ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  * support for audit of ipc object properties and permission changes
  17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  18  *
  19  * namespaces support
  20  * OpenVZ, SWsoft Inc.
  21  * Pavel Emelianov <xemul@openvz.org>
  22  *
  23  * Better ipc lock (kern_ipc_perm.lock) handling
  24  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
  25  */
  26
  27 #include <linux/slab.h>
  28 #include <linux/mm.h>
  29 #include <linux/hugetlb.h>
  30 #include <linux/shm.h>
  31 #include <linux/init.h>
  32 #include <linux/file.h>
  33 #include <linux/mman.h>
  34 #include <linux/shmem_fs.h>
  35 #include <linux/security.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/audit.h>
  38 #include <linux/capability.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/rwsem.h>
  42 #include <linux/nsproxy.h>
  43 #include <linux/mount.h>
  44 #include <linux/ipc_namespace.h>
  45
  46 #include <linux/uaccess.h>
  47
  48 #include "util.h"
  49
  50 struct shm_file_data {
  51         int id;
  52         struct ipc_namespace *ns;
  53         struct file *file;
  54         const struct vm_operations_struct *vm_ops;
  55 };
  56
  57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  58
  59 static const struct file_operations shm_file_operations;
  60 static const struct vm_operations_struct shm_vm_ops;
  61
  62 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  63
  64 #define shm_unlock(shp)                 \
  65         ipc_unlock(&(shp)->shm_perm)
  66
  67 static int newseg(struct ipc_namespace *, struct ipc_params *);
  68 static void shm_open(struct vm_area_struct *vma);
  69 static void shm_close(struct vm_area_struct *vma);
  70 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
  71 #ifdef CONFIG_PROC_FS
  72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  73 #endif
  74
  75 void shm_init_ns(struct ipc_namespace *ns)
  76 {
  77         ns->shm_ctlmax = SHMMAX;
  78         ns->shm_ctlall = SHMALL;
  79         ns->shm_ctlmni = SHMMNI;
  80         ns->shm_rmid_forced = 0;
  81         ns->shm_tot = 0;
  82         ipc_init_ids(&shm_ids(ns));
  83 }
  84
  85 /*
  86  * Called with shm_ids.rwsem (writer) and the shp structure locked.
  87  * Only shm_ids.rwsem remains locked on exit.
  88  */
  89 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  90 {
  91         struct shmid_kernel *shp;
  92         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  93
  94         if (shp->shm_nattch) {
  95                 shp->shm_perm.mode |= SHM_DEST;
  96                 /* Do not find it any more */
  97                 shp->shm_perm.key = IPC_PRIVATE;
  98                 shm_unlock(shp);
  99         } else
 100                 shm_destroy(ns, shp);
 101 }
 102
 103 #ifdef CONFIG_IPC_NS
 104 void shm_exit_ns(struct ipc_namespace *ns)
 105 {
 106         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 107         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 108 }
 109 #endif
 110
 111 static int __init ipc_ns_init(void)
 112 {
 113         shm_init_ns(&init_ipc_ns);
 114         return 0;
 115 }
 116
 117 pure_initcall(ipc_ns_init);
 118
 119 void __init shm_init(void)
 120 {
 121         ipc_init_proc_interface("sysvipc/shm",
 122 #if BITS_PER_LONG <= 32
 123                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 124 #else
 125                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 126 #endif
 127                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 128 }
 129
 130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 131 {
 132         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 133
 134         if (IS_ERR(ipcp))
 135                 return ERR_CAST(ipcp);
 136
 137         return container_of(ipcp, struct shmid_kernel, shm_perm);
 138 }
 139
 140 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 141 {
 142         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 143
 144         if (IS_ERR(ipcp))
 145                 return ERR_CAST(ipcp);
 146
 147         return container_of(ipcp, struct shmid_kernel, shm_perm);
 148 }
 149
 150 /*
 151  * shm_lock_(check_) routines are called in the paths where the rwsem
 152  * is not necessarily held.
 153  */
 154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 155 {
 156         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 157
 158         /*
 159          * Callers of shm_lock() must validate the status of the returned ipc
 160          * object pointer (as returned by ipc_lock()), and error out as
 161          * appropriate.
 162          */
 163         if (IS_ERR(ipcp))
 164                 return (void *)ipcp;
 165         return container_of(ipcp, struct shmid_kernel, shm_perm);
 166 }
 167
 168 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 169 {
 170         rcu_read_lock();
 171         ipc_lock_object(&ipcp->shm_perm);
 172 }
 173
 174 static void shm_rcu_free(struct rcu_head *head)
 175 {
 176         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
 177         struct shmid_kernel *shp = ipc_rcu_to_struct(p);
 178
 179         security_shm_free(shp);
 180         ipc_rcu_free(head);
 181 }
 182
 183 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 184 {
 185         list_del(&s->shm_clist);
 186         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 187 }
 188
 189
 190 static int __shm_open(struct vm_area_struct *vma)
 191 {
 192         struct file *file = vma->vm_file;
 193         struct shm_file_data *sfd = shm_file_data(file);
 194         struct shmid_kernel *shp;
 195
 196         shp = shm_lock(sfd->ns, sfd->id);
 197
 198         if (IS_ERR(shp))
 199                 return PTR_ERR(shp);
 200
 201         shp->shm_atim = get_seconds();
 202         shp->shm_lprid = task_tgid_vnr(current);
 203         shp->shm_nattch++;
 204         shm_unlock(shp);
 205         return 0;
 206 }
 207
 208 /* This is called by fork, once for every shm attach. */
 209 static void shm_open(struct vm_area_struct *vma)
 210 {
 211         int err = __shm_open(vma);
 212         /*
 213          * We raced in the idr lookup or with shm_destroy().
 214          * Either way, the ID is busted.
 215          */
 216         WARN_ON_ONCE(err);
 217 }
 218
 219 /*
 220  * shm_destroy - free the struct shmid_kernel
 221  *
 222  * @ns: namespace
 223  * @shp: struct to free
 224  *
 225  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 226  * but returns with shp unlocked and freed.
 227  */
 228 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 229 {
 230         struct file *shm_file;
 231
 232         shm_file = shp->shm_file;
 233         shp->shm_file = NULL;
 234         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 235         shm_rmid(ns, shp);
 236         shm_unlock(shp);
 237         if (!is_file_hugepages(shm_file))
 238                 shmem_lock(shm_file, 0, shp->mlock_user);
 239         else if (shp->mlock_user)
 240                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 241                                 shp->mlock_user);
 242         fput(shm_file);
 243         ipc_rcu_putref(shp, shm_rcu_free);
 244 }
 245
 246 /*
 247  * shm_may_destroy - identifies whether shm segment should be destroyed now
 248  *
 249  * Returns true if and only if there are no active users of the segment and
 250  * one of the following is true:
 251  *
 252  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 253  *
 254  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 255  */
 256 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 257 {
 258         return (shp->shm_nattch == 0) &&
 259                (ns->shm_rmid_forced ||
 260                 (shp->shm_perm.mode & SHM_DEST));
 261 }
 262
 263 /*
 264  * remove the attach descriptor vma.
 265  * free memory for segment if it is marked destroyed.
 266  * The descriptor has already been removed from the current->mm->mmap list
 267  * and will later be kfree()d.
 268  */
 269 static void shm_close(struct vm_area_struct *vma)
 270 {
 271         struct file *file = vma->vm_file;
 272         struct shm_file_data *sfd = shm_file_data(file);
 273         struct shmid_kernel *shp;
 274         struct ipc_namespace *ns = sfd->ns;
 275
 276         down_write(&shm_ids(ns).rwsem);
 277         /* remove from the list of attaches of the shm segment */
 278         shp = shm_lock(ns, sfd->id);
 279
 280         /*
 281          * We raced in the idr lookup or with shm_destroy().
 282          * Either way, the ID is busted.
 283          */
 284         if (WARN_ON_ONCE(IS_ERR(shp)))
 285                 goto done; /* no-op */
 286
 287         shp->shm_lprid = task_tgid_vnr(current);
 288         shp->shm_dtim = get_seconds();
 289         shp->shm_nattch--;
 290         if (shm_may_destroy(ns, shp))
 291                 shm_destroy(ns, shp);
 292         else
 293                 shm_unlock(shp);
 294 done:
 295         up_write(&shm_ids(ns).rwsem);
 296 }
 297
 298 /* Called with ns->shm_ids(ns).rwsem locked */
 299 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 300 {
 301         struct ipc_namespace *ns = data;
 302         struct kern_ipc_perm *ipcp = p;
 303         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 304
 305         /*
 306          * We want to destroy segments without users and with already
 307          * exit'ed originating process.
 308          *
 309          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 310          */
 311         if (shp->shm_creator != NULL)
 312                 return 0;
 313
 314         if (shm_may_destroy(ns, shp)) {
 315                 shm_lock_by_ptr(shp);
 316                 shm_destroy(ns, shp);
 317         }
 318         return 0;
 319 }
 320
 321 void shm_destroy_orphaned(struct ipc_namespace *ns)
 322 {
 323         down_write(&shm_ids(ns).rwsem);
 324         if (shm_ids(ns).in_use)
 325                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 326         up_write(&shm_ids(ns).rwsem);
 327 }
 328
 329 /* Locking assumes this will only be called with task == current */
 330 void exit_shm(struct task_struct *task)
 331 {
 332         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 333         struct shmid_kernel *shp, *n;
 334
 335         if (list_empty(&task->sysvshm.shm_clist))
 336                 return;
 337
 338         /*
 339          * If kernel.shm_rmid_forced is not set then only keep track of
 340          * which shmids are orphaned, so that a later set of the sysctl
 341          * can clean them up.
 342          */
 343         if (!ns->shm_rmid_forced) {
 344                 down_read(&shm_ids(ns).rwsem);
 345                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
 346                         shp->shm_creator = NULL;
 347                 /*
 348                  * Only under read lock but we are only called on current
 349                  * so no entry on the list will be shared.
 350                  */
 351                 list_del(&task->sysvshm.shm_clist);
 352                 up_read(&shm_ids(ns).rwsem);
 353                 return;
 354         }
 355
 356         /*
 357          * Destroy all already created segments, that were not yet mapped,
 358          * and mark any mapped as orphan to cover the sysctl toggling.
 359          * Destroy is skipped if shm_may_destroy() returns false.
 360          */
 361         down_write(&shm_ids(ns).rwsem);
 362         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
 363                 shp->shm_creator = NULL;
 364
 365                 if (shm_may_destroy(ns, shp)) {
 366                         shm_lock_by_ptr(shp);
 367                         shm_destroy(ns, shp);
 368                 }
 369         }
 370
 371         /* Remove the list head from any segments still attached. */
 372         list_del(&task->sysvshm.shm_clist);
 373         up_write(&shm_ids(ns).rwsem);
 374 }
 375
 376 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 377 {
 378         struct file *file = vma->vm_file;
 379         struct shm_file_data *sfd = shm_file_data(file);
 380
 381         return sfd->vm_ops->fault(vma, vmf);
 382 }
 383
 384 #ifdef CONFIG_NUMA
 385 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 386 {
 387         struct file *file = vma->vm_file;
 388         struct shm_file_data *sfd = shm_file_data(file);
 389         int err = 0;
 390         if (sfd->vm_ops->set_policy)
 391                 err = sfd->vm_ops->set_policy(vma, new);
 392         return err;
 393 }
 394
 395 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 396                                         unsigned long addr)
 397 {
 398         struct file *file = vma->vm_file;
 399         struct shm_file_data *sfd = shm_file_data(file);
 400         struct mempolicy *pol = NULL;
 401
 402         if (sfd->vm_ops->get_policy)
 403                 pol = sfd->vm_ops->get_policy(vma, addr);
 404         else if (vma->vm_policy)
 405                 pol = vma->vm_policy;
 406
 407         return pol;
 408 }
 409 #endif
 410
 411 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 412 {
 413         struct shm_file_data *sfd = shm_file_data(file);
 414         int ret;
 415
 416         /*
 417          * In case of remap_file_pages() emulation, the file can represent
 418          * removed IPC ID: propogate shm_lock() error to caller.
 419          */
 420         ret =__shm_open(vma);
 421         if (ret)
 422                 return ret;
 423
 424         ret = sfd->file->f_op->mmap(sfd->file, vma);
 425         if (ret) {
 426                 shm_close(vma);
 427                 return ret;
 428         }
 429         sfd->vm_ops = vma->vm_ops;
 430 #ifdef CONFIG_MMU
 431         WARN_ON(!sfd->vm_ops->fault);
 432 #endif
 433         vma->vm_ops = &shm_vm_ops;
 434         return 0;
 435 }
 436
 437 static int shm_release(struct inode *ino, struct file *file)
 438 {
 439         struct shm_file_data *sfd = shm_file_data(file);
 440
 441         put_ipc_ns(sfd->ns);
 442         shm_file_data(file) = NULL;
 443         kfree(sfd);
 444         return 0;
 445 }
 446
 447 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 448 {
 449         struct shm_file_data *sfd = shm_file_data(file);
 450
 451         if (!sfd->file->f_op->fsync)
 452                 return -EINVAL;
 453         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 454 }
 455
 456 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 457                           loff_t len)
 458 {
 459         struct shm_file_data *sfd = shm_file_data(file);
 460
 461         if (!sfd->file->f_op->fallocate)
 462                 return -EOPNOTSUPP;
 463         return sfd->file->f_op->fallocate(file, mode, offset, len);
 464 }
 465
 466 static unsigned long shm_get_unmapped_area(struct file *file,
 467         unsigned long addr, unsigned long len, unsigned long pgoff,
 468         unsigned long flags)
 469 {
 470         struct shm_file_data *sfd = shm_file_data(file);
 471         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 472                                                 pgoff, flags);
 473 }
 474
 475 static const struct file_operations shm_file_operations = {
 476         .mmap           = shm_mmap,
 477         .fsync          = shm_fsync,
 478         .release        = shm_release,
 479         .get_unmapped_area      = shm_get_unmapped_area,
 480         .llseek         = noop_llseek,
 481         .fallocate      = shm_fallocate,
 482 };
 483
 484 /*
 485  * shm_file_operations_huge is now identical to shm_file_operations,
 486  * but we keep it distinct for the sake of is_file_shm_hugepages().
 487  */
 488 static const struct file_operations shm_file_operations_huge = {
 489         .mmap           = shm_mmap,
 490         .fsync          = shm_fsync,
 491         .release        = shm_release,
 492         .get_unmapped_area      = shm_get_unmapped_area,
 493         .llseek         = noop_llseek,
 494         .fallocate      = shm_fallocate,
 495 };
 496
 497 bool is_file_shm_hugepages(struct file *file)
 498 {
 499         return file->f_op == &shm_file_operations_huge;
 500 }
 501
 502 static const struct vm_operations_struct shm_vm_ops = {
 503         .open   = shm_open,     /* callback for a new vm-area open */
 504         .close  = shm_close,    /* callback for when the vm-area is released */
 505         .fault  = shm_fault,
 506 #if defined(CONFIG_NUMA)
 507         .set_policy = shm_set_policy,
 508         .get_policy = shm_get_policy,
 509 #endif
 510 };
 511
 512 /**
 513  * newseg - Create a new shared memory segment
 514  * @ns: namespace
 515  * @params: ptr to the structure that contains key, size and shmflg
 516  *
 517  * Called with shm_ids.rwsem held as a writer.
 518  */
 519 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 520 {
 521         key_t key = params->key;
 522         int shmflg = params->flg;
 523         size_t size = params->u.size;
 524         int error;
 525         struct shmid_kernel *shp;
 526         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 527         struct file *file;
 528         char name[13];
 529         int id;
 530         vm_flags_t acctflag = 0;
 531
 532         if (size < SHMMIN || size > ns->shm_ctlmax)
 533                 return -EINVAL;
 534
 535         if (numpages << PAGE_SHIFT < size)
 536                 return -ENOSPC;
 537
 538         if (ns->shm_tot + numpages < ns->shm_tot ||
 539                         ns->shm_tot + numpages > ns->shm_ctlall)
 540                 return -ENOSPC;
 541
 542         shp = ipc_rcu_alloc(sizeof(*shp));
 543         if (!shp)
 544                 return -ENOMEM;
 545
 546         shp->shm_perm.key = key;
 547         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 548         shp->mlock_user = NULL;
 549
 550         shp->shm_perm.security = NULL;
 551         error = security_shm_alloc(shp);
 552         if (error) {
 553                 ipc_rcu_putref(shp, ipc_rcu_free);
 554                 return error;
 555         }
 556
 557         sprintf(name, "SYSV%08x", key);
 558         if (shmflg & SHM_HUGETLB) {
 559                 struct hstate *hs;
 560                 size_t hugesize;
 561
 562                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 563                 if (!hs) {
 564                         error = -EINVAL;
 565                         goto no_file;
 566                 }
 567                 hugesize = ALIGN(size, huge_page_size(hs));
 568
 569                 /* hugetlb_file_setup applies strict accounting */
 570                 if (shmflg & SHM_NORESERVE)
 571                         acctflag = VM_NORESERVE;
 572                 file = hugetlb_file_setup(name, hugesize, acctflag,
 573                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 574                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 575         } else {
 576                 /*
 577                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 578                  * if it's asked for.
 579                  */
 580                 if  ((shmflg & SHM_NORESERVE) &&
 581                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 582                         acctflag = VM_NORESERVE;
 583                 file = shmem_kernel_file_setup(name, size, acctflag);
 584         }
 585         error = PTR_ERR(file);
 586         if (IS_ERR(file))
 587                 goto no_file;
 588
 589         shp->shm_cprid = task_tgid_vnr(current);
 590         shp->shm_lprid = 0;
 591         shp->shm_atim = shp->shm_dtim = 0;
 592         shp->shm_ctim = get_seconds();
 593         shp->shm_segsz = size;
 594         shp->shm_nattch = 0;
 595         shp->shm_file = file;
 596         shp->shm_creator = current;
 597
 598         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 599         if (id < 0) {
 600                 error = id;
 601                 goto no_id;
 602         }
 603
 604         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 605
 606         /*
 607          * shmid gets reported as "inode#" in /proc/pid/maps.
 608          * proc-ps tools use this. Changing this will break them.
 609          */
 610         file_inode(file)->i_ino = shp->shm_perm.id;
 611
 612         ns->shm_tot += numpages;
 613         error = shp->shm_perm.id;
 614
 615         ipc_unlock_object(&shp->shm_perm);
 616         rcu_read_unlock();
 617         return error;
 618
 619 no_id:
 620         if (is_file_hugepages(file) && shp->mlock_user)
 621                 user_shm_unlock(size, shp->mlock_user);
 622         fput(file);
 623 no_file:
 624         ipc_rcu_putref(shp, shm_rcu_free);
 625         return error;
 626 }
 627
 628 /*
 629  * Called with shm_ids.rwsem and ipcp locked.
 630  */
 631 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 632 {
 633         struct shmid_kernel *shp;
 634
 635         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 636         return security_shm_associate(shp, shmflg);
 637 }
 638
 639 /*
 640  * Called with shm_ids.rwsem and ipcp locked.
 641  */
 642 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 643                                 struct ipc_params *params)
 644 {
 645         struct shmid_kernel *shp;
 646
 647         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 648         if (shp->shm_segsz < params->u.size)
 649                 return -EINVAL;
 650
 651         return 0;
 652 }
 653
 654 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 655 {
 656         struct ipc_namespace *ns;
 657         static const struct ipc_ops shm_ops = {
 658                 .getnew = newseg,
 659                 .associate = shm_security,
 660                 .more_checks = shm_more_checks,
 661         };
 662         struct ipc_params shm_params;
 663
 664         ns = current->nsproxy->ipc_ns;
 665
 666         shm_params.key = key;
 667         shm_params.flg = shmflg;
 668         shm_params.u.size = size;
 669
 670         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 671 }
 672
 673 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 674 {
 675         switch (version) {
 676         case IPC_64:
 677                 return copy_to_user(buf, in, sizeof(*in));
 678         case IPC_OLD:
 679             {
 680                 struct shmid_ds out;
 681
 682                 memset(&out, 0, sizeof(out));
 683                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 684                 out.shm_segsz   = in->shm_segsz;
 685                 out.shm_atime   = in->shm_atime;
 686                 out.shm_dtime   = in->shm_dtime;
 687                 out.shm_ctime   = in->shm_ctime;
 688                 out.shm_cpid    = in->shm_cpid;
 689                 out.shm_lpid    = in->shm_lpid;
 690                 out.shm_nattch  = in->shm_nattch;
 691
 692                 return copy_to_user(buf, &out, sizeof(out));
 693             }
 694         default:
 695                 return -EINVAL;
 696         }
 697 }
 698
 699 static inline unsigned long
 700 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 701 {
 702         switch (version) {
 703         case IPC_64:
 704                 if (copy_from_user(out, buf, sizeof(*out)))
 705                         return -EFAULT;
 706                 return 0;
 707         case IPC_OLD:
 708             {
 709                 struct shmid_ds tbuf_old;
 710
 711                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 712                         return -EFAULT;
 713
 714                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 715                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 716                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 717
 718                 return 0;
 719             }
 720         default:
 721                 return -EINVAL;
 722         }
 723 }
 724
 725 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 726 {
 727         switch (version) {
 728         case IPC_64:
 729                 return copy_to_user(buf, in, sizeof(*in));
 730         case IPC_OLD:
 731             {
 732                 struct shminfo out;
 733
 734                 if (in->shmmax > INT_MAX)
 735                         out.shmmax = INT_MAX;
 736                 else
 737                         out.shmmax = (int)in->shmmax;
 738
 739                 out.shmmin      = in->shmmin;
 740                 out.shmmni      = in->shmmni;
 741                 out.shmseg      = in->shmseg;
 742                 out.shmall      = in->shmall;
 743
 744                 return copy_to_user(buf, &out, sizeof(out));
 745             }
 746         default:
 747                 return -EINVAL;
 748         }
 749 }
 750
 751 /*
 752  * Calculate and add used RSS and swap pages of a shm.
 753  * Called with shm_ids.rwsem held as a reader
 754  */
 755 static void shm_add_rss_swap(struct shmid_kernel *shp,
 756         unsigned long *rss_add, unsigned long *swp_add)
 757 {
 758         struct inode *inode;
 759
 760         inode = file_inode(shp->shm_file);
 761
 762         if (is_file_hugepages(shp->shm_file)) {
 763                 struct address_space *mapping = inode->i_mapping;
 764                 struct hstate *h = hstate_file(shp->shm_file);
 765                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 766         } else {
 767 #ifdef CONFIG_SHMEM
 768                 struct shmem_inode_info *info = SHMEM_I(inode);
 769                 spin_lock_irq(&info->lock);
 770                 *rss_add += inode->i_mapping->nrpages;
 771                 *swp_add += info->swapped;
 772                 spin_unlock_irq(&info->lock);
 773 #else
 774                 *rss_add += inode->i_mapping->nrpages;
 775 #endif
 776         }
 777 }
 778
 779 /*
 780  * Called with shm_ids.rwsem held as a reader
 781  */
 782 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 783                 unsigned long *swp)
 784 {
 785         int next_id;
 786         int total, in_use;
 787
 788         *rss = 0;
 789         *swp = 0;
 790
 791         in_use = shm_ids(ns).in_use;
 792
 793         for (total = 0, next_id = 0; total < in_use; next_id++) {
 794                 struct kern_ipc_perm *ipc;
 795                 struct shmid_kernel *shp;
 796
 797                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 798                 if (ipc == NULL)
 799                         continue;
 800                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 801
 802                 shm_add_rss_swap(shp, rss, swp);
 803
 804                 total++;
 805         }
 806 }
 807
 808 /*
 809  * This function handles some shmctl commands which require the rwsem
 810  * to be held in write mode.
 811  * NOTE: no locks must be held, the rwsem is taken inside this function.
 812  */
 813 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 814                        struct shmid_ds __user *buf, int version)
 815 {
 816         struct kern_ipc_perm *ipcp;
 817         struct shmid64_ds shmid64;
 818         struct shmid_kernel *shp;
 819         int err;
 820
 821         if (cmd == IPC_SET) {
 822                 if (copy_shmid_from_user(&shmid64, buf, version))
 823                         return -EFAULT;
 824         }
 825
 826         down_write(&shm_ids(ns).rwsem);
 827         rcu_read_lock();
 828
 829         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 830                                       &shmid64.shm_perm, 0);
 831         if (IS_ERR(ipcp)) {
 832                 err = PTR_ERR(ipcp);
 833                 goto out_unlock1;
 834         }
 835
 836         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 837
 838         err = security_shm_shmctl(shp, cmd);
 839         if (err)
 840                 goto out_unlock1;
 841
 842         switch (cmd) {
 843         case IPC_RMID:
 844                 ipc_lock_object(&shp->shm_perm);
 845                 /* do_shm_rmid unlocks the ipc object and rcu */
 846                 do_shm_rmid(ns, ipcp);
 847                 goto out_up;
 848         case IPC_SET:
 849                 ipc_lock_object(&shp->shm_perm);
 850                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
 851                 if (err)
 852                         goto out_unlock0;
 853                 shp->shm_ctim = get_seconds();
 854                 break;
 855         default:
 856                 err = -EINVAL;
 857                 goto out_unlock1;
 858         }
 859
 860 out_unlock0:
 861         ipc_unlock_object(&shp->shm_perm);
 862 out_unlock1:
 863         rcu_read_unlock();
 864 out_up:
 865         up_write(&shm_ids(ns).rwsem);
 866         return err;
 867 }
 868
 869 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
 870                          int cmd, int version, void __user *buf)
 871 {
 872         int err;
 873         struct shmid_kernel *shp;
 874
 875         /* preliminary security checks for *_INFO */
 876         if (cmd == IPC_INFO || cmd == SHM_INFO) {
 877                 err = security_shm_shmctl(NULL, cmd);
 878                 if (err)
 879                         return err;
 880         }
 881
 882         switch (cmd) {
 883         case IPC_INFO:
 884         {
 885                 struct shminfo64 shminfo;
 886
 887                 memset(&shminfo, 0, sizeof(shminfo));
 888                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 889                 shminfo.shmmax = ns->shm_ctlmax;
 890                 shminfo.shmall = ns->shm_ctlall;
 891
 892                 shminfo.shmmin = SHMMIN;
 893                 if (copy_shminfo_to_user(buf, &shminfo, version))
 894                         return -EFAULT;
 895
 896                 down_read(&shm_ids(ns).rwsem);
 897                 err = ipc_get_maxid(&shm_ids(ns));
 898                 up_read(&shm_ids(ns).rwsem);
 899
 900                 if (err < 0)
 901                         err = 0;
 902                 goto out;
 903         }
 904         case SHM_INFO:
 905         {
 906                 struct shm_info shm_info;
 907
 908                 memset(&shm_info, 0, sizeof(shm_info));
 909                 down_read(&shm_ids(ns).rwsem);
 910                 shm_info.used_ids = shm_ids(ns).in_use;
 911                 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
 912                 shm_info.shm_tot = ns->shm_tot;
 913                 shm_info.swap_attempts = 0;
 914                 shm_info.swap_successes = 0;
 915                 err = ipc_get_maxid(&shm_ids(ns));
 916                 up_read(&shm_ids(ns).rwsem);
 917                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 918                         err = -EFAULT;
 919                         goto out;
 920                 }
 921
 922                 err = err < 0 ? 0 : err;
 923                 goto out;
 924         }
 925         case SHM_STAT:
 926         case IPC_STAT:
 927         {
 928                 struct shmid64_ds tbuf;
 929                 int result;
 930
 931                 rcu_read_lock();
 932                 if (cmd == SHM_STAT) {
 933                         shp = shm_obtain_object(ns, shmid);
 934                         if (IS_ERR(shp)) {
 935                                 err = PTR_ERR(shp);
 936                                 goto out_unlock;
 937                         }
 938                         result = shp->shm_perm.id;
 939                 } else {
 940                         shp = shm_obtain_object_check(ns, shmid);
 941                         if (IS_ERR(shp)) {
 942                                 err = PTR_ERR(shp);
 943                                 goto out_unlock;
 944                         }
 945                         result = 0;
 946                 }
 947
 948                 err = -EACCES;
 949                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 950                         goto out_unlock;
 951
 952                 err = security_shm_shmctl(shp, cmd);
 953                 if (err)
 954                         goto out_unlock;
 955
 956                 memset(&tbuf, 0, sizeof(tbuf));
 957                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 958                 tbuf.shm_segsz  = shp->shm_segsz;
 959                 tbuf.shm_atime  = shp->shm_atim;
 960                 tbuf.shm_dtime  = shp->shm_dtim;
 961                 tbuf.shm_ctime  = shp->shm_ctim;
 962                 tbuf.shm_cpid   = shp->shm_cprid;
 963                 tbuf.shm_lpid   = shp->shm_lprid;
 964                 tbuf.shm_nattch = shp->shm_nattch;
 965                 rcu_read_unlock();
 966
 967                 if (copy_shmid_to_user(buf, &tbuf, version))
 968                         err = -EFAULT;
 969                 else
 970                         err = result;
 971                 goto out;
 972         }
 973         default:
 974                 return -EINVAL;
 975         }
 976
 977 out_unlock:
 978         rcu_read_unlock();
 979 out:
 980         return err;
 981 }
 982
 983 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 984 {
 985         struct shmid_kernel *shp;
 986         int err, version;
 987         struct ipc_namespace *ns;
 988
 989         if (cmd < 0 || shmid < 0)
 990                 return -EINVAL;
 991
 992         version = ipc_parse_version(&cmd);
 993         ns = current->nsproxy->ipc_ns;
 994
 995         switch (cmd) {
 996         case IPC_INFO:
 997         case SHM_INFO:
 998         case SHM_STAT:
 999         case IPC_STAT:
1000                 return shmctl_nolock(ns, shmid, cmd, version, buf);
1001         case IPC_RMID:
1002         case IPC_SET:
1003                 return shmctl_down(ns, shmid, cmd, buf, version);
1004         case SHM_LOCK:
1005         case SHM_UNLOCK:
1006         {
1007                 struct file *shm_file;
1008
1009                 rcu_read_lock();
1010                 shp = shm_obtain_object_check(ns, shmid);
1011                 if (IS_ERR(shp)) {
1012                         err = PTR_ERR(shp);
1013                         goto out_unlock1;
1014                 }
1015
1016                 audit_ipc_obj(&(shp->shm_perm));
1017                 err = security_shm_shmctl(shp, cmd);
1018                 if (err)
1019                         goto out_unlock1;
1020
1021                 ipc_lock_object(&shp->shm_perm);
1022
1023                 /* check if shm_destroy() is tearing down shp */
1024                 if (!ipc_valid_object(&shp->shm_perm)) {
1025                         err = -EIDRM;
1026                         goto out_unlock0;
1027                 }
1028
1029                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1030                         kuid_t euid = current_euid();
1031                         if (!uid_eq(euid, shp->shm_perm.uid) &&
1032                             !uid_eq(euid, shp->shm_perm.cuid)) {
1033                                 err = -EPERM;
1034                                 goto out_unlock0;
1035                         }
1036                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1037                                 err = -EPERM;
1038                                 goto out_unlock0;
1039                         }
1040                 }
1041
1042                 shm_file = shp->shm_file;
1043                 if (is_file_hugepages(shm_file))
1044                         goto out_unlock0;
1045
1046                 if (cmd == SHM_LOCK) {
1047                         struct user_struct *user = current_user();
1048                         err = shmem_lock(shm_file, 1, user);
1049                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1050                                 shp->shm_perm.mode |= SHM_LOCKED;
1051                                 shp->mlock_user = user;
1052                         }
1053                         goto out_unlock0;
1054                 }
1055
1056                 /* SHM_UNLOCK */
1057                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1058                         goto out_unlock0;
1059                 shmem_lock(shm_file, 0, shp->mlock_user);
1060                 shp->shm_perm.mode &= ~SHM_LOCKED;
1061                 shp->mlock_user = NULL;
1062                 get_file(shm_file);
1063                 ipc_unlock_object(&shp->shm_perm);
1064                 rcu_read_unlock();
1065                 shmem_unlock_mapping(shm_file->f_mapping);
1066
1067                 fput(shm_file);
1068                 return err;
1069         }
1070         default:
1071                 return -EINVAL;
1072         }
1073
1074 out_unlock0:
1075         ipc_unlock_object(&shp->shm_perm);
1076 out_unlock1:
1077         rcu_read_unlock();
1078         return err;
1079 }
1080
1081 /*
1082  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1083  *
1084  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1085  * "raddr" thing points to kernel space, and there has to be a wrapper around
1086  * this.
1087  */
1088 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1089               unsigned long shmlba)
1090 {
1091         struct shmid_kernel *shp;
1092         unsigned long addr;
1093         unsigned long size;
1094         struct file *file;
1095         int    err;
1096         unsigned long flags;
1097         unsigned long prot;
1098         int acc_mode;
1099         struct ipc_namespace *ns;
1100         struct shm_file_data *sfd;
1101         struct path path;
1102         fmode_t f_mode;
1103         unsigned long populate = 0;
1104
1105         err = -EINVAL;
1106         if (shmid < 0)
1107                 goto out;
1108         else if ((addr = (ulong)shmaddr)) {
1109                 if (addr & (shmlba - 1)) {
1110                         if (shmflg & SHM_RND)
1111                                 addr &= ~(shmlba - 1);     /* round down */
1112                         else
1113 #ifndef __ARCH_FORCE_SHMLBA
1114                                 if (addr & ~PAGE_MASK)
1115 #endif
1116                                         goto out;
1117                 }
1118                 flags = MAP_SHARED | MAP_FIXED;
1119         } else {
1120                 if ((shmflg & SHM_REMAP))
1121                         goto out;
1122
1123                 flags = MAP_SHARED;
1124         }
1125
1126         if (shmflg & SHM_RDONLY) {
1127                 prot = PROT_READ;
1128                 acc_mode = S_IRUGO;
1129                 f_mode = FMODE_READ;
1130         } else {
1131                 prot = PROT_READ | PROT_WRITE;
1132                 acc_mode = S_IRUGO | S_IWUGO;
1133                 f_mode = FMODE_READ | FMODE_WRITE;
1134         }
1135         if (shmflg & SHM_EXEC) {
1136                 prot |= PROT_EXEC;
1137                 acc_mode |= S_IXUGO;
1138         }
1139
1140         /*
1141          * We cannot rely on the fs check since SYSV IPC does have an
1142          * additional creator id...
1143          */
1144         ns = current->nsproxy->ipc_ns;
1145         rcu_read_lock();
1146         shp = shm_obtain_object_check(ns, shmid);
1147         if (IS_ERR(shp)) {
1148                 err = PTR_ERR(shp);
1149                 goto out_unlock;
1150         }
1151
1152         err = -EACCES;
1153         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1154                 goto out_unlock;
1155
1156         err = security_shm_shmat(shp, shmaddr, shmflg);
1157         if (err)
1158                 goto out_unlock;
1159
1160         ipc_lock_object(&shp->shm_perm);
1161
1162         /* check if shm_destroy() is tearing down shp */
1163         if (!ipc_valid_object(&shp->shm_perm)) {
1164                 ipc_unlock_object(&shp->shm_perm);
1165                 err = -EIDRM;
1166                 goto out_unlock;
1167         }
1168
1169         path = shp->shm_file->f_path;
1170         path_get(&path);
1171         shp->shm_nattch++;
1172         size = i_size_read(d_inode(path.dentry));
1173         ipc_unlock_object(&shp->shm_perm);
1174         rcu_read_unlock();
1175
1176         err = -ENOMEM;
1177         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1178         if (!sfd) {
1179                 path_put(&path);
1180                 goto out_nattch;
1181         }
1182
1183         file = alloc_file(&path, f_mode,
1184                           is_file_hugepages(shp->shm_file) ?
1185                                 &shm_file_operations_huge :
1186                                 &shm_file_operations);
1187         err = PTR_ERR(file);
1188         if (IS_ERR(file)) {
1189                 kfree(sfd);
1190                 path_put(&path);
1191                 goto out_nattch;
1192         }
1193
1194         file->private_data = sfd;
1195         file->f_mapping = shp->shm_file->f_mapping;
1196         sfd->id = shp->shm_perm.id;
1197         sfd->ns = get_ipc_ns(ns);
1198         sfd->file = shp->shm_file;
1199         sfd->vm_ops = NULL;
1200
1201         err = security_mmap_file(file, prot, flags);
1202         if (err)
1203                 goto out_fput;
1204
1205         if (down_write_killable(&current->mm->mmap_sem)) {
1206                 err = -EINTR;
1207                 goto out_fput;
1208         }
1209
1210         if (addr && !(shmflg & SHM_REMAP)) {
1211                 err = -EINVAL;
1212                 if (addr + size < addr)
1213                         goto invalid;
1214
1215                 if (find_vma_intersection(current->mm, addr, addr + size))
1216                         goto invalid;
1217         }
1218
1219         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1220         *raddr = addr;
1221         err = 0;
1222         if (IS_ERR_VALUE(addr))
1223                 err = (long)addr;
1224 invalid:
1225         up_write(&current->mm->mmap_sem);
1226         if (populate)
1227                 mm_populate(addr, populate);
1228
1229 out_fput:
1230         fput(file);
1231
1232 out_nattch:
1233         down_write(&shm_ids(ns).rwsem);
1234         shp = shm_lock(ns, shmid);
1235         shp->shm_nattch--;
1236         if (shm_may_destroy(ns, shp))
1237                 shm_destroy(ns, shp);
1238         else
1239                 shm_unlock(shp);
1240         up_write(&shm_ids(ns).rwsem);
1241         return err;
1242
1243 out_unlock:
1244         rcu_read_unlock();
1245 out:
1246         return err;
1247 }
1248
1249 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1250 {
1251         unsigned long ret;
1252         long err;
1253
1254         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1255         if (err)
1256                 return err;
1257         force_successful_syscall_return();
1258         return (long)ret;
1259 }
1260
1261 /*
1262  * detach and kill segment if marked destroyed.
1263  * The work is done in shm_close.
1264  */
1265 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1266 {
1267         struct mm_struct *mm = current->mm;
1268         struct vm_area_struct *vma;
1269         unsigned long addr = (unsigned long)shmaddr;
1270         int retval = -EINVAL;
1271 #ifdef CONFIG_MMU
1272         loff_t size = 0;
1273         struct file *file;
1274         struct vm_area_struct *next;
1275 #endif
1276
1277         if (addr & ~PAGE_MASK)
1278                 return retval;
1279
1280         if (down_write_killable(&mm->mmap_sem))
1281                 return -EINTR;
1282
1283         /*
1284          * This function tries to be smart and unmap shm segments that
1285          * were modified by partial mlock or munmap calls:
1286          * - It first determines the size of the shm segment that should be
1287          *   unmapped: It searches for a vma that is backed by shm and that
1288          *   started at address shmaddr. It records it's size and then unmaps
1289          *   it.
1290          * - Then it unmaps all shm vmas that started at shmaddr and that
1291          *   are within the initially determined size and that are from the
1292          *   same shm segment from which we determined the size.
1293          * Errors from do_munmap are ignored: the function only fails if
1294          * it's called with invalid parameters or if it's called to unmap
1295          * a part of a vma. Both calls in this function are for full vmas,
1296          * the parameters are directly copied from the vma itself and always
1297          * valid - therefore do_munmap cannot fail. (famous last words?)
1298          */
1299         /*
1300          * If it had been mremap()'d, the starting address would not
1301          * match the usual checks anyway. So assume all vma's are
1302          * above the starting address given.
1303          */
1304         vma = find_vma(mm, addr);
1305
1306 #ifdef CONFIG_MMU
1307         while (vma) {
1308                 next = vma->vm_next;
1309
1310                 /*
1311                  * Check if the starting address would match, i.e. it's
1312                  * a fragment created by mprotect() and/or munmap(), or it
1313                  * otherwise it starts at this address with no hassles.
1314                  */
1315                 if ((vma->vm_ops == &shm_vm_ops) &&
1316                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1317
1318                         /*
1319                          * Record the file of the shm segment being
1320                          * unmapped.  With mremap(), someone could place
1321                          * page from another segment but with equal offsets
1322                          * in the range we are unmapping.
1323                          */
1324                         file = vma->vm_file;
1325                         size = i_size_read(file_inode(vma->vm_file));
1326                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1327                         /*
1328                          * We discovered the size of the shm segment, so
1329                          * break out of here and fall through to the next
1330                          * loop that uses the size information to stop
1331                          * searching for matching vma's.
1332                          */
1333                         retval = 0;
1334                         vma = next;
1335                         break;
1336                 }
1337                 vma = next;
1338         }
1339
1340         /*
1341          * We need look no further than the maximum address a fragment
1342          * could possibly have landed at. Also cast things to loff_t to
1343          * prevent overflows and make comparisons vs. equal-width types.
1344          */
1345         size = PAGE_ALIGN(size);
1346         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1347                 next = vma->vm_next;
1348
1349                 /* finding a matching vma now does not alter retval */
1350                 if ((vma->vm_ops == &shm_vm_ops) &&
1351                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1352                     (vma->vm_file == file))
1353                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1354                 vma = next;
1355         }
1356
1357 #else /* CONFIG_MMU */
1358         /* under NOMMU conditions, the exact address to be destroyed must be
1359          * given */
1360         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1361                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1362                 retval = 0;
1363         }
1364
1365 #endif
1366
1367         up_write(&mm->mmap_sem);
1368         return retval;
1369 }
1370
1371 #ifdef CONFIG_PROC_FS
1372 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1373 {
1374         struct user_namespace *user_ns = seq_user_ns(s);
1375         struct shmid_kernel *shp = it;
1376         unsigned long rss = 0, swp = 0;
1377
1378         shm_add_rss_swap(shp, &rss, &swp);
1379
1380 #if BITS_PER_LONG <= 32
1381 #define SIZE_SPEC "%10lu"
1382 #else
1383 #define SIZE_SPEC "%21lu"
1384 #endif
1385
1386         seq_printf(s,
1387                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1388                    "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1389                    SIZE_SPEC " " SIZE_SPEC "\n",
1390                    shp->shm_perm.key,
1391                    shp->shm_perm.id,
1392                    shp->shm_perm.mode,
1393                    shp->shm_segsz,
1394                    shp->shm_cprid,
1395                    shp->shm_lprid,
1396                    shp->shm_nattch,
1397                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1398                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1399                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1400                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1401                    shp->shm_atim,
1402                    shp->shm_dtim,
1403                    shp->shm_ctim,
1404                    rss * PAGE_SIZE,
1405                    swp * PAGE_SIZE);
1406
1407         return 0;
1408 }
1409 #endif