ipc/shm.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/ipc/shm.c
   4  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   5  *       Many improvements/fixes by Bruno Haible.
   6  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   7  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   8  *
   9  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  10  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  11  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  12  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  13  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  14  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  15  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  *
  24  * Better ipc lock (kern_ipc_perm.lock) handling
  25  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
  26  */
  27
  28 #include <linux/slab.h>
  29 #include <linux/mm.h>
  30 #include <linux/hugetlb.h>
  31 #include <linux/shm.h>
  32 #include <linux/init.h>
  33 #include <linux/file.h>
  34 #include <linux/mman.h>
  35 #include <linux/shmem_fs.h>
  36 #include <linux/security.h>
  37 #include <linux/syscalls.h>
  38 #include <linux/audit.h>
  39 #include <linux/capability.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/seq_file.h>
  42 #include <linux/rwsem.h>
  43 #include <linux/nsproxy.h>
  44 #include <linux/mount.h>
  45 #include <linux/ipc_namespace.h>
  46 #include <linux/rhashtable.h>
  47
  48 #include <linux/uaccess.h>
  49
  50 #include "util.h"
  51
  52 struct shmid_kernel /* private to the kernel */
  53 {
  54         struct kern_ipc_perm    shm_perm;
  55         struct file             *shm_file;
  56         unsigned long           shm_nattch;
  57         unsigned long           shm_segsz;
  58         time64_t                shm_atim;
  59         time64_t                shm_dtim;
  60         time64_t                shm_ctim;
  61         struct pid              *shm_cprid;
  62         struct pid              *shm_lprid;
  63         struct user_struct      *mlock_user;
  64
  65         /*
  66          * The task created the shm object, for
  67          * task_lock(shp->shm_creator)
  68          */
  69         struct task_struct      *shm_creator;
  70
  71         /*
  72          * List by creator. task_lock(->shm_creator) required for read/write.
  73          * If list_empty(), then the creator is dead already.
  74          */
  75         struct list_head        shm_clist;
  76         struct ipc_namespace    *ns;
  77 } __randomize_layout;
  78
  79 /* shm_mode upper byte flags */
  80 #define SHM_DEST        01000   /* segment will be destroyed on last detach */
  81 #define SHM_LOCKED      02000   /* segment will not be swapped */
  82
  83 struct shm_file_data {
  84         int id;
  85         struct ipc_namespace *ns;
  86         struct file *file;
  87         const struct vm_operations_struct *vm_ops;
  88 };
  89
  90 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  91
  92 static const struct file_operations shm_file_operations;
  93 static const struct vm_operations_struct shm_vm_ops;
  94
  95 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  96
  97 #define shm_unlock(shp)                 \
  98         ipc_unlock(&(shp)->shm_perm)
  99
 100 static int newseg(struct ipc_namespace *, struct ipc_params *);
 101 static void shm_open(struct vm_area_struct *vma);
 102 static void shm_close(struct vm_area_struct *vma);
 103 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
 104 #ifdef CONFIG_PROC_FS
 105 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
 106 #endif
 107
 108 void shm_init_ns(struct ipc_namespace *ns)
 109 {
 110         ns->shm_ctlmax = SHMMAX;
 111         ns->shm_ctlall = SHMALL;
 112         ns->shm_ctlmni = SHMMNI;
 113         ns->shm_rmid_forced = 0;
 114         ns->shm_tot = 0;
 115         ipc_init_ids(&shm_ids(ns));
 116 }
 117
 118 /*
 119  * Called with shm_ids.rwsem (writer) and the shp structure locked.
 120  * Only shm_ids.rwsem remains locked on exit.
 121  */
 122 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 123 {
 124         struct shmid_kernel *shp;
 125
 126         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 127         WARN_ON(ns != shp->ns);
 128
 129         if (shp->shm_nattch) {
 130                 shp->shm_perm.mode |= SHM_DEST;
 131                 /* Do not find it any more */
 132                 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
 133                 shm_unlock(shp);
 134         } else
 135                 shm_destroy(ns, shp);
 136 }
 137
 138 #ifdef CONFIG_IPC_NS
 139 void shm_exit_ns(struct ipc_namespace *ns)
 140 {
 141         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 142         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 143         rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
 144 }
 145 #endif
 146
 147 static int __init ipc_ns_init(void)
 148 {
 149         shm_init_ns(&init_ipc_ns);
 150         return 0;
 151 }
 152
 153 pure_initcall(ipc_ns_init);
 154
 155 void __init shm_init(void)
 156 {
 157         ipc_init_proc_interface("sysvipc/shm",
 158 #if BITS_PER_LONG <= 32
 159                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 160 #else
 161                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 162 #endif
 163                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 164 }
 165
 166 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 167 {
 168         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 169
 170         if (IS_ERR(ipcp))
 171                 return ERR_CAST(ipcp);
 172
 173         return container_of(ipcp, struct shmid_kernel, shm_perm);
 174 }
 175
 176 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 177 {
 178         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 179
 180         if (IS_ERR(ipcp))
 181                 return ERR_CAST(ipcp);
 182
 183         return container_of(ipcp, struct shmid_kernel, shm_perm);
 184 }
 185
 186 /*
 187  * shm_lock_(check_) routines are called in the paths where the rwsem
 188  * is not necessarily held.
 189  */
 190 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 191 {
 192         struct kern_ipc_perm *ipcp;
 193
 194         rcu_read_lock();
 195         ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 196         if (IS_ERR(ipcp))
 197                 goto err;
 198
 199         ipc_lock_object(ipcp);
 200         /*
 201          * ipc_rmid() may have already freed the ID while ipc_lock_object()
 202          * was spinning: here verify that the structure is still valid.
 203          * Upon races with RMID, return -EIDRM, thus indicating that
 204          * the ID points to a removed identifier.
 205          */
 206         if (ipc_valid_object(ipcp)) {
 207                 /* return a locked ipc object upon success */
 208                 return container_of(ipcp, struct shmid_kernel, shm_perm);
 209         }
 210
 211         ipc_unlock_object(ipcp);
 212         ipcp = ERR_PTR(-EIDRM);
 213 err:
 214         rcu_read_unlock();
 215         /*
 216          * Callers of shm_lock() must validate the status of the returned ipc
 217          * object pointer and error out as appropriate.
 218          */
 219         return ERR_CAST(ipcp);
 220 }
 221
 222 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 223 {
 224         rcu_read_lock();
 225         ipc_lock_object(&ipcp->shm_perm);
 226 }
 227
 228 static void shm_rcu_free(struct rcu_head *head)
 229 {
 230         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 231                                                         rcu);
 232         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 233                                                         shm_perm);
 234         security_shm_free(&shp->shm_perm);
 235         kvfree(shp);
 236 }
 237
 238 /*
 239  * It has to be called with shp locked.
 240  * It must be called before ipc_rmid()
 241  */
 242 static inline void shm_clist_rm(struct shmid_kernel *shp)
 243 {
 244         struct task_struct *creator;
 245
 246         /* ensure that shm_creator does not disappear */
 247         rcu_read_lock();
 248
 249         /*
 250          * A concurrent exit_shm may do a list_del_init() as well.
 251          * Just do nothing if exit_shm already did the work
 252          */
 253         if (!list_empty(&shp->shm_clist)) {
 254                 /*
 255                  * shp->shm_creator is guaranteed to be valid *only*
 256                  * if shp->shm_clist is not empty.
 257                  */
 258                 creator = shp->shm_creator;
 259
 260                 task_lock(creator);
 261                 /*
 262                  * list_del_init() is a nop if the entry was already removed
 263                  * from the list.
 264                  */
 265                 list_del_init(&shp->shm_clist);
 266                 task_unlock(creator);
 267         }
 268         rcu_read_unlock();
 269 }
 270
 271 static inline void shm_rmid(struct shmid_kernel *s)
 272 {
 273         shm_clist_rm(s);
 274         ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
 275 }
 276
 277
 278 static int __shm_open(struct vm_area_struct *vma)
 279 {
 280         struct file *file = vma->vm_file;
 281         struct shm_file_data *sfd = shm_file_data(file);
 282         struct shmid_kernel *shp;
 283
 284         shp = shm_lock(sfd->ns, sfd->id);
 285
 286         if (IS_ERR(shp))
 287                 return PTR_ERR(shp);
 288
 289         if (shp->shm_file != sfd->file) {
 290                 /* ID was reused */
 291                 shm_unlock(shp);
 292                 return -EINVAL;
 293         }
 294
 295         shp->shm_atim = ktime_get_real_seconds();
 296         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 297         shp->shm_nattch++;
 298         shm_unlock(shp);
 299         return 0;
 300 }
 301
 302 /* This is called by fork, once for every shm attach. */
 303 static void shm_open(struct vm_area_struct *vma)
 304 {
 305         int err = __shm_open(vma);
 306         /*
 307          * We raced in the idr lookup or with shm_destroy().
 308          * Either way, the ID is busted.
 309          */
 310         WARN_ON_ONCE(err);
 311 }
 312
 313 /*
 314  * shm_destroy - free the struct shmid_kernel
 315  *
 316  * @ns: namespace
 317  * @shp: struct to free
 318  *
 319  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 320  * but returns with shp unlocked and freed.
 321  */
 322 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 323 {
 324         struct file *shm_file;
 325
 326         shm_file = shp->shm_file;
 327         shp->shm_file = NULL;
 328         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 329         shm_rmid(shp);
 330         shm_unlock(shp);
 331         if (!is_file_hugepages(shm_file))
 332                 shmem_lock(shm_file, 0, shp->mlock_user);
 333         else if (shp->mlock_user)
 334                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 335                                 shp->mlock_user);
 336         fput(shm_file);
 337         ipc_update_pid(&shp->shm_cprid, NULL);
 338         ipc_update_pid(&shp->shm_lprid, NULL);
 339         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 340 }
 341
 342 /*
 343  * shm_may_destroy - identifies whether shm segment should be destroyed now
 344  *
 345  * Returns true if and only if there are no active users of the segment and
 346  * one of the following is true:
 347  *
 348  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 349  *
 350  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 351  */
 352 static bool shm_may_destroy(struct shmid_kernel *shp)
 353 {
 354         return (shp->shm_nattch == 0) &&
 355                (shp->ns->shm_rmid_forced ||
 356                 (shp->shm_perm.mode & SHM_DEST));
 357 }
 358
 359 /*
 360  * remove the attach descriptor vma.
 361  * free memory for segment if it is marked destroyed.
 362  * The descriptor has already been removed from the current->mm->mmap list
 363  * and will later be kfree()d.
 364  */
 365 static void shm_close(struct vm_area_struct *vma)
 366 {
 367         struct file *file = vma->vm_file;
 368         struct shm_file_data *sfd = shm_file_data(file);
 369         struct shmid_kernel *shp;
 370         struct ipc_namespace *ns = sfd->ns;
 371
 372         down_write(&shm_ids(ns).rwsem);
 373         /* remove from the list of attaches of the shm segment */
 374         shp = shm_lock(ns, sfd->id);
 375
 376         /*
 377          * We raced in the idr lookup or with shm_destroy().
 378          * Either way, the ID is busted.
 379          */
 380         if (WARN_ON_ONCE(IS_ERR(shp)))
 381                 goto done; /* no-op */
 382
 383         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 384         shp->shm_dtim = ktime_get_real_seconds();
 385         shp->shm_nattch--;
 386         if (shm_may_destroy(shp))
 387                 shm_destroy(ns, shp);
 388         else
 389                 shm_unlock(shp);
 390 done:
 391         up_write(&shm_ids(ns).rwsem);
 392 }
 393
 394 /* Called with ns->shm_ids(ns).rwsem locked */
 395 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 396 {
 397         struct ipc_namespace *ns = data;
 398         struct kern_ipc_perm *ipcp = p;
 399         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 400
 401         /*
 402          * We want to destroy segments without users and with already
 403          * exit'ed originating process.
 404          *
 405          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 406          */
 407         if (!list_empty(&shp->shm_clist))
 408                 return 0;
 409
 410         if (shm_may_destroy(shp)) {
 411                 shm_lock_by_ptr(shp);
 412                 shm_destroy(ns, shp);
 413         }
 414         return 0;
 415 }
 416
 417 void shm_destroy_orphaned(struct ipc_namespace *ns)
 418 {
 419         down_write(&shm_ids(ns).rwsem);
 420         if (shm_ids(ns).in_use)
 421                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 422         up_write(&shm_ids(ns).rwsem);
 423 }
 424
 425 /* Locking assumes this will only be called with task == current */
 426 void exit_shm(struct task_struct *task)
 427 {
 428         for (;;) {
 429                 struct shmid_kernel *shp;
 430                 struct ipc_namespace *ns;
 431
 432                 task_lock(task);
 433
 434                 if (list_empty(&task->sysvshm.shm_clist)) {
 435                         task_unlock(task);
 436                         break;
 437                 }
 438
 439                 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
 440                                 shm_clist);
 441
 442                 /*
 443                  * 1) Get pointer to the ipc namespace. It is worth to say
 444                  * that this pointer is guaranteed to be valid because
 445                  * shp lifetime is always shorter than namespace lifetime
 446                  * in which shp lives.
 447                  * We taken task_lock it means that shp won't be freed.
 448                  */
 449                 ns = shp->ns;
 450
 451                 /*
 452                  * 2) If kernel.shm_rmid_forced is not set then only keep track of
 453                  * which shmids are orphaned, so that a later set of the sysctl
 454                  * can clean them up.
 455                  */
 456                 if (!ns->shm_rmid_forced)
 457                         goto unlink_continue;
 458
 459                 /*
 460                  * 3) get a reference to the namespace.
 461                  *    The refcount could be already 0. If it is 0, then
 462                  *    the shm objects will be free by free_ipc_work().
 463                  */
 464                 ns = get_ipc_ns_not_zero(ns);
 465                 if (!ns) {
 466 unlink_continue:
 467                         list_del_init(&shp->shm_clist);
 468                         task_unlock(task);
 469                         continue;
 470                 }
 471
 472                 /*
 473                  * 4) get a reference to shp.
 474                  *   This cannot fail: shm_clist_rm() is called before
 475                  *   ipc_rmid(), thus the refcount cannot be 0.
 476                  */
 477                 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
 478
 479                 /*
 480                  * 5) unlink the shm segment from the list of segments
 481                  *    created by current.
 482                  *    This must be done last. After unlinking,
 483                  *    only the refcounts obtained above prevent IPC_RMID
 484                  *    from destroying the segment or the namespace.
 485                  */
 486                 list_del_init(&shp->shm_clist);
 487
 488                 task_unlock(task);
 489
 490                 /*
 491                  * 6) we have all references
 492                  *    Thus lock & if needed destroy shp.
 493                  */
 494                 down_write(&shm_ids(ns).rwsem);
 495                 shm_lock_by_ptr(shp);
 496                 /*
 497                  * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
 498                  * safe to call ipc_rcu_putref here
 499                  */
 500                 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 501
 502                 if (ipc_valid_object(&shp->shm_perm)) {
 503                         if (shm_may_destroy(shp))
 504                                 shm_destroy(ns, shp);
 505                         else
 506                                 shm_unlock(shp);
 507                 } else {
 508                         /*
 509                          * Someone else deleted the shp from namespace
 510                          * idr/kht while we have waited.
 511                          * Just unlock and continue.
 512                          */
 513                         shm_unlock(shp);
 514                 }
 515
 516                 up_write(&shm_ids(ns).rwsem);
 517                 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
 518         }
 519 }
 520
 521 static vm_fault_t shm_fault(struct vm_fault *vmf)
 522 {
 523         struct file *file = vmf->vma->vm_file;
 524         struct shm_file_data *sfd = shm_file_data(file);
 525
 526         return sfd->vm_ops->fault(vmf);
 527 }
 528
 529 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
 530 {
 531         struct file *file = vma->vm_file;
 532         struct shm_file_data *sfd = shm_file_data(file);
 533
 534         if (sfd->vm_ops->split)
 535                 return sfd->vm_ops->split(vma, addr);
 536
 537         return 0;
 538 }
 539
 540 static unsigned long shm_pagesize(struct vm_area_struct *vma)
 541 {
 542         struct file *file = vma->vm_file;
 543         struct shm_file_data *sfd = shm_file_data(file);
 544
 545         if (sfd->vm_ops->pagesize)
 546                 return sfd->vm_ops->pagesize(vma);
 547
 548         return PAGE_SIZE;
 549 }
 550
 551 #ifdef CONFIG_NUMA
 552 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 553 {
 554         struct file *file = vma->vm_file;
 555         struct shm_file_data *sfd = shm_file_data(file);
 556         int err = 0;
 557
 558         if (sfd->vm_ops->set_policy)
 559                 err = sfd->vm_ops->set_policy(vma, new);
 560         return err;
 561 }
 562
 563 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 564                                         unsigned long addr)
 565 {
 566         struct file *file = vma->vm_file;
 567         struct shm_file_data *sfd = shm_file_data(file);
 568         struct mempolicy *pol = NULL;
 569
 570         if (sfd->vm_ops->get_policy)
 571                 pol = sfd->vm_ops->get_policy(vma, addr);
 572         else if (vma->vm_policy)
 573                 pol = vma->vm_policy;
 574
 575         return pol;
 576 }
 577 #endif
 578
 579 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 580 {
 581         struct shm_file_data *sfd = shm_file_data(file);
 582         int ret;
 583
 584         /*
 585          * In case of remap_file_pages() emulation, the file can represent an
 586          * IPC ID that was removed, and possibly even reused by another shm
 587          * segment already.  Propagate this case as an error to caller.
 588          */
 589         ret = __shm_open(vma);
 590         if (ret)
 591                 return ret;
 592
 593         ret = call_mmap(sfd->file, vma);
 594         if (ret) {
 595                 shm_close(vma);
 596                 return ret;
 597         }
 598         sfd->vm_ops = vma->vm_ops;
 599 #ifdef CONFIG_MMU
 600         WARN_ON(!sfd->vm_ops->fault);
 601 #endif
 602         vma->vm_ops = &shm_vm_ops;
 603         return 0;
 604 }
 605
 606 static int shm_release(struct inode *ino, struct file *file)
 607 {
 608         struct shm_file_data *sfd = shm_file_data(file);
 609
 610         put_ipc_ns(sfd->ns);
 611         fput(sfd->file);
 612         shm_file_data(file) = NULL;
 613         kfree(sfd);
 614         return 0;
 615 }
 616
 617 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 618 {
 619         struct shm_file_data *sfd = shm_file_data(file);
 620
 621         if (!sfd->file->f_op->fsync)
 622                 return -EINVAL;
 623         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 624 }
 625
 626 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 627                           loff_t len)
 628 {
 629         struct shm_file_data *sfd = shm_file_data(file);
 630
 631         if (!sfd->file->f_op->fallocate)
 632                 return -EOPNOTSUPP;
 633         return sfd->file->f_op->fallocate(file, mode, offset, len);
 634 }
 635
 636 static unsigned long shm_get_unmapped_area(struct file *file,
 637         unsigned long addr, unsigned long len, unsigned long pgoff,
 638         unsigned long flags)
 639 {
 640         struct shm_file_data *sfd = shm_file_data(file);
 641
 642         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 643                                                 pgoff, flags);
 644 }
 645
 646 static const struct file_operations shm_file_operations = {
 647         .mmap           = shm_mmap,
 648         .fsync          = shm_fsync,
 649         .release        = shm_release,
 650         .get_unmapped_area      = shm_get_unmapped_area,
 651         .llseek         = noop_llseek,
 652         .fallocate      = shm_fallocate,
 653 };
 654
 655 /*
 656  * shm_file_operations_huge is now identical to shm_file_operations,
 657  * but we keep it distinct for the sake of is_file_shm_hugepages().
 658  */
 659 static const struct file_operations shm_file_operations_huge = {
 660         .mmap           = shm_mmap,
 661         .fsync          = shm_fsync,
 662         .release        = shm_release,
 663         .get_unmapped_area      = shm_get_unmapped_area,
 664         .llseek         = noop_llseek,
 665         .fallocate      = shm_fallocate,
 666 };
 667
 668 bool is_file_shm_hugepages(struct file *file)
 669 {
 670         return file->f_op == &shm_file_operations_huge;
 671 }
 672
 673 static const struct vm_operations_struct shm_vm_ops = {
 674         .open   = shm_open,     /* callback for a new vm-area open */
 675         .close  = shm_close,    /* callback for when the vm-area is released */
 676         .fault  = shm_fault,
 677         .split  = shm_split,
 678         .pagesize = shm_pagesize,
 679 #if defined(CONFIG_NUMA)
 680         .set_policy = shm_set_policy,
 681         .get_policy = shm_get_policy,
 682 #endif
 683 };
 684
 685 /**
 686  * newseg - Create a new shared memory segment
 687  * @ns: namespace
 688  * @params: ptr to the structure that contains key, size and shmflg
 689  *
 690  * Called with shm_ids.rwsem held as a writer.
 691  */
 692 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 693 {
 694         key_t key = params->key;
 695         int shmflg = params->flg;
 696         size_t size = params->u.size;
 697         int error;
 698         struct shmid_kernel *shp;
 699         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 700         struct file *file;
 701         char name[13];
 702         vm_flags_t acctflag = 0;
 703
 704         if (size < SHMMIN || size > ns->shm_ctlmax)
 705                 return -EINVAL;
 706
 707         if (numpages << PAGE_SHIFT < size)
 708                 return -ENOSPC;
 709
 710         if (ns->shm_tot + numpages < ns->shm_tot ||
 711                         ns->shm_tot + numpages > ns->shm_ctlall)
 712                 return -ENOSPC;
 713
 714         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
 715         if (unlikely(!shp))
 716                 return -ENOMEM;
 717
 718         shp->shm_perm.key = key;
 719         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 720         shp->mlock_user = NULL;
 721
 722         shp->shm_perm.security = NULL;
 723         error = security_shm_alloc(&shp->shm_perm);
 724         if (error) {
 725                 kvfree(shp);
 726                 return error;
 727         }
 728
 729         sprintf(name, "SYSV%08x", key);
 730         if (shmflg & SHM_HUGETLB) {
 731                 struct hstate *hs;
 732                 size_t hugesize;
 733
 734                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 735                 if (!hs) {
 736                         error = -EINVAL;
 737                         goto no_file;
 738                 }
 739                 hugesize = ALIGN(size, huge_page_size(hs));
 740
 741                 /* hugetlb_file_setup applies strict accounting */
 742                 if (shmflg & SHM_NORESERVE)
 743                         acctflag = VM_NORESERVE;
 744                 file = hugetlb_file_setup(name, hugesize, acctflag,
 745                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 746                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 747         } else {
 748                 /*
 749                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 750                  * if it's asked for.
 751                  */
 752                 if  ((shmflg & SHM_NORESERVE) &&
 753                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 754                         acctflag = VM_NORESERVE;
 755                 file = shmem_kernel_file_setup(name, size, acctflag);
 756         }
 757         error = PTR_ERR(file);
 758         if (IS_ERR(file))
 759                 goto no_file;
 760
 761         shp->shm_cprid = get_pid(task_tgid(current));
 762         shp->shm_lprid = NULL;
 763         shp->shm_atim = shp->shm_dtim = 0;
 764         shp->shm_ctim = ktime_get_real_seconds();
 765         shp->shm_segsz = size;
 766         shp->shm_nattch = 0;
 767         shp->shm_file = file;
 768         shp->shm_creator = current;
 769
 770         /* ipc_addid() locks shp upon success. */
 771         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 772         if (error < 0)
 773                 goto no_id;
 774
 775         shp->ns = ns;
 776
 777         task_lock(current);
 778         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 779         task_unlock(current);
 780
 781         /*
 782          * shmid gets reported as "inode#" in /proc/pid/maps.
 783          * proc-ps tools use this. Changing this will break them.
 784          */
 785         file_inode(file)->i_ino = shp->shm_perm.id;
 786
 787         ns->shm_tot += numpages;
 788         error = shp->shm_perm.id;
 789
 790         ipc_unlock_object(&shp->shm_perm);
 791         rcu_read_unlock();
 792         return error;
 793
 794 no_id:
 795         ipc_update_pid(&shp->shm_cprid, NULL);
 796         ipc_update_pid(&shp->shm_lprid, NULL);
 797         if (is_file_hugepages(file) && shp->mlock_user)
 798                 user_shm_unlock(size, shp->mlock_user);
 799         fput(file);
 800         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 801         return error;
 802 no_file:
 803         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 804         return error;
 805 }
 806
 807 /*
 808  * Called with shm_ids.rwsem and ipcp locked.
 809  */
 810 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 811                                 struct ipc_params *params)
 812 {
 813         struct shmid_kernel *shp;
 814
 815         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 816         if (shp->shm_segsz < params->u.size)
 817                 return -EINVAL;
 818
 819         return 0;
 820 }
 821
 822 long ksys_shmget(key_t key, size_t size, int shmflg)
 823 {
 824         struct ipc_namespace *ns;
 825         static const struct ipc_ops shm_ops = {
 826                 .getnew = newseg,
 827                 .associate = security_shm_associate,
 828                 .more_checks = shm_more_checks,
 829         };
 830         struct ipc_params shm_params;
 831
 832         ns = current->nsproxy->ipc_ns;
 833
 834         shm_params.key = key;
 835         shm_params.flg = shmflg;
 836         shm_params.u.size = size;
 837
 838         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 839 }
 840
 841 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 842 {
 843         return ksys_shmget(key, size, shmflg);
 844 }
 845
 846 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 847 {
 848         switch (version) {
 849         case IPC_64:
 850                 return copy_to_user(buf, in, sizeof(*in));
 851         case IPC_OLD:
 852             {
 853                 struct shmid_ds out;
 854
 855                 memset(&out, 0, sizeof(out));
 856                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 857                 out.shm_segsz   = in->shm_segsz;
 858                 out.shm_atime   = in->shm_atime;
 859                 out.shm_dtime   = in->shm_dtime;
 860                 out.shm_ctime   = in->shm_ctime;
 861                 out.shm_cpid    = in->shm_cpid;
 862                 out.shm_lpid    = in->shm_lpid;
 863                 out.shm_nattch  = in->shm_nattch;
 864
 865                 return copy_to_user(buf, &out, sizeof(out));
 866             }
 867         default:
 868                 return -EINVAL;
 869         }
 870 }
 871
 872 static inline unsigned long
 873 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 874 {
 875         switch (version) {
 876         case IPC_64:
 877                 if (copy_from_user(out, buf, sizeof(*out)))
 878                         return -EFAULT;
 879                 return 0;
 880         case IPC_OLD:
 881             {
 882                 struct shmid_ds tbuf_old;
 883
 884                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 885                         return -EFAULT;
 886
 887                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 888                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 889                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 890
 891                 return 0;
 892             }
 893         default:
 894                 return -EINVAL;
 895         }
 896 }
 897
 898 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 899 {
 900         switch (version) {
 901         case IPC_64:
 902                 return copy_to_user(buf, in, sizeof(*in));
 903         case IPC_OLD:
 904             {
 905                 struct shminfo out;
 906
 907                 if (in->shmmax > INT_MAX)
 908                         out.shmmax = INT_MAX;
 909                 else
 910                         out.shmmax = (int)in->shmmax;
 911
 912                 out.shmmin      = in->shmmin;
 913                 out.shmmni      = in->shmmni;
 914                 out.shmseg      = in->shmseg;
 915                 out.shmall      = in->shmall;
 916
 917                 return copy_to_user(buf, &out, sizeof(out));
 918             }
 919         default:
 920                 return -EINVAL;
 921         }
 922 }
 923
 924 /*
 925  * Calculate and add used RSS and swap pages of a shm.
 926  * Called with shm_ids.rwsem held as a reader
 927  */
 928 static void shm_add_rss_swap(struct shmid_kernel *shp,
 929         unsigned long *rss_add, unsigned long *swp_add)
 930 {
 931         struct inode *inode;
 932
 933         inode = file_inode(shp->shm_file);
 934
 935         if (is_file_hugepages(shp->shm_file)) {
 936                 struct address_space *mapping = inode->i_mapping;
 937                 struct hstate *h = hstate_file(shp->shm_file);
 938                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 939         } else {
 940 #ifdef CONFIG_SHMEM
 941                 struct shmem_inode_info *info = SHMEM_I(inode);
 942
 943                 spin_lock_irq(&info->lock);
 944                 *rss_add += inode->i_mapping->nrpages;
 945                 *swp_add += info->swapped;
 946                 spin_unlock_irq(&info->lock);
 947 #else
 948                 *rss_add += inode->i_mapping->nrpages;
 949 #endif
 950         }
 951 }
 952
 953 /*
 954  * Called with shm_ids.rwsem held as a reader
 955  */
 956 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 957                 unsigned long *swp)
 958 {
 959         int next_id;
 960         int total, in_use;
 961
 962         *rss = 0;
 963         *swp = 0;
 964
 965         in_use = shm_ids(ns).in_use;
 966
 967         for (total = 0, next_id = 0; total < in_use; next_id++) {
 968                 struct kern_ipc_perm *ipc;
 969                 struct shmid_kernel *shp;
 970
 971                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 972                 if (ipc == NULL)
 973                         continue;
 974                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 975
 976                 shm_add_rss_swap(shp, rss, swp);
 977
 978                 total++;
 979         }
 980 }
 981
 982 /*
 983  * This function handles some shmctl commands which require the rwsem
 984  * to be held in write mode.
 985  * NOTE: no locks must be held, the rwsem is taken inside this function.
 986  */
 987 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 988                        struct shmid64_ds *shmid64)
 989 {
 990         struct kern_ipc_perm *ipcp;
 991         struct shmid_kernel *shp;
 992         int err;
 993
 994         down_write(&shm_ids(ns).rwsem);
 995         rcu_read_lock();
 996
 997         ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
 998                                       &shmid64->shm_perm, 0);
 999         if (IS_ERR(ipcp)) {
1000                 err = PTR_ERR(ipcp);
1001                 goto out_unlock1;
1002         }
1003
1004         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1005
1006         err = security_shm_shmctl(&shp->shm_perm, cmd);
1007         if (err)
1008                 goto out_unlock1;
1009
1010         switch (cmd) {
1011         case IPC_RMID:
1012                 ipc_lock_object(&shp->shm_perm);
1013                 /* do_shm_rmid unlocks the ipc object and rcu */
1014                 do_shm_rmid(ns, ipcp);
1015                 goto out_up;
1016         case IPC_SET:
1017                 ipc_lock_object(&shp->shm_perm);
1018                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1019                 if (err)
1020                         goto out_unlock0;
1021                 shp->shm_ctim = ktime_get_real_seconds();
1022                 break;
1023         default:
1024                 err = -EINVAL;
1025                 goto out_unlock1;
1026         }
1027
1028 out_unlock0:
1029         ipc_unlock_object(&shp->shm_perm);
1030 out_unlock1:
1031         rcu_read_unlock();
1032 out_up:
1033         up_write(&shm_ids(ns).rwsem);
1034         return err;
1035 }
1036
1037 static int shmctl_ipc_info(struct ipc_namespace *ns,
1038                            struct shminfo64 *shminfo)
1039 {
1040         int err = security_shm_shmctl(NULL, IPC_INFO);
1041         if (!err) {
1042                 memset(shminfo, 0, sizeof(*shminfo));
1043                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1044                 shminfo->shmmax = ns->shm_ctlmax;
1045                 shminfo->shmall = ns->shm_ctlall;
1046                 shminfo->shmmin = SHMMIN;
1047                 down_read(&shm_ids(ns).rwsem);
1048                 err = ipc_get_maxidx(&shm_ids(ns));
1049                 up_read(&shm_ids(ns).rwsem);
1050                 if (err < 0)
1051                         err = 0;
1052         }
1053         return err;
1054 }
1055
1056 static int shmctl_shm_info(struct ipc_namespace *ns,
1057                            struct shm_info *shm_info)
1058 {
1059         int err = security_shm_shmctl(NULL, SHM_INFO);
1060         if (!err) {
1061                 memset(shm_info, 0, sizeof(*shm_info));
1062                 down_read(&shm_ids(ns).rwsem);
1063                 shm_info->used_ids = shm_ids(ns).in_use;
1064                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1065                 shm_info->shm_tot = ns->shm_tot;
1066                 shm_info->swap_attempts = 0;
1067                 shm_info->swap_successes = 0;
1068                 err = ipc_get_maxidx(&shm_ids(ns));
1069                 up_read(&shm_ids(ns).rwsem);
1070                 if (err < 0)
1071                         err = 0;
1072         }
1073         return err;
1074 }
1075
1076 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1077                         int cmd, struct shmid64_ds *tbuf)
1078 {
1079         struct shmid_kernel *shp;
1080         int err;
1081
1082         memset(tbuf, 0, sizeof(*tbuf));
1083
1084         rcu_read_lock();
1085         if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1086                 shp = shm_obtain_object(ns, shmid);
1087                 if (IS_ERR(shp)) {
1088                         err = PTR_ERR(shp);
1089                         goto out_unlock;
1090                 }
1091         } else { /* IPC_STAT */
1092                 shp = shm_obtain_object_check(ns, shmid);
1093                 if (IS_ERR(shp)) {
1094                         err = PTR_ERR(shp);
1095                         goto out_unlock;
1096                 }
1097         }
1098
1099         /*
1100          * Semantically SHM_STAT_ANY ought to be identical to
1101          * that functionality provided by the /proc/sysvipc/
1102          * interface. As such, only audit these calls and
1103          * do not do traditional S_IRUGO permission checks on
1104          * the ipc object.
1105          */
1106         if (cmd == SHM_STAT_ANY)
1107                 audit_ipc_obj(&shp->shm_perm);
1108         else {
1109                 err = -EACCES;
1110                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1111                         goto out_unlock;
1112         }
1113
1114         err = security_shm_shmctl(&shp->shm_perm, cmd);
1115         if (err)
1116                 goto out_unlock;
1117
1118         ipc_lock_object(&shp->shm_perm);
1119
1120         if (!ipc_valid_object(&shp->shm_perm)) {
1121                 ipc_unlock_object(&shp->shm_perm);
1122                 err = -EIDRM;
1123                 goto out_unlock;
1124         }
1125
1126         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1127         tbuf->shm_segsz = shp->shm_segsz;
1128         tbuf->shm_atime = shp->shm_atim;
1129         tbuf->shm_dtime = shp->shm_dtim;
1130         tbuf->shm_ctime = shp->shm_ctim;
1131 #ifndef CONFIG_64BIT
1132         tbuf->shm_atime_high = shp->shm_atim >> 32;
1133         tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1134         tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1135 #endif
1136         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
1137         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
1138         tbuf->shm_nattch = shp->shm_nattch;
1139
1140         if (cmd == IPC_STAT) {
1141                 /*
1142                  * As defined in SUS:
1143                  * Return 0 on success
1144                  */
1145                 err = 0;
1146         } else {
1147                 /*
1148                  * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1149                  * Return the full id, including the sequence number
1150                  */
1151                 err = shp->shm_perm.id;
1152         }
1153
1154         ipc_unlock_object(&shp->shm_perm);
1155 out_unlock:
1156         rcu_read_unlock();
1157         return err;
1158 }
1159
1160 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1161 {
1162         struct shmid_kernel *shp;
1163         struct file *shm_file;
1164         int err;
1165
1166         rcu_read_lock();
1167         shp = shm_obtain_object_check(ns, shmid);
1168         if (IS_ERR(shp)) {
1169                 err = PTR_ERR(shp);
1170                 goto out_unlock1;
1171         }
1172
1173         audit_ipc_obj(&(shp->shm_perm));
1174         err = security_shm_shmctl(&shp->shm_perm, cmd);
1175         if (err)
1176                 goto out_unlock1;
1177
1178         ipc_lock_object(&shp->shm_perm);
1179
1180         /* check if shm_destroy() is tearing down shp */
1181         if (!ipc_valid_object(&shp->shm_perm)) {
1182                 err = -EIDRM;
1183                 goto out_unlock0;
1184         }
1185
1186         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1187                 kuid_t euid = current_euid();
1188
1189                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1190                     !uid_eq(euid, shp->shm_perm.cuid)) {
1191                         err = -EPERM;
1192                         goto out_unlock0;
1193                 }
1194                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1195                         err = -EPERM;
1196                         goto out_unlock0;
1197                 }
1198         }
1199
1200         shm_file = shp->shm_file;
1201         if (is_file_hugepages(shm_file))
1202                 goto out_unlock0;
1203
1204         if (cmd == SHM_LOCK) {
1205                 struct user_struct *user = current_user();
1206
1207                 err = shmem_lock(shm_file, 1, user);
1208                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1209                         shp->shm_perm.mode |= SHM_LOCKED;
1210                         shp->mlock_user = user;
1211                 }
1212                 goto out_unlock0;
1213         }
1214
1215         /* SHM_UNLOCK */
1216         if (!(shp->shm_perm.mode & SHM_LOCKED))
1217                 goto out_unlock0;
1218         shmem_lock(shm_file, 0, shp->mlock_user);
1219         shp->shm_perm.mode &= ~SHM_LOCKED;
1220         shp->mlock_user = NULL;
1221         get_file(shm_file);
1222         ipc_unlock_object(&shp->shm_perm);
1223         rcu_read_unlock();
1224         shmem_unlock_mapping(shm_file->f_mapping);
1225
1226         fput(shm_file);
1227         return err;
1228
1229 out_unlock0:
1230         ipc_unlock_object(&shp->shm_perm);
1231 out_unlock1:
1232         rcu_read_unlock();
1233         return err;
1234 }
1235
1236 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1237 {
1238         int err, version;
1239         struct ipc_namespace *ns;
1240         struct shmid64_ds sem64;
1241
1242         if (cmd < 0 || shmid < 0)
1243                 return -EINVAL;
1244
1245         version = ipc_parse_version(&cmd);
1246         ns = current->nsproxy->ipc_ns;
1247
1248         switch (cmd) {
1249         case IPC_INFO: {
1250                 struct shminfo64 shminfo;
1251                 err = shmctl_ipc_info(ns, &shminfo);
1252                 if (err < 0)
1253                         return err;
1254                 if (copy_shminfo_to_user(buf, &shminfo, version))
1255                         err = -EFAULT;
1256                 return err;
1257         }
1258         case SHM_INFO: {
1259                 struct shm_info shm_info;
1260                 err = shmctl_shm_info(ns, &shm_info);
1261                 if (err < 0)
1262                         return err;
1263                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1264                         err = -EFAULT;
1265                 return err;
1266         }
1267         case SHM_STAT:
1268         case SHM_STAT_ANY:
1269         case IPC_STAT: {
1270                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1271                 if (err < 0)
1272                         return err;
1273                 if (copy_shmid_to_user(buf, &sem64, version))
1274                         err = -EFAULT;
1275                 return err;
1276         }
1277         case IPC_SET:
1278                 if (copy_shmid_from_user(&sem64, buf, version))
1279                         return -EFAULT;
1280                 /* fallthru */
1281         case IPC_RMID:
1282                 return shmctl_down(ns, shmid, cmd, &sem64);
1283         case SHM_LOCK:
1284         case SHM_UNLOCK:
1285                 return shmctl_do_lock(ns, shmid, cmd);
1286         default:
1287                 return -EINVAL;
1288         }
1289 }
1290
1291 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1292 {
1293         return ksys_shmctl(shmid, cmd, buf);
1294 }
1295
1296 #ifdef CONFIG_COMPAT
1297
1298 struct compat_shmid_ds {
1299         struct compat_ipc_perm shm_perm;
1300         int shm_segsz;
1301         compat_time_t shm_atime;
1302         compat_time_t shm_dtime;
1303         compat_time_t shm_ctime;
1304         compat_ipc_pid_t shm_cpid;
1305         compat_ipc_pid_t shm_lpid;
1306         unsigned short shm_nattch;
1307         unsigned short shm_unused;
1308         compat_uptr_t shm_unused2;
1309         compat_uptr_t shm_unused3;
1310 };
1311
1312 struct compat_shminfo64 {
1313         compat_ulong_t shmmax;
1314         compat_ulong_t shmmin;
1315         compat_ulong_t shmmni;
1316         compat_ulong_t shmseg;
1317         compat_ulong_t shmall;
1318         compat_ulong_t __unused1;
1319         compat_ulong_t __unused2;
1320         compat_ulong_t __unused3;
1321         compat_ulong_t __unused4;
1322 };
1323
1324 struct compat_shm_info {
1325         compat_int_t used_ids;
1326         compat_ulong_t shm_tot, shm_rss, shm_swp;
1327         compat_ulong_t swap_attempts, swap_successes;
1328 };
1329
1330 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1331                                         int version)
1332 {
1333         if (in->shmmax > INT_MAX)
1334                 in->shmmax = INT_MAX;
1335         if (version == IPC_64) {
1336                 struct compat_shminfo64 info;
1337                 memset(&info, 0, sizeof(info));
1338                 info.shmmax = in->shmmax;
1339                 info.shmmin = in->shmmin;
1340                 info.shmmni = in->shmmni;
1341                 info.shmseg = in->shmseg;
1342                 info.shmall = in->shmall;
1343                 return copy_to_user(buf, &info, sizeof(info));
1344         } else {
1345                 struct shminfo info;
1346                 memset(&info, 0, sizeof(info));
1347                 info.shmmax = in->shmmax;
1348                 info.shmmin = in->shmmin;
1349                 info.shmmni = in->shmmni;
1350                 info.shmseg = in->shmseg;
1351                 info.shmall = in->shmall;
1352                 return copy_to_user(buf, &info, sizeof(info));
1353         }
1354 }
1355
1356 static int put_compat_shm_info(struct shm_info *ip,
1357                                 struct compat_shm_info __user *uip)
1358 {
1359         struct compat_shm_info info;
1360
1361         memset(&info, 0, sizeof(info));
1362         info.used_ids = ip->used_ids;
1363         info.shm_tot = ip->shm_tot;
1364         info.shm_rss = ip->shm_rss;
1365         info.shm_swp = ip->shm_swp;
1366         info.swap_attempts = ip->swap_attempts;
1367         info.swap_successes = ip->swap_successes;
1368         return copy_to_user(uip, &info, sizeof(info));
1369 }
1370
1371 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1372                                         int version)
1373 {
1374         if (version == IPC_64) {
1375                 struct compat_shmid64_ds v;
1376                 memset(&v, 0, sizeof(v));
1377                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1378                 v.shm_atime      = lower_32_bits(in->shm_atime);
1379                 v.shm_atime_high = upper_32_bits(in->shm_atime);
1380                 v.shm_dtime      = lower_32_bits(in->shm_dtime);
1381                 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1382                 v.shm_ctime      = lower_32_bits(in->shm_ctime);
1383                 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1384                 v.shm_segsz = in->shm_segsz;
1385                 v.shm_nattch = in->shm_nattch;
1386                 v.shm_cpid = in->shm_cpid;
1387                 v.shm_lpid = in->shm_lpid;
1388                 return copy_to_user(buf, &v, sizeof(v));
1389         } else {
1390                 struct compat_shmid_ds v;
1391                 memset(&v, 0, sizeof(v));
1392                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1393                 v.shm_perm.key = in->shm_perm.key;
1394                 v.shm_atime = in->shm_atime;
1395                 v.shm_dtime = in->shm_dtime;
1396                 v.shm_ctime = in->shm_ctime;
1397                 v.shm_segsz = in->shm_segsz;
1398                 v.shm_nattch = in->shm_nattch;
1399                 v.shm_cpid = in->shm_cpid;
1400                 v.shm_lpid = in->shm_lpid;
1401                 return copy_to_user(buf, &v, sizeof(v));
1402         }
1403 }
1404
1405 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1406                                         int version)
1407 {
1408         memset(out, 0, sizeof(*out));
1409         if (version == IPC_64) {
1410                 struct compat_shmid64_ds __user *p = buf;
1411                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1412         } else {
1413                 struct compat_shmid_ds __user *p = buf;
1414                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1415         }
1416 }
1417
1418 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1419 {
1420         struct ipc_namespace *ns;
1421         struct shmid64_ds sem64;
1422         int version = compat_ipc_parse_version(&cmd);
1423         int err;
1424
1425         ns = current->nsproxy->ipc_ns;
1426
1427         if (cmd < 0 || shmid < 0)
1428                 return -EINVAL;
1429
1430         switch (cmd) {
1431         case IPC_INFO: {
1432                 struct shminfo64 shminfo;
1433                 err = shmctl_ipc_info(ns, &shminfo);
1434                 if (err < 0)
1435                         return err;
1436                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1437                         err = -EFAULT;
1438                 return err;
1439         }
1440         case SHM_INFO: {
1441                 struct shm_info shm_info;
1442                 err = shmctl_shm_info(ns, &shm_info);
1443                 if (err < 0)
1444                         return err;
1445                 if (put_compat_shm_info(&shm_info, uptr))
1446                         err = -EFAULT;
1447                 return err;
1448         }
1449         case IPC_STAT:
1450         case SHM_STAT_ANY:
1451         case SHM_STAT:
1452                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1453                 if (err < 0)
1454                         return err;
1455                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1456                         err = -EFAULT;
1457                 return err;
1458
1459         case IPC_SET:
1460                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1461                         return -EFAULT;
1462                 /* fallthru */
1463         case IPC_RMID:
1464                 return shmctl_down(ns, shmid, cmd, &sem64);
1465         case SHM_LOCK:
1466         case SHM_UNLOCK:
1467                 return shmctl_do_lock(ns, shmid, cmd);
1468                 break;
1469         default:
1470                 return -EINVAL;
1471         }
1472         return err;
1473 }
1474
1475 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1476 {
1477         return compat_ksys_shmctl(shmid, cmd, uptr);
1478 }
1479 #endif
1480
1481 /*
1482  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1483  *
1484  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1485  * "raddr" thing points to kernel space, and there has to be a wrapper around
1486  * this.
1487  */
1488 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1489               ulong *raddr, unsigned long shmlba)
1490 {
1491         struct shmid_kernel *shp;
1492         unsigned long addr = (unsigned long)shmaddr;
1493         unsigned long size;
1494         struct file *file, *base;
1495         int    err;
1496         unsigned long flags = MAP_SHARED;
1497         unsigned long prot;
1498         int acc_mode;
1499         struct ipc_namespace *ns;
1500         struct shm_file_data *sfd;
1501         int f_flags;
1502         unsigned long populate = 0;
1503
1504         err = -EINVAL;
1505         if (shmid < 0)
1506                 goto out;
1507
1508         if (addr) {
1509                 if (addr & (shmlba - 1)) {
1510                         if (shmflg & SHM_RND) {
1511                                 addr &= ~(shmlba - 1);  /* round down */
1512
1513                                 /*
1514                                  * Ensure that the round-down is non-nil
1515                                  * when remapping. This can happen for
1516                                  * cases when addr < shmlba.
1517                                  */
1518                                 if (!addr && (shmflg & SHM_REMAP))
1519                                         goto out;
1520                         } else
1521 #ifndef __ARCH_FORCE_SHMLBA
1522                                 if (addr & ~PAGE_MASK)
1523 #endif
1524                                         goto out;
1525                 }
1526
1527                 flags |= MAP_FIXED;
1528         } else if ((shmflg & SHM_REMAP))
1529                 goto out;
1530
1531         if (shmflg & SHM_RDONLY) {
1532                 prot = PROT_READ;
1533                 acc_mode = S_IRUGO;
1534                 f_flags = O_RDONLY;
1535         } else {
1536                 prot = PROT_READ | PROT_WRITE;
1537                 acc_mode = S_IRUGO | S_IWUGO;
1538                 f_flags = O_RDWR;
1539         }
1540         if (shmflg & SHM_EXEC) {
1541                 prot |= PROT_EXEC;
1542                 acc_mode |= S_IXUGO;
1543         }
1544
1545         /*
1546          * We cannot rely on the fs check since SYSV IPC does have an
1547          * additional creator id...
1548          */
1549         ns = current->nsproxy->ipc_ns;
1550         rcu_read_lock();
1551         shp = shm_obtain_object_check(ns, shmid);
1552         if (IS_ERR(shp)) {
1553                 err = PTR_ERR(shp);
1554                 goto out_unlock;
1555         }
1556
1557         err = -EACCES;
1558         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1559                 goto out_unlock;
1560
1561         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1562         if (err)
1563                 goto out_unlock;
1564
1565         ipc_lock_object(&shp->shm_perm);
1566
1567         /* check if shm_destroy() is tearing down shp */
1568         if (!ipc_valid_object(&shp->shm_perm)) {
1569                 ipc_unlock_object(&shp->shm_perm);
1570                 err = -EIDRM;
1571                 goto out_unlock;
1572         }
1573
1574         /*
1575          * We need to take a reference to the real shm file to prevent the
1576          * pointer from becoming stale in cases where the lifetime of the outer
1577          * file extends beyond that of the shm segment.  It's not usually
1578          * possible, but it can happen during remap_file_pages() emulation as
1579          * that unmaps the memory, then does ->mmap() via file reference only.
1580          * We'll deny the ->mmap() if the shm segment was since removed, but to
1581          * detect shm ID reuse we need to compare the file pointers.
1582          */
1583         base = get_file(shp->shm_file);
1584         shp->shm_nattch++;
1585         size = i_size_read(file_inode(base));
1586         ipc_unlock_object(&shp->shm_perm);
1587         rcu_read_unlock();
1588
1589         err = -ENOMEM;
1590         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1591         if (!sfd) {
1592                 fput(base);
1593                 goto out_nattch;
1594         }
1595
1596         file = alloc_file_clone(base, f_flags,
1597                           is_file_hugepages(base) ?
1598                                 &shm_file_operations_huge :
1599                                 &shm_file_operations);
1600         err = PTR_ERR(file);
1601         if (IS_ERR(file)) {
1602                 kfree(sfd);
1603                 fput(base);
1604                 goto out_nattch;
1605         }
1606
1607         sfd->id = shp->shm_perm.id;
1608         sfd->ns = get_ipc_ns(ns);
1609         sfd->file = base;
1610         sfd->vm_ops = NULL;
1611         file->private_data = sfd;
1612
1613         err = security_mmap_file(file, prot, flags);
1614         if (err)
1615                 goto out_fput;
1616
1617         if (down_write_killable(&current->mm->mmap_sem)) {
1618                 err = -EINTR;
1619                 goto out_fput;
1620         }
1621
1622         if (addr && !(shmflg & SHM_REMAP)) {
1623                 err = -EINVAL;
1624                 if (addr + size < addr)
1625                         goto invalid;
1626
1627                 if (find_vma_intersection(current->mm, addr, addr + size))
1628                         goto invalid;
1629         }
1630
1631         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1632         *raddr = addr;
1633         err = 0;
1634         if (IS_ERR_VALUE(addr))
1635                 err = (long)addr;
1636 invalid:
1637         up_write(&current->mm->mmap_sem);
1638         if (populate)
1639                 mm_populate(addr, populate);
1640
1641 out_fput:
1642         fput(file);
1643
1644 out_nattch:
1645         down_write(&shm_ids(ns).rwsem);
1646         shp = shm_lock(ns, shmid);
1647         shp->shm_nattch--;
1648
1649         if (shm_may_destroy(shp))
1650                 shm_destroy(ns, shp);
1651         else
1652                 shm_unlock(shp);
1653         up_write(&shm_ids(ns).rwsem);
1654         return err;
1655
1656 out_unlock:
1657         rcu_read_unlock();
1658 out:
1659         return err;
1660 }
1661
1662 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1663 {
1664         unsigned long ret;
1665         long err;
1666
1667         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1668         if (err)
1669                 return err;
1670         force_successful_syscall_return();
1671         return (long)ret;
1672 }
1673
1674 #ifdef CONFIG_COMPAT
1675
1676 #ifndef COMPAT_SHMLBA
1677 #define COMPAT_SHMLBA   SHMLBA
1678 #endif
1679
1680 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1681 {
1682         unsigned long ret;
1683         long err;
1684
1685         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1686         if (err)
1687                 return err;
1688         force_successful_syscall_return();
1689         return (long)ret;
1690 }
1691 #endif
1692
1693 /*
1694  * detach and kill segment if marked destroyed.
1695  * The work is done in shm_close.
1696  */
1697 long ksys_shmdt(char __user *shmaddr)
1698 {
1699         struct mm_struct *mm = current->mm;
1700         struct vm_area_struct *vma;
1701         unsigned long addr = (unsigned long)shmaddr;
1702         int retval = -EINVAL;
1703 #ifdef CONFIG_MMU
1704         loff_t size = 0;
1705         struct file *file;
1706         struct vm_area_struct *next;
1707 #endif
1708
1709         if (addr & ~PAGE_MASK)
1710                 return retval;
1711
1712         if (down_write_killable(&mm->mmap_sem))
1713                 return -EINTR;
1714
1715         /*
1716          * This function tries to be smart and unmap shm segments that
1717          * were modified by partial mlock or munmap calls:
1718          * - It first determines the size of the shm segment that should be
1719          *   unmapped: It searches for a vma that is backed by shm and that
1720          *   started at address shmaddr. It records it's size and then unmaps
1721          *   it.
1722          * - Then it unmaps all shm vmas that started at shmaddr and that
1723          *   are within the initially determined size and that are from the
1724          *   same shm segment from which we determined the size.
1725          * Errors from do_munmap are ignored: the function only fails if
1726          * it's called with invalid parameters or if it's called to unmap
1727          * a part of a vma. Both calls in this function are for full vmas,
1728          * the parameters are directly copied from the vma itself and always
1729          * valid - therefore do_munmap cannot fail. (famous last words?)
1730          */
1731         /*
1732          * If it had been mremap()'d, the starting address would not
1733          * match the usual checks anyway. So assume all vma's are
1734          * above the starting address given.
1735          */
1736         vma = find_vma(mm, addr);
1737
1738 #ifdef CONFIG_MMU
1739         while (vma) {
1740                 next = vma->vm_next;
1741
1742                 /*
1743                  * Check if the starting address would match, i.e. it's
1744                  * a fragment created by mprotect() and/or munmap(), or it
1745                  * otherwise it starts at this address with no hassles.
1746                  */
1747                 if ((vma->vm_ops == &shm_vm_ops) &&
1748                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1749
1750                         /*
1751                          * Record the file of the shm segment being
1752                          * unmapped.  With mremap(), someone could place
1753                          * page from another segment but with equal offsets
1754                          * in the range we are unmapping.
1755                          */
1756                         file = vma->vm_file;
1757                         size = i_size_read(file_inode(vma->vm_file));
1758                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1759                         /*
1760                          * We discovered the size of the shm segment, so
1761                          * break out of here and fall through to the next
1762                          * loop that uses the size information to stop
1763                          * searching for matching vma's.
1764                          */
1765                         retval = 0;
1766                         vma = next;
1767                         break;
1768                 }
1769                 vma = next;
1770         }
1771
1772         /*
1773          * We need look no further than the maximum address a fragment
1774          * could possibly have landed at. Also cast things to loff_t to
1775          * prevent overflows and make comparisons vs. equal-width types.
1776          */
1777         size = PAGE_ALIGN(size);
1778         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1779                 next = vma->vm_next;
1780
1781                 /* finding a matching vma now does not alter retval */
1782                 if ((vma->vm_ops == &shm_vm_ops) &&
1783                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1784                     (vma->vm_file == file))
1785                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1786                 vma = next;
1787         }
1788
1789 #else   /* CONFIG_MMU */
1790         /* under NOMMU conditions, the exact address to be destroyed must be
1791          * given
1792          */
1793         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1794                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1795                 retval = 0;
1796         }
1797
1798 #endif
1799
1800         up_write(&mm->mmap_sem);
1801         return retval;
1802 }
1803
1804 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1805 {
1806         return ksys_shmdt(shmaddr);
1807 }
1808
1809 #ifdef CONFIG_PROC_FS
1810 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1811 {
1812         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1813         struct user_namespace *user_ns = seq_user_ns(s);
1814         struct kern_ipc_perm *ipcp = it;
1815         struct shmid_kernel *shp;
1816         unsigned long rss = 0, swp = 0;
1817
1818         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1819         shm_add_rss_swap(shp, &rss, &swp);
1820
1821 #if BITS_PER_LONG <= 32
1822 #define SIZE_SPEC "%10lu"
1823 #else
1824 #define SIZE_SPEC "%21lu"
1825 #endif
1826
1827         seq_printf(s,
1828                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1829                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1830                    SIZE_SPEC " " SIZE_SPEC "\n",
1831                    shp->shm_perm.key,
1832                    shp->shm_perm.id,
1833                    shp->shm_perm.mode,
1834                    shp->shm_segsz,
1835                    pid_nr_ns(shp->shm_cprid, pid_ns),
1836                    pid_nr_ns(shp->shm_lprid, pid_ns),
1837                    shp->shm_nattch,
1838                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1839                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1840                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1841                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1842                    shp->shm_atim,
1843                    shp->shm_dtim,
1844                    shp->shm_ctim,
1845                    rss * PAGE_SIZE,
1846                    swp * PAGE_SIZE);
1847
1848         return 0;
1849 }
1850 #endif