_perm.mode &= ~SHM_LOCKED;
1081                 shp->mlock_user = NULL;
1082                 get_file(shm_file);
1083                 ipc_unlock_object(&shp->shm_perm);
1084                 rcu_read_unlock();
1085                 shmem_unlock_mapping(shm_file->f_mapping);
1086
1087                 fput(shm_file);
1088                 return err;
1089         }
1090         default:
1091                 return -EINVAL;
1092         }
1093
1094 out_unlock0:
1095         ipc_unlock_object(&shp->shm_perm);
1096 out_unlock1:
1097         rcu_read_unlock();
1098         return err;
1099 }
1100
1101 /*
1102  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1103  *
1104  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1105  * "raddr" thing points to kernel space, and there has to be a wrapper around
1106  * this.
1107  */
1108 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1109               ulong *raddr, unsigned long shmlba)
1110 {
1111         struct shmid_kernel *shp;
1112         unsigned long addr = (unsigned long)shmaddr;
1113         unsigned long size;
1114         struct file *file;
1115         int    err;
1116         unsigned long flags = MAP_SHARED;
1117         unsigned long prot;
1118         int acc_mode;
1119         struct ipc_namespace *ns;
1120         struct shm_file_data *sfd;
1121         struct path path;
1122         fmode_t f_mode;
1123         unsigned long populate = 0;
1124
1125         err = -EINVAL;
1126         if (shmid < 0)
1127                 goto out;
1128
1129         if (addr) {
1130                 if (addr & (shmlba - 1)) {
1131                         /*
1132                          * Round down to the nearest multiple of shmlba.
1133                          * For sane do_mmap_pgoff() parameters, avoid
1134                          * round downs that trigger nil-page and MAP_FIXED.
1135                          */
1136                         if ((shmflg & SHM_RND) && addr >= shmlba)
1137                                 addr &= ~(shmlba - 1);
1138                         else
1139 #ifndef __ARCH_FORCE_SHMLBA
1140                                 if (addr & ~PAGE_MASK)
1141 #endif
1142                                         goto out;
1143                 }
1144
1145                 flags |= MAP_FIXED;
1146         } else if ((shmflg & SHM_REMAP))
1147                 goto out;
1148
1149         if (shmflg & SHM_RDONLY) {
1150                 prot = PROT_READ;
1151                 acc_mode = S_IRUGO;
1152                 f_mode = FMODE_READ;
1153         } else {
1154                 prot = PROT_READ | PROT_WRITE;
1155                 acc_mode = S_IRUGO | S_IWUGO;
1156                 f_mode = FMODE_READ | FMODE_WRITE;
1157         }
1158         if (shmflg & SHM_EXEC) {
1159                 prot |= PROT_EXEC;
1160                 acc_mode |= S_IXUGO;
1161         }
1162
1163         /*
1164          * We cannot rely on the fs check since SYSV IPC does have an
1165          * additional creator id...
1166          */
1167         ns = current->nsproxy->ipc_ns;
1168         rcu_read_lock();
1169         shp = shm_obtain_object_check(ns, shmid);
1170         if (IS_ERR(shp)) {
1171                 err = PTR_ERR(shp);
1172                 goto out_unlock;
1173         }
1174
1175         err = -EACCES;
1176         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1177                 goto out_unlock;
1178
1179         err = security_shm_shmat(shp, shmaddr, shmflg);
1180         if (err)
1181                 goto out_unlock;
1182
1183         ipc_lock_object(&shp->shm_perm);
1184
1185         /* check if shm_destroy() is tearing down shp */
1186         if (!ipc_valid_object(&shp->shm_perm)) {
1187                 ipc_unlock_object(&shp->shm_perm);
1188                 err = -EIDRM;
1189                 goto out_unlock;
1190         }
1191
1192         path = shp->shm_file->f_path;
1193         path_get(&path);
1194         shp->shm_nattch++;
1195         size = i_size_read(d_inode(path.dentry));
1196         ipc_unlock_object(&shp->shm_perm);
1197         rcu_read_unlock();
1198
1199         err = -ENOMEM;
1200         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1201         if (!sfd) {
1202                 path_put(&path);
1203                 goto out_nattch;
1204         }
1205
1206         file = alloc_file(&path, f_mode,
1207                           is_file_hugepages(shp->shm_file) ?
1208                                 &shm_file_operations_huge :
1209                                 &shm_file_operations);
1210         err = PTR_ERR(file);
1211         if (IS_ERR(file)) {
1212                 kfree(sfd);
1213                 path_put(&path);
1214                 goto out_nattch;
1215         }
1216
1217         file->private_data = sfd;
1218         file->f_mapping = shp->shm_file->f_mapping;
1219         sfd->id = shp->shm_perm.id;
1220         sfd->ns = get_ipc_ns(ns);
1221         sfd->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         if (down_write_killable(&current->mm->mmap_sem)) {
1229                 err = -EINTR;
1230                 goto out_fput;
1231         }
1232
1233         if (addr && !(shmflg & SHM_REMAP)) {
1234                 err = -EINVAL;
1235                 if (addr + size < addr)
1236                         goto invalid;
1237
1238                 if (find_vma_intersection(current->mm, addr, addr + size))
1239                         goto invalid;
1240         }
1241
1242         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1243         *raddr = addr;
1244         err = 0;
1245         if (IS_ERR_VALUE(addr))
1246                 err = (long)addr;
1247 invalid:
1248         up_write(&current->mm->mmap_sem);
1249         if (populate)
1250                 mm_populate(addr, populate);
1251
1252 out_fput:
1253         fput(file);
1254
1255 out_nattch:
1256         down_write(&shm_ids(ns).rwsem);
1257         shp = shm_lock(ns, shmid);
1258         shp->shm_nattch--;
1259         if (shm_may_destroy(ns, shp))
1260                 shm_destroy(ns, shp);
1261         else
1262                 shm_unlock(shp);
1263         up_write(&shm_ids(ns).rwsem);
1264         return err;
1265
1266 out_unlock:
1267         rcu_read_unlock();
1268 out:
1269         return err;
1270 }
1271
1272 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1273 {
1274         unsigned long ret;
1275         long err;
1276
1277         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1278         if (err)
1279                 return err;
1280         force_successful_syscall_return();
1281         return (long)ret;
1282 }
1283
1284 /*
1285  * detach and kill segment if marked destroyed.
1286  * The work is done in shm_close.
1287  */
1288 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1289 {
1290         struct mm_struct *mm = current->mm;
1291         struct vm_area_struct *vma;
1292         unsigned long addr = (unsigned long)shmaddr;
1293         int retval = -EINVAL;
1294 #ifdef CONFIG_MMU
1295         loff_t size = 0;
1296         struct file *file;
1297         struct vm_area_struct *next;
1298 #endif
1299
1300         if (addr & ~PAGE_MASK)
1301                 return retval;
1302
1303         if (down_write_killable(&mm->mmap_sem))
1304                 return -EINTR;
1305
1306         /*
1307          * This function tries to be smart and unmap shm segments that
1308          * were modified by partial mlock or munmap calls:
1309          * - It first determines the size of the shm segment that should be
1310          *   unmapped: It searches for a vma that is backed by shm and that
1311          *   started at address shmaddr. It records it's size and then unmaps
1312          *   it.
1313          * - Then it unmaps all shm vmas that started at shmaddr and that
1314          *   are within the initially determined size and that are from the
1315          *   same shm segment from which we determined the size.
1316          * Errors from do_munmap are ignored: the function only fails if
1317          * it's called with invalid parameters or if it's called to unmap
1318          * a part of a vma. Both calls in this function are for full vmas,
1319          * the parameters are directly copied from the vma itself and always
1320          * valid - therefore do_munmap cannot fail. (famous last words?)
1321          */
1322         /*
1323          * If it had been mremap()'d, the starting address would not
1324          * match the usual checks anyway. So assume all vma's are
1325          * above the starting address given.
1326          */
1327         vma = find_vma(mm, addr);
1328
1329 #ifdef CONFIG_MMU
1330         while (vma) {
1331                 next = vma->vm_next;
1332
1333                 /*
1334                  * Check if the starting address would match, i.e. it's
1335                  * a fragment created by mprotect() and/or munmap(), or it
1336                  * otherwise it starts at this address with no hassles.
1337                  */
1338                 if ((vma->vm_ops == &shm_vm_ops) &&
1339                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1340
1341                         /*
1342                          * Record the file of the shm segment being
1343                          * unmapped.  With mremap(), someone could place
1344                          * page from another segment but with equal offsets
1345                          * in the range we are unmapping.
1346                          */
1347                         file = vma->vm_file;
1348                         size = i_size_read(file_inode(vma->vm_file));
1349                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1350                         /*
1351                          * We discovered the size of the shm segment, so
1352                          * break out of here and fall through to the next
1353                          * loop that uses the size information to stop
1354                          * searching for matching vma's.
1355                          */
1356                         retval = 0;
1357                         vma = next;
1358                         break;
1359                 }
1360                 vma = next;
1361         }
1362
1363         /*
1364          * We need look no further than the maximum address a fragment
1365          * could possibly have landed at. Also cast things to loff_t to
1366          * prevent overflows and make comparisons vs. equal-width types.
1367          */
1368         size = PAGE_ALIGN(size);
1369         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1370                 next = vma->vm_next;
1371
1372                 /* finding a matching vma now does not alter retval */
1373                 if ((vma->vm_ops == &shm_vm_ops) &&
1374                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1375                     (vma->vm_file == file))
1376                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1377                 vma = next;
1378         }
1379
1380 #else   /* CONFIG_MMU */
1381         /* under NOMMU conditions, the exact address to be destroyed must be
1382          * given
1383          */
1384         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1385                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1386                 retval = 0;
1387         }
1388
1389 #endif
1390
1391         up_write(&mm->mmap_sem);
1392         return retval;
1393 }
1394
1395 #ifdef CONFIG_PROC_FS
1396 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1397 {
1398         struct user_namespace *user_ns = seq_user_ns(s);
1399         struct shmid_kernel *shp = it;
1400         unsigned long rss = 0, swp = 0;
1401
1402         shm_add_rss_swap(shp, &rss, &swp);
1403
1404 #if BITS_PER_LONG <= 32
1405 #define SIZE_SPEC "%10lu"
1406 #else
1407 #define SIZE_SPEC "%21lu"
1408 #endif
1409
1410         seq_printf(s,
1411                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1412                    "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1413                    SIZE_SPEC " " SIZE_SPEC "\n",
1414                    shp->shm_perm.key,
1415                    shp->shm_perm.id,
1416                    shp->shm_perm.mode,
1417                    shp->shm_segsz,
1418                    shp->shm_cprid,
1419                    shp->shm_lprid,
1420                    shp->shm_nattch,
1421                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1422                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1423                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1424                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1425                    shp->shm_atim,
1426                    shp->shm_dtim,
1427                    shp->shm_ctim,
1428                    rss * PAGE_SIZE,
1429                    swp * PAGE_SIZE);
1430
1431         return 0;
1432 }
1433 #endif