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 ucounts          *mlock_ucounts;
  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         kfree(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_ucounts);
 333         fput(shm_file);
 334         ipc_update_pid(&shp->shm_cprid, NULL);
 335         ipc_update_pid(&shp->shm_lprid, NULL);
 336         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 337 }
 338
 339 /*
 340  * shm_may_destroy - identifies whether shm segment should be destroyed now
 341  *
 342  * Returns true if and only if there are no active users of the segment and
 343  * one of the following is true:
 344  *
 345  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 346  *
 347  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 348  */
 349 static bool shm_may_destroy(struct shmid_kernel *shp)
 350 {
 351         return (shp->shm_nattch == 0) &&
 352                (shp->ns->shm_rmid_forced ||
 353                 (shp->shm_perm.mode & SHM_DEST));
 354 }
 355
 356 /*
 357  * remove the attach descriptor vma.
 358  * free memory for segment if it is marked destroyed.
 359  * The descriptor has already been removed from the current->mm->mmap list
 360  * and will later be kfree()d.
 361  */
 362 static void shm_close(struct vm_area_struct *vma)
 363 {
 364         struct file *file = vma->vm_file;
 365         struct shm_file_data *sfd = shm_file_data(file);
 366         struct shmid_kernel *shp;
 367         struct ipc_namespace *ns = sfd->ns;
 368
 369         down_write(&shm_ids(ns).rwsem);
 370         /* remove from the list of attaches of the shm segment */
 371         shp = shm_lock(ns, sfd->id);
 372
 373         /*
 374          * We raced in the idr lookup or with shm_destroy().
 375          * Either way, the ID is busted.
 376          */
 377         if (WARN_ON_ONCE(IS_ERR(shp)))
 378                 goto done; /* no-op */
 379
 380         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 381         shp->shm_dtim = ktime_get_real_seconds();
 382         shp->shm_nattch--;
 383         if (shm_may_destroy(shp))
 384                 shm_destroy(ns, shp);
 385         else
 386                 shm_unlock(shp);
 387 done:
 388         up_write(&shm_ids(ns).rwsem);
 389 }
 390
 391 /* Called with ns->shm_ids(ns).rwsem locked */
 392 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 393 {
 394         struct ipc_namespace *ns = data;
 395         struct kern_ipc_perm *ipcp = p;
 396         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 397
 398         /*
 399          * We want to destroy segments without users and with already
 400          * exit'ed originating process.
 401          *
 402          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 403          */
 404         if (!list_empty(&shp->shm_clist))
 405                 return 0;
 406
 407         if (shm_may_destroy(shp)) {
 408                 shm_lock_by_ptr(shp);
 409                 shm_destroy(ns, shp);
 410         }
 411         return 0;
 412 }
 413
 414 void shm_destroy_orphaned(struct ipc_namespace *ns)
 415 {
 416         down_write(&shm_ids(ns).rwsem);
 417         if (shm_ids(ns).in_use)
 418                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 419         up_write(&shm_ids(ns).rwsem);
 420 }
 421
 422 /* Locking assumes this will only be called with task == current */
 423 void exit_shm(struct task_struct *task)
 424 {
 425         for (;;) {
 426                 struct shmid_kernel *shp;
 427                 struct ipc_namespace *ns;
 428
 429                 task_lock(task);
 430
 431                 if (list_empty(&task->sysvshm.shm_clist)) {
 432                         task_unlock(task);
 433                         break;
 434                 }
 435
 436                 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
 437                                 shm_clist);
 438
 439                 /*
 440                  * 1) Get pointer to the ipc namespace. It is worth to say
 441                  * that this pointer is guaranteed to be valid because
 442                  * shp lifetime is always shorter than namespace lifetime
 443                  * in which shp lives.
 444                  * We taken task_lock it means that shp won't be freed.
 445                  */
 446                 ns = shp->ns;
 447
 448                 /*
 449                  * 2) If kernel.shm_rmid_forced is not set then only keep track of
 450                  * which shmids are orphaned, so that a later set of the sysctl
 451                  * can clean them up.
 452                  */
 453                 if (!ns->shm_rmid_forced)
 454                         goto unlink_continue;
 455
 456                 /*
 457                  * 3) get a reference to the namespace.
 458                  *    The refcount could be already 0. If it is 0, then
 459                  *    the shm objects will be free by free_ipc_work().
 460                  */
 461                 ns = get_ipc_ns_not_zero(ns);
 462                 if (!ns) {
 463 unlink_continue:
 464                         list_del_init(&shp->shm_clist);
 465                         task_unlock(task);
 466                         continue;
 467                 }
 468
 469                 /*
 470                  * 4) get a reference to shp.
 471                  *   This cannot fail: shm_clist_rm() is called before
 472                  *   ipc_rmid(), thus the refcount cannot be 0.
 473                  */
 474                 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
 475
 476                 /*
 477                  * 5) unlink the shm segment from the list of segments
 478                  *    created by current.
 479                  *    This must be done last. After unlinking,
 480                  *    only the refcounts obtained above prevent IPC_RMID
 481                  *    from destroying the segment or the namespace.
 482                  */
 483                 list_del_init(&shp->shm_clist);
 484
 485                 task_unlock(task);
 486
 487                 /*
 488                  * 6) we have all references
 489                  *    Thus lock & if needed destroy shp.
 490                  */
 491                 down_write(&shm_ids(ns).rwsem);
 492                 shm_lock_by_ptr(shp);
 493                 /*
 494                  * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
 495                  * safe to call ipc_rcu_putref here
 496                  */
 497                 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 498
 499                 if (ipc_valid_object(&shp->shm_perm)) {
 500                         if (shm_may_destroy(shp))
 501                                 shm_destroy(ns, shp);
 502                         else
 503                                 shm_unlock(shp);
 504                 } else {
 505                         /*
 506                          * Someone else deleted the shp from namespace
 507                          * idr/kht while we have waited.
 508                          * Just unlock and continue.
 509                          */
 510                         shm_unlock(shp);
 511                 }
 512
 513                 up_write(&shm_ids(ns).rwsem);
 514                 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
 515         }
 516 }
 517
 518 static vm_fault_t shm_fault(struct vm_fault *vmf)
 519 {
 520         struct file *file = vmf->vma->vm_file;
 521         struct shm_file_data *sfd = shm_file_data(file);
 522
 523         return sfd->vm_ops->fault(vmf);
 524 }
 525
 526 static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
 527 {
 528         struct file *file = vma->vm_file;
 529         struct shm_file_data *sfd = shm_file_data(file);
 530
 531         if (sfd->vm_ops->may_split)
 532                 return sfd->vm_ops->may_split(vma, addr);
 533
 534         return 0;
 535 }
 536
 537 static unsigned long shm_pagesize(struct vm_area_struct *vma)
 538 {
 539         struct file *file = vma->vm_file;
 540         struct shm_file_data *sfd = shm_file_data(file);
 541
 542         if (sfd->vm_ops->pagesize)
 543                 return sfd->vm_ops->pagesize(vma);
 544
 545         return PAGE_SIZE;
 546 }
 547
 548 #ifdef CONFIG_NUMA
 549 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 550 {
 551         struct file *file = vma->vm_file;
 552         struct shm_file_data *sfd = shm_file_data(file);
 553         int err = 0;
 554
 555         if (sfd->vm_ops->set_policy)
 556                 err = sfd->vm_ops->set_policy(vma, new);
 557         return err;
 558 }
 559
 560 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 561                                         unsigned long addr)
 562 {
 563         struct file *file = vma->vm_file;
 564         struct shm_file_data *sfd = shm_file_data(file);
 565         struct mempolicy *pol = NULL;
 566
 567         if (sfd->vm_ops->get_policy)
 568                 pol = sfd->vm_ops->get_policy(vma, addr);
 569         else if (vma->vm_policy)
 570                 pol = vma->vm_policy;
 571
 572         return pol;
 573 }
 574 #endif
 575
 576 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 577 {
 578         struct shm_file_data *sfd = shm_file_data(file);
 579         int ret;
 580
 581         /*
 582          * In case of remap_file_pages() emulation, the file can represent an
 583          * IPC ID that was removed, and possibly even reused by another shm
 584          * segment already.  Propagate this case as an error to caller.
 585          */
 586         ret = __shm_open(vma);
 587         if (ret)
 588                 return ret;
 589
 590         ret = call_mmap(sfd->file, vma);
 591         if (ret) {
 592                 shm_close(vma);
 593                 return ret;
 594         }
 595         sfd->vm_ops = vma->vm_ops;
 596 #ifdef CONFIG_MMU
 597         WARN_ON(!sfd->vm_ops->fault);
 598 #endif
 599         vma->vm_ops = &shm_vm_ops;
 600         return 0;
 601 }
 602
 603 static int shm_release(struct inode *ino, struct file *file)
 604 {
 605         struct shm_file_data *sfd = shm_file_data(file);
 606
 607         put_ipc_ns(sfd->ns);
 608         fput(sfd->file);
 609         shm_file_data(file) = NULL;
 610         kfree(sfd);
 611         return 0;
 612 }
 613
 614 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 615 {
 616         struct shm_file_data *sfd = shm_file_data(file);
 617
 618         if (!sfd->file->f_op->fsync)
 619                 return -EINVAL;
 620         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 621 }
 622
 623 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 624                           loff_t len)
 625 {
 626         struct shm_file_data *sfd = shm_file_data(file);
 627
 628         if (!sfd->file->f_op->fallocate)
 629                 return -EOPNOTSUPP;
 630         return sfd->file->f_op->fallocate(file, mode, offset, len);
 631 }
 632
 633 static unsigned long shm_get_unmapped_area(struct file *file,
 634         unsigned long addr, unsigned long len, unsigned long pgoff,
 635         unsigned long flags)
 636 {
 637         struct shm_file_data *sfd = shm_file_data(file);
 638
 639         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 640                                                 pgoff, flags);
 641 }
 642
 643 static const struct file_operations shm_file_operations = {
 644         .mmap           = shm_mmap,
 645         .fsync          = shm_fsync,
 646         .release        = shm_release,
 647         .get_unmapped_area      = shm_get_unmapped_area,
 648         .llseek         = noop_llseek,
 649         .fallocate      = shm_fallocate,
 650 };
 651
 652 /*
 653  * shm_file_operations_huge is now identical to shm_file_operations,
 654  * but we keep it distinct for the sake of is_file_shm_hugepages().
 655  */
 656 static const struct file_operations shm_file_operations_huge = {
 657         .mmap           = shm_mmap,
 658         .fsync          = shm_fsync,
 659         .release        = shm_release,
 660         .get_unmapped_area      = shm_get_unmapped_area,
 661         .llseek         = noop_llseek,
 662         .fallocate      = shm_fallocate,
 663 };
 664
 665 bool is_file_shm_hugepages(struct file *file)
 666 {
 667         return file->f_op == &shm_file_operations_huge;
 668 }
 669
 670 static const struct vm_operations_struct shm_vm_ops = {
 671         .open   = shm_open,     /* callback for a new vm-area open */
 672         .close  = shm_close,    /* callback for when the vm-area is released */
 673         .fault  = shm_fault,
 674         .may_split = shm_may_split,
 675         .pagesize = shm_pagesize,
 676 #if defined(CONFIG_NUMA)
 677         .set_policy = shm_set_policy,
 678         .get_policy = shm_get_policy,
 679 #endif
 680 };
 681
 682 /**
 683  * newseg - Create a new shared memory segment
 684  * @ns: namespace
 685  * @params: ptr to the structure that contains key, size and shmflg
 686  *
 687  * Called with shm_ids.rwsem held as a writer.
 688  */
 689 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 690 {
 691         key_t key = params->key;
 692         int shmflg = params->flg;
 693         size_t size = params->u.size;
 694         int error;
 695         struct shmid_kernel *shp;
 696         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 697         struct file *file;
 698         char name[13];
 699         vm_flags_t acctflag = 0;
 700
 701         if (size < SHMMIN || size > ns->shm_ctlmax)
 702                 return -EINVAL;
 703
 704         if (numpages << PAGE_SHIFT < size)
 705                 return -ENOSPC;
 706
 707         if (ns->shm_tot + numpages < ns->shm_tot ||
 708                         ns->shm_tot + numpages > ns->shm_ctlall)
 709                 return -ENOSPC;
 710
 711         shp = kmalloc(sizeof(*shp), GFP_KERNEL_ACCOUNT);
 712         if (unlikely(!shp))
 713                 return -ENOMEM;
 714
 715         shp->shm_perm.key = key;
 716         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 717         shp->mlock_ucounts = NULL;
 718
 719         shp->shm_perm.security = NULL;
 720         error = security_shm_alloc(&shp->shm_perm);
 721         if (error) {
 722                 kfree(shp);
 723                 return error;
 724         }
 725
 726         sprintf(name, "SYSV%08x", key);
 727         if (shmflg & SHM_HUGETLB) {
 728                 struct hstate *hs;
 729                 size_t hugesize;
 730
 731                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 732                 if (!hs) {
 733                         error = -EINVAL;
 734                         goto no_file;
 735                 }
 736                 hugesize = ALIGN(size, huge_page_size(hs));
 737
 738                 /* hugetlb_file_setup applies strict accounting */
 739                 if (shmflg & SHM_NORESERVE)
 740                         acctflag = VM_NORESERVE;
 741                 file = hugetlb_file_setup(name, hugesize, acctflag,
 742                                 HUGETLB_SHMFS_INODE, (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 743         } else {
 744                 /*
 745                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 746                  * if it's asked for.
 747                  */
 748                 if  ((shmflg & SHM_NORESERVE) &&
 749                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 750                         acctflag = VM_NORESERVE;
 751                 file = shmem_kernel_file_setup(name, size, acctflag);
 752         }
 753         error = PTR_ERR(file);
 754         if (IS_ERR(file))
 755                 goto no_file;
 756
 757         shp->shm_cprid = get_pid(task_tgid(current));
 758         shp->shm_lprid = NULL;
 759         shp->shm_atim = shp->shm_dtim = 0;
 760         shp->shm_ctim = ktime_get_real_seconds();
 761         shp->shm_segsz = size;
 762         shp->shm_nattch = 0;
 763         shp->shm_file = file;
 764         shp->shm_creator = current;
 765
 766         /* ipc_addid() locks shp upon success. */
 767         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 768         if (error < 0)
 769                 goto no_id;
 770
 771         shp->ns = ns;
 772
 773         task_lock(current);
 774         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 775         task_unlock(current);
 776
 777         /*
 778          * shmid gets reported as "inode#" in /proc/pid/maps.
 779          * proc-ps tools use this. Changing this will break them.
 780          */
 781         file_inode(file)->i_ino = shp->shm_perm.id;
 782
 783         ns->shm_tot += numpages;
 784         error = shp->shm_perm.id;
 785
 786         ipc_unlock_object(&shp->shm_perm);
 787         rcu_read_unlock();
 788         return error;
 789
 790 no_id:
 791         ipc_update_pid(&shp->shm_cprid, NULL);
 792         ipc_update_pid(&shp->shm_lprid, NULL);
 793         fput(file);
 794         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 795         return error;
 796 no_file:
 797         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 798         return error;
 799 }
 800
 801 /*
 802  * Called with shm_ids.rwsem and ipcp locked.
 803  */
 804 static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
 805 {
 806         struct shmid_kernel *shp;
 807
 808         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 809         if (shp->shm_segsz < params->u.size)
 810                 return -EINVAL;
 811
 812         return 0;
 813 }
 814
 815 long ksys_shmget(key_t key, size_t size, int shmflg)
 816 {
 817         struct ipc_namespace *ns;
 818         static const struct ipc_ops shm_ops = {
 819                 .getnew = newseg,
 820                 .associate = security_shm_associate,
 821                 .more_checks = shm_more_checks,
 822         };
 823         struct ipc_params shm_params;
 824
 825         ns = current->nsproxy->ipc_ns;
 826
 827         shm_params.key = key;
 828         shm_params.flg = shmflg;
 829         shm_params.u.size = size;
 830
 831         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 832 }
 833
 834 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 835 {
 836         return ksys_shmget(key, size, shmflg);
 837 }
 838
 839 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 840 {
 841         switch (version) {
 842         case IPC_64:
 843                 return copy_to_user(buf, in, sizeof(*in));
 844         case IPC_OLD:
 845             {
 846                 struct shmid_ds out;
 847
 848                 memset(&out, 0, sizeof(out));
 849                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 850                 out.shm_segsz   = in->shm_segsz;
 851                 out.shm_atime   = in->shm_atime;
 852                 out.shm_dtime   = in->shm_dtime;
 853                 out.shm_ctime   = in->shm_ctime;
 854                 out.shm_cpid    = in->shm_cpid;
 855                 out.shm_lpid    = in->shm_lpid;
 856                 out.shm_nattch  = in->shm_nattch;
 857
 858                 return copy_to_user(buf, &out, sizeof(out));
 859             }
 860         default:
 861                 return -EINVAL;
 862         }
 863 }
 864
 865 static inline unsigned long
 866 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 867 {
 868         switch (version) {
 869         case IPC_64:
 870                 if (copy_from_user(out, buf, sizeof(*out)))
 871                         return -EFAULT;
 872                 return 0;
 873         case IPC_OLD:
 874             {
 875                 struct shmid_ds tbuf_old;
 876
 877                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 878                         return -EFAULT;
 879
 880                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 881                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 882                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 883
 884                 return 0;
 885             }
 886         default:
 887                 return -EINVAL;
 888         }
 889 }
 890
 891 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 892 {
 893         switch (version) {
 894         case IPC_64:
 895                 return copy_to_user(buf, in, sizeof(*in));
 896         case IPC_OLD:
 897             {
 898                 struct shminfo out;
 899
 900                 if (in->shmmax > INT_MAX)
 901                         out.shmmax = INT_MAX;
 902                 else
 903                         out.shmmax = (int)in->shmmax;
 904
 905                 out.shmmin      = in->shmmin;
 906                 out.shmmni      = in->shmmni;
 907                 out.shmseg      = in->shmseg;
 908                 out.shmall      = in->shmall;
 909
 910                 return copy_to_user(buf, &out, sizeof(out));
 911             }
 912         default:
 913                 return -EINVAL;
 914         }
 915 }
 916
 917 /*
 918  * Calculate and add used RSS and swap pages of a shm.
 919  * Called with shm_ids.rwsem held as a reader
 920  */
 921 static void shm_add_rss_swap(struct shmid_kernel *shp,
 922         unsigned long *rss_add, unsigned long *swp_add)
 923 {
 924         struct inode *inode;
 925
 926         inode = file_inode(shp->shm_file);
 927
 928         if (is_file_hugepages(shp->shm_file)) {
 929                 struct address_space *mapping = inode->i_mapping;
 930                 struct hstate *h = hstate_file(shp->shm_file);
 931                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 932         } else {
 933 #ifdef CONFIG_SHMEM
 934                 struct shmem_inode_info *info = SHMEM_I(inode);
 935
 936                 spin_lock_irq(&info->lock);
 937                 *rss_add += inode->i_mapping->nrpages;
 938                 *swp_add += info->swapped;
 939                 spin_unlock_irq(&info->lock);
 940 #else
 941                 *rss_add += inode->i_mapping->nrpages;
 942 #endif
 943         }
 944 }
 945
 946 /*
 947  * Called with shm_ids.rwsem held as a reader
 948  */
 949 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 950                 unsigned long *swp)
 951 {
 952         int next_id;
 953         int total, in_use;
 954
 955         *rss = 0;
 956         *swp = 0;
 957
 958         in_use = shm_ids(ns).in_use;
 959
 960         for (total = 0, next_id = 0; total < in_use; next_id++) {
 961                 struct kern_ipc_perm *ipc;
 962                 struct shmid_kernel *shp;
 963
 964                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 965                 if (ipc == NULL)
 966                         continue;
 967                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 968
 969                 shm_add_rss_swap(shp, rss, swp);
 970
 971                 total++;
 972         }
 973 }
 974
 975 /*
 976  * This function handles some shmctl commands which require the rwsem
 977  * to be held in write mode.
 978  * NOTE: no locks must be held, the rwsem is taken inside this function.
 979  */
 980 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 981                        struct shmid64_ds *shmid64)
 982 {
 983         struct kern_ipc_perm *ipcp;
 984         struct shmid_kernel *shp;
 985         int err;
 986
 987         down_write(&shm_ids(ns).rwsem);
 988         rcu_read_lock();
 989
 990         ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
 991                                       &shmid64->shm_perm, 0);
 992         if (IS_ERR(ipcp)) {
 993                 err = PTR_ERR(ipcp);
 994                 goto out_unlock1;
 995         }
 996
 997         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 998
 999         err = security_shm_shmctl(&shp->shm_perm, cmd);
1000         if (err)
1001                 goto out_unlock1;
1002
1003         switch (cmd) {
1004         case IPC_RMID:
1005                 ipc_lock_object(&shp->shm_perm);
1006                 /* do_shm_rmid unlocks the ipc object and rcu */
1007                 do_shm_rmid(ns, ipcp);
1008                 goto out_up;
1009         case IPC_SET:
1010                 ipc_lock_object(&shp->shm_perm);
1011                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1012                 if (err)
1013                         goto out_unlock0;
1014                 shp->shm_ctim = ktime_get_real_seconds();
1015                 break;
1016         default:
1017                 err = -EINVAL;
1018                 goto out_unlock1;
1019         }
1020
1021 out_unlock0:
1022         ipc_unlock_object(&shp->shm_perm);
1023 out_unlock1:
1024         rcu_read_unlock();
1025 out_up:
1026         up_write(&shm_ids(ns).rwsem);
1027         return err;
1028 }
1029
1030 static int shmctl_ipc_info(struct ipc_namespace *ns,
1031                            struct shminfo64 *shminfo)
1032 {
1033         int err = security_shm_shmctl(NULL, IPC_INFO);
1034         if (!err) {
1035                 memset(shminfo, 0, sizeof(*shminfo));
1036                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1037                 shminfo->shmmax = ns->shm_ctlmax;
1038                 shminfo->shmall = ns->shm_ctlall;
1039                 shminfo->shmmin = SHMMIN;
1040                 down_read(&shm_ids(ns).rwsem);
1041                 err = ipc_get_maxidx(&shm_ids(ns));
1042                 up_read(&shm_ids(ns).rwsem);
1043                 if (err < 0)
1044                         err = 0;
1045         }
1046         return err;
1047 }
1048
1049 static int shmctl_shm_info(struct ipc_namespace *ns,
1050                            struct shm_info *shm_info)
1051 {
1052         int err = security_shm_shmctl(NULL, SHM_INFO);
1053         if (!err) {
1054                 memset(shm_info, 0, sizeof(*shm_info));
1055                 down_read(&shm_ids(ns).rwsem);
1056                 shm_info->used_ids = shm_ids(ns).in_use;
1057                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1058                 shm_info->shm_tot = ns->shm_tot;
1059                 shm_info->swap_attempts = 0;
1060                 shm_info->swap_successes = 0;
1061                 err = ipc_get_maxidx(&shm_ids(ns));
1062                 up_read(&shm_ids(ns).rwsem);
1063                 if (err < 0)
1064                         err = 0;
1065         }
1066         return err;
1067 }
1068
1069 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1070                         int cmd, struct shmid64_ds *tbuf)
1071 {
1072         struct shmid_kernel *shp;
1073         int err;
1074
1075         memset(tbuf, 0, sizeof(*tbuf));
1076
1077         rcu_read_lock();
1078         if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1079                 shp = shm_obtain_object(ns, shmid);
1080                 if (IS_ERR(shp)) {
1081                         err = PTR_ERR(shp);
1082                         goto out_unlock;
1083                 }
1084         } else { /* IPC_STAT */
1085                 shp = shm_obtain_object_check(ns, shmid);
1086                 if (IS_ERR(shp)) {
1087                         err = PTR_ERR(shp);
1088                         goto out_unlock;
1089                 }
1090         }
1091
1092         /*
1093          * Semantically SHM_STAT_ANY ought to be identical to
1094          * that functionality provided by the /proc/sysvipc/
1095          * interface. As such, only audit these calls and
1096          * do not do traditional S_IRUGO permission checks on
1097          * the ipc object.
1098          */
1099         if (cmd == SHM_STAT_ANY)
1100                 audit_ipc_obj(&shp->shm_perm);
1101         else {
1102                 err = -EACCES;
1103                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1104                         goto out_unlock;
1105         }
1106
1107         err = security_shm_shmctl(&shp->shm_perm, cmd);
1108         if (err)
1109                 goto out_unlock;
1110
1111         ipc_lock_object(&shp->shm_perm);
1112
1113         if (!ipc_valid_object(&shp->shm_perm)) {
1114                 ipc_unlock_object(&shp->shm_perm);
1115                 err = -EIDRM;
1116                 goto out_unlock;
1117         }
1118
1119         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1120         tbuf->shm_segsz = shp->shm_segsz;
1121         tbuf->shm_atime = shp->shm_atim;
1122         tbuf->shm_dtime = shp->shm_dtim;
1123         tbuf->shm_ctime = shp->shm_ctim;
1124 #ifndef CONFIG_64BIT
1125         tbuf->shm_atime_high = shp->shm_atim >> 32;
1126         tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1127         tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1128 #endif
1129         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
1130         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
1131         tbuf->shm_nattch = shp->shm_nattch;
1132
1133         if (cmd == IPC_STAT) {
1134                 /*
1135                  * As defined in SUS:
1136                  * Return 0 on success
1137                  */
1138                 err = 0;
1139         } else {
1140                 /*
1141                  * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1142                  * Return the full id, including the sequence number
1143                  */
1144                 err = shp->shm_perm.id;
1145         }
1146
1147         ipc_unlock_object(&shp->shm_perm);
1148 out_unlock:
1149         rcu_read_unlock();
1150         return err;
1151 }
1152
1153 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1154 {
1155         struct shmid_kernel *shp;
1156         struct file *shm_file;
1157         int err;
1158
1159         rcu_read_lock();
1160         shp = shm_obtain_object_check(ns, shmid);
1161         if (IS_ERR(shp)) {
1162                 err = PTR_ERR(shp);
1163                 goto out_unlock1;
1164         }
1165
1166         audit_ipc_obj(&(shp->shm_perm));
1167         err = security_shm_shmctl(&shp->shm_perm, cmd);
1168         if (err)
1169                 goto out_unlock1;
1170
1171         ipc_lock_object(&shp->shm_perm);
1172
1173         /* check if shm_destroy() is tearing down shp */
1174         if (!ipc_valid_object(&shp->shm_perm)) {
1175                 err = -EIDRM;
1176                 goto out_unlock0;
1177         }
1178
1179         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1180                 kuid_t euid = current_euid();
1181
1182                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1183                     !uid_eq(euid, shp->shm_perm.cuid)) {
1184                         err = -EPERM;
1185                         goto out_unlock0;
1186                 }
1187                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1188                         err = -EPERM;
1189                         goto out_unlock0;
1190                 }
1191         }
1192
1193         shm_file = shp->shm_file;
1194         if (is_file_hugepages(shm_file))
1195                 goto out_unlock0;
1196
1197         if (cmd == SHM_LOCK) {
1198                 struct ucounts *ucounts = current_ucounts();
1199
1200                 err = shmem_lock(shm_file, 1, ucounts);
1201                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1202                         shp->shm_perm.mode |= SHM_LOCKED;
1203                         shp->mlock_ucounts = ucounts;
1204                 }
1205                 goto out_unlock0;
1206         }
1207
1208         /* SHM_UNLOCK */
1209         if (!(shp->shm_perm.mode & SHM_LOCKED))
1210                 goto out_unlock0;
1211         shmem_lock(shm_file, 0, shp->mlock_ucounts);
1212         shp->shm_perm.mode &= ~SHM_LOCKED;
1213         shp->mlock_ucounts = NULL;
1214         get_file(shm_file);
1215         ipc_unlock_object(&shp->shm_perm);
1216         rcu_read_unlock();
1217         shmem_unlock_mapping(shm_file->f_mapping);
1218
1219         fput(shm_file);
1220         return err;
1221
1222 out_unlock0:
1223         ipc_unlock_object(&shp->shm_perm);
1224 out_unlock1:
1225         rcu_read_unlock();
1226         return err;
1227 }
1228
1229 static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1230 {
1231         int err;
1232         struct ipc_namespace *ns;
1233         struct shmid64_ds sem64;
1234
1235         if (cmd < 0 || shmid < 0)
1236                 return -EINVAL;
1237
1238         ns = current->nsproxy->ipc_ns;
1239
1240         switch (cmd) {
1241         case IPC_INFO: {
1242                 struct shminfo64 shminfo;
1243                 err = shmctl_ipc_info(ns, &shminfo);
1244                 if (err < 0)
1245                         return err;
1246                 if (copy_shminfo_to_user(buf, &shminfo, version))
1247                         err = -EFAULT;
1248                 return err;
1249         }
1250         case SHM_INFO: {
1251                 struct shm_info shm_info;
1252                 err = shmctl_shm_info(ns, &shm_info);
1253                 if (err < 0)
1254                         return err;
1255                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1256                         err = -EFAULT;
1257                 return err;
1258         }
1259         case SHM_STAT:
1260         case SHM_STAT_ANY:
1261         case IPC_STAT: {
1262                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1263                 if (err < 0)
1264                         return err;
1265                 if (copy_shmid_to_user(buf, &sem64, version))
1266                         err = -EFAULT;
1267                 return err;
1268         }
1269         case IPC_SET:
1270                 if (copy_shmid_from_user(&sem64, buf, version))
1271                         return -EFAULT;
1272                 fallthrough;
1273         case IPC_RMID:
1274                 return shmctl_down(ns, shmid, cmd, &sem64);
1275         case SHM_LOCK:
1276         case SHM_UNLOCK:
1277                 return shmctl_do_lock(ns, shmid, cmd);
1278         default:
1279                 return -EINVAL;
1280         }
1281 }
1282
1283 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1284 {
1285         return ksys_shmctl(shmid, cmd, buf, IPC_64);
1286 }
1287
1288 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1289 long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1290 {
1291         int version = ipc_parse_version(&cmd);
1292
1293         return ksys_shmctl(shmid, cmd, buf, version);
1294 }
1295
1296 SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1297 {
1298         return ksys_old_shmctl(shmid, cmd, buf);
1299 }
1300 #endif
1301
1302 #ifdef CONFIG_COMPAT
1303
1304 struct compat_shmid_ds {
1305         struct compat_ipc_perm shm_perm;
1306         int shm_segsz;
1307         old_time32_t shm_atime;
1308         old_time32_t shm_dtime;
1309         old_time32_t shm_ctime;
1310         compat_ipc_pid_t shm_cpid;
1311         compat_ipc_pid_t shm_lpid;
1312         unsigned short shm_nattch;
1313         unsigned short shm_unused;
1314         compat_uptr_t shm_unused2;
1315         compat_uptr_t shm_unused3;
1316 };
1317
1318 struct compat_shminfo64 {
1319         compat_ulong_t shmmax;
1320         compat_ulong_t shmmin;
1321         compat_ulong_t shmmni;
1322         compat_ulong_t shmseg;
1323         compat_ulong_t shmall;
1324         compat_ulong_t __unused1;
1325         compat_ulong_t __unused2;
1326         compat_ulong_t __unused3;
1327         compat_ulong_t __unused4;
1328 };
1329
1330 struct compat_shm_info {
1331         compat_int_t used_ids;
1332         compat_ulong_t shm_tot, shm_rss, shm_swp;
1333         compat_ulong_t swap_attempts, swap_successes;
1334 };
1335
1336 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1337                                         int version)
1338 {
1339         if (in->shmmax > INT_MAX)
1340                 in->shmmax = INT_MAX;
1341         if (version == IPC_64) {
1342                 struct compat_shminfo64 info;
1343                 memset(&info, 0, sizeof(info));
1344                 info.shmmax = in->shmmax;
1345                 info.shmmin = in->shmmin;
1346                 info.shmmni = in->shmmni;
1347                 info.shmseg = in->shmseg;
1348                 info.shmall = in->shmall;
1349                 return copy_to_user(buf, &info, sizeof(info));
1350         } else {
1351                 struct shminfo info;
1352                 memset(&info, 0, sizeof(info));
1353                 info.shmmax = in->shmmax;
1354                 info.shmmin = in->shmmin;
1355                 info.shmmni = in->shmmni;
1356                 info.shmseg = in->shmseg;
1357                 info.shmall = in->shmall;
1358                 return copy_to_user(buf, &info, sizeof(info));
1359         }
1360 }
1361
1362 static int put_compat_shm_info(struct shm_info *ip,
1363                                 struct compat_shm_info __user *uip)
1364 {
1365         struct compat_shm_info info;
1366
1367         memset(&info, 0, sizeof(info));
1368         info.used_ids = ip->used_ids;
1369         info.shm_tot = ip->shm_tot;
1370         info.shm_rss = ip->shm_rss;
1371         info.shm_swp = ip->shm_swp;
1372         info.swap_attempts = ip->swap_attempts;
1373         info.swap_successes = ip->swap_successes;
1374         return copy_to_user(uip, &info, sizeof(info));
1375 }
1376
1377 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1378                                         int version)
1379 {
1380         if (version == IPC_64) {
1381                 struct compat_shmid64_ds v;
1382                 memset(&v, 0, sizeof(v));
1383                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1384                 v.shm_atime      = lower_32_bits(in->shm_atime);
1385                 v.shm_atime_high = upper_32_bits(in->shm_atime);
1386                 v.shm_dtime      = lower_32_bits(in->shm_dtime);
1387                 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1388                 v.shm_ctime      = lower_32_bits(in->shm_ctime);
1389                 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1390                 v.shm_segsz = in->shm_segsz;
1391                 v.shm_nattch = in->shm_nattch;
1392                 v.shm_cpid = in->shm_cpid;
1393                 v.shm_lpid = in->shm_lpid;
1394                 return copy_to_user(buf, &v, sizeof(v));
1395         } else {
1396                 struct compat_shmid_ds v;
1397                 memset(&v, 0, sizeof(v));
1398                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1399                 v.shm_perm.key = in->shm_perm.key;
1400                 v.shm_atime = in->shm_atime;
1401                 v.shm_dtime = in->shm_dtime;
1402                 v.shm_ctime = in->shm_ctime;
1403                 v.shm_segsz = in->shm_segsz;
1404                 v.shm_nattch = in->shm_nattch;
1405                 v.shm_cpid = in->shm_cpid;
1406                 v.shm_lpid = in->shm_lpid;
1407                 return copy_to_user(buf, &v, sizeof(v));
1408         }
1409 }
1410
1411 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1412                                         int version)
1413 {
1414         memset(out, 0, sizeof(*out));
1415         if (version == IPC_64) {
1416                 struct compat_shmid64_ds __user *p = buf;
1417                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1418         } else {
1419                 struct compat_shmid_ds __user *p = buf;
1420                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1421         }
1422 }
1423
1424 static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1425 {
1426         struct ipc_namespace *ns;
1427         struct shmid64_ds sem64;
1428         int err;
1429
1430         ns = current->nsproxy->ipc_ns;
1431
1432         if (cmd < 0 || shmid < 0)
1433                 return -EINVAL;
1434
1435         switch (cmd) {
1436         case IPC_INFO: {
1437                 struct shminfo64 shminfo;
1438                 err = shmctl_ipc_info(ns, &shminfo);
1439                 if (err < 0)
1440                         return err;
1441                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1442                         err = -EFAULT;
1443                 return err;
1444         }
1445         case SHM_INFO: {
1446                 struct shm_info shm_info;
1447                 err = shmctl_shm_info(ns, &shm_info);
1448                 if (err < 0)
1449                         return err;
1450                 if (put_compat_shm_info(&shm_info, uptr))
1451                         err = -EFAULT;
1452                 return err;
1453         }
1454         case IPC_STAT:
1455         case SHM_STAT_ANY:
1456         case SHM_STAT:
1457                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1458                 if (err < 0)
1459                         return err;
1460                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1461                         err = -EFAULT;
1462                 return err;
1463
1464         case IPC_SET:
1465                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1466                         return -EFAULT;
1467                 fallthrough;
1468         case IPC_RMID:
1469                 return shmctl_down(ns, shmid, cmd, &sem64);
1470         case SHM_LOCK:
1471         case SHM_UNLOCK:
1472                 return shmctl_do_lock(ns, shmid, cmd);
1473         default:
1474                 return -EINVAL;
1475         }
1476         return err;
1477 }
1478
1479 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1480 {
1481         return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1482 }
1483
1484 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1485 long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1486 {
1487         int version = compat_ipc_parse_version(&cmd);
1488
1489         return compat_ksys_shmctl(shmid, cmd, uptr, version);
1490 }
1491
1492 COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1493 {
1494         return compat_ksys_old_shmctl(shmid, cmd, uptr);
1495 }
1496 #endif
1497 #endif
1498
1499 /*
1500  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1501  *
1502  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1503  * "raddr" thing points to kernel space, and there has to be a wrapper around
1504  * this.
1505  */
1506 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1507               ulong *raddr, unsigned long shmlba)
1508 {
1509         struct shmid_kernel *shp;
1510         unsigned long addr = (unsigned long)shmaddr;
1511         unsigned long size;
1512         struct file *file, *base;
1513         int    err;
1514         unsigned long flags = MAP_SHARED;
1515         unsigned long prot;
1516         int acc_mode;
1517         struct ipc_namespace *ns;
1518         struct shm_file_data *sfd;
1519         int f_flags;
1520         unsigned long populate = 0;
1521
1522         err = -EINVAL;
1523         if (shmid < 0)
1524                 goto out;
1525
1526         if (addr) {
1527                 if (addr & (shmlba - 1)) {
1528                         if (shmflg & SHM_RND) {
1529                                 addr &= ~(shmlba - 1);  /* round down */
1530
1531                                 /*
1532                                  * Ensure that the round-down is non-nil
1533                                  * when remapping. This can happen for
1534                                  * cases when addr < shmlba.
1535                                  */
1536                                 if (!addr && (shmflg & SHM_REMAP))
1537                                         goto out;
1538                         } else
1539 #ifndef __ARCH_FORCE_SHMLBA
1540                                 if (addr & ~PAGE_MASK)
1541 #endif
1542                                         goto out;
1543                 }
1544
1545                 flags |= MAP_FIXED;
1546         } else if ((shmflg & SHM_REMAP))
1547                 goto out;
1548
1549         if (shmflg & SHM_RDONLY) {
1550                 prot = PROT_READ;
1551                 acc_mode = S_IRUGO;
1552                 f_flags = O_RDONLY;
1553         } else {
1554                 prot = PROT_READ | PROT_WRITE;
1555                 acc_mode = S_IRUGO | S_IWUGO;
1556                 f_flags = O_RDWR;
1557         }
1558         if (shmflg & SHM_EXEC) {
1559                 prot |= PROT_EXEC;
1560                 acc_mode |= S_IXUGO;
1561         }
1562
1563         /*
1564          * We cannot rely on the fs check since SYSV IPC does have an
1565          * additional creator id...
1566          */
1567         ns = current->nsproxy->ipc_ns;
1568         rcu_read_lock();
1569         shp = shm_obtain_object_check(ns, shmid);
1570         if (IS_ERR(shp)) {
1571                 err = PTR_ERR(shp);
1572                 goto out_unlock;
1573         }
1574
1575         err = -EACCES;
1576         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1577                 goto out_unlock;
1578
1579         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1580         if (err)
1581                 goto out_unlock;
1582
1583         ipc_lock_object(&shp->shm_perm);
1584
1585         /* check if shm_destroy() is tearing down shp */
1586         if (!ipc_valid_object(&shp->shm_perm)) {
1587                 ipc_unlock_object(&shp->shm_perm);
1588                 err = -EIDRM;
1589                 goto out_unlock;
1590         }
1591
1592         /*
1593          * We need to take a reference to the real shm file to prevent the
1594          * pointer from becoming stale in cases where the lifetime of the outer
1595          * file extends beyond that of the shm segment.  It's not usually
1596          * possible, but it can happen during remap_file_pages() emulation as
1597          * that unmaps the memory, then does ->mmap() via file reference only.
1598          * We'll deny the ->mmap() if the shm segment was since removed, but to
1599          * detect shm ID reuse we need to compare the file pointers.
1600          */
1601         base = get_file(shp->shm_file);
1602         shp->shm_nattch++;
1603         size = i_size_read(file_inode(base));
1604         ipc_unlock_object(&shp->shm_perm);
1605         rcu_read_unlock();
1606
1607         err = -ENOMEM;
1608         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1609         if (!sfd) {
1610                 fput(base);
1611                 goto out_nattch;
1612         }
1613
1614         file = alloc_file_clone(base, f_flags,
1615                           is_file_hugepages(base) ?
1616                                 &shm_file_operations_huge :
1617                                 &shm_file_operations);
1618         err = PTR_ERR(file);
1619         if (IS_ERR(file)) {
1620                 kfree(sfd);
1621                 fput(base);
1622                 goto out_nattch;
1623         }
1624
1625         sfd->id = shp->shm_perm.id;
1626         sfd->ns = get_ipc_ns(ns);
1627         sfd->file = base;
1628         sfd->vm_ops = NULL;
1629         file->private_data = sfd;
1630
1631         err = security_mmap_file(file, prot, flags);
1632         if (err)
1633                 goto out_fput;
1634
1635         if (mmap_write_lock_killable(current->mm)) {
1636                 err = -EINTR;
1637                 goto out_fput;
1638         }
1639
1640         if (addr && !(shmflg & SHM_REMAP)) {
1641                 err = -EINVAL;
1642                 if (addr + size < addr)
1643                         goto invalid;
1644
1645                 if (find_vma_intersection(current->mm, addr, addr + size))
1646                         goto invalid;
1647         }
1648
1649         addr = do_mmap(file, addr, size, prot, flags, 0, &populate, NULL);
1650         *raddr = addr;
1651         err = 0;
1652         if (IS_ERR_VALUE(addr))
1653                 err = (long)addr;
1654 invalid:
1655         mmap_write_unlock(current->mm);
1656         if (populate)
1657                 mm_populate(addr, populate);
1658
1659 out_fput:
1660         fput(file);
1661
1662 out_nattch:
1663         down_write(&shm_ids(ns).rwsem);
1664         shp = shm_lock(ns, shmid);
1665         shp->shm_nattch--;
1666
1667         if (shm_may_destroy(shp))
1668                 shm_destroy(ns, shp);
1669         else
1670                 shm_unlock(shp);
1671         up_write(&shm_ids(ns).rwsem);
1672         return err;
1673
1674 out_unlock:
1675         rcu_read_unlock();
1676 out:
1677         return err;
1678 }
1679
1680 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1681 {
1682         unsigned long ret;
1683         long err;
1684
1685         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1686         if (err)
1687                 return err;
1688         force_successful_syscall_return();
1689         return (long)ret;
1690 }
1691
1692 #ifdef CONFIG_COMPAT
1693
1694 #ifndef COMPAT_SHMLBA
1695 #define COMPAT_SHMLBA   SHMLBA
1696 #endif
1697
1698 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1699 {
1700         unsigned long ret;
1701         long err;
1702
1703         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1704         if (err)
1705                 return err;
1706         force_successful_syscall_return();
1707         return (long)ret;
1708 }
1709 #endif
1710
1711 /*
1712  * detach and kill segment if marked destroyed.
1713  * The work is done in shm_close.
1714  */
1715 long ksys_shmdt(char __user *shmaddr)
1716 {
1717         struct mm_struct *mm = current->mm;
1718         struct vm_area_struct *vma;
1719         unsigned long addr = (unsigned long)shmaddr;
1720         int retval = -EINVAL;
1721 #ifdef CONFIG_MMU
1722         loff_t size = 0;
1723         struct file *file;
1724         VMA_ITERATOR(vmi, mm, addr);
1725 #endif
1726
1727         if (addr & ~PAGE_MASK)
1728                 return retval;
1729
1730         if (mmap_write_lock_killable(mm))
1731                 return -EINTR;
1732
1733         /*
1734          * This function tries to be smart and unmap shm segments that
1735          * were modified by partial mlock or munmap calls:
1736          * - It first determines the size of the shm segment that should be
1737          *   unmapped: It searches for a vma that is backed by shm and that
1738          *   started at address shmaddr. It records it's size and then unmaps
1739          *   it.
1740          * - Then it unmaps all shm vmas that started at shmaddr and that
1741          *   are within the initially determined size and that are from the
1742          *   same shm segment from which we determined the size.
1743          * Errors from do_munmap are ignored: the function only fails if
1744          * it's called with invalid parameters or if it's called to unmap
1745          * a part of a vma. Both calls in this function are for full vmas,
1746          * the parameters are directly copied from the vma itself and always
1747          * valid - therefore do_munmap cannot fail. (famous last words?)
1748          */
1749         /*
1750          * If it had been mremap()'d, the starting address would not
1751          * match the usual checks anyway. So assume all vma's are
1752          * above the starting address given.
1753          */
1754
1755 #ifdef CONFIG_MMU
1756         for_each_vma(vmi, vma) {
1757                 /*
1758                  * Check if the starting address would match, i.e. it's
1759                  * a fragment created by mprotect() and/or munmap(), or it
1760                  * otherwise it starts at this address with no hassles.
1761                  */
1762                 if ((vma->vm_ops == &shm_vm_ops) &&
1763                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1764
1765                         /*
1766                          * Record the file of the shm segment being
1767                          * unmapped.  With mremap(), someone could place
1768                          * page from another segment but with equal offsets
1769                          * in the range we are unmapping.
1770                          */
1771                         file = vma->vm_file;
1772                         size = i_size_read(file_inode(vma->vm_file));
1773                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1774                         mas_pause(&vmi.mas);
1775                         /*
1776                          * We discovered the size of the shm segment, so
1777                          * break out of here and fall through to the next
1778                          * loop that uses the size information to stop
1779                          * searching for matching vma's.
1780                          */
1781                         retval = 0;
1782                         vma = vma_next(&vmi);
1783                         break;
1784                 }
1785         }
1786
1787         /*
1788          * We need look no further than the maximum address a fragment
1789          * could possibly have landed at. Also cast things to loff_t to
1790          * prevent overflows and make comparisons vs. equal-width types.
1791          */
1792         size = PAGE_ALIGN(size);
1793         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1794                 /* finding a matching vma now does not alter retval */
1795                 if ((vma->vm_ops == &shm_vm_ops) &&
1796                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1797                     (vma->vm_file == file)) {
1798                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1799                         mas_pause(&vmi.mas);
1800                 }
1801
1802                 vma = vma_next(&vmi);
1803         }
1804
1805 #else   /* CONFIG_MMU */
1806         vma = vma_lookup(mm, addr);
1807         /* under NOMMU conditions, the exact address to be destroyed must be
1808          * given
1809          */
1810         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1811                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1812                 retval = 0;
1813         }
1814
1815 #endif
1816
1817         mmap_write_unlock(mm);
1818         return retval;
1819 }
1820
1821 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1822 {
1823         return ksys_shmdt(shmaddr);
1824 }
1825
1826 #ifdef CONFIG_PROC_FS
1827 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1828 {
1829         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1830         struct user_namespace *user_ns = seq_user_ns(s);
1831         struct kern_ipc_perm *ipcp = it;
1832         struct shmid_kernel *shp;
1833         unsigned long rss = 0, swp = 0;
1834
1835         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1836         shm_add_rss_swap(shp, &rss, &swp);
1837
1838 #if BITS_PER_LONG <= 32
1839 #define SIZE_SPEC "%10lu"
1840 #else
1841 #define SIZE_SPEC "%21lu"
1842 #endif
1843
1844         seq_printf(s,
1845                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1846                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1847                    SIZE_SPEC " " SIZE_SPEC "\n",
1848                    shp->shm_perm.key,
1849                    shp->shm_perm.id,
1850                    shp->shm_perm.mode,
1851                    shp->shm_segsz,
1852                    pid_nr_ns(shp->shm_cprid, pid_ns),
1853                    pid_nr_ns(shp->shm_lprid, pid_ns),
1854                    shp->shm_nattch,
1855                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1856                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1857                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1858                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1859                    shp->shm_atim,
1860                    shp->shm_dtim,
1861                    shp->shm_ctim,
1862                    rss * PAGE_SIZE,
1863                    swp * PAGE_SIZE);
1864
1865         return 0;
1866 }
1867 #endif