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