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 <uapi/linux/shm.h>
  33 #include <linux/init.h>
  34 #include <linux/file.h>
  35 #include <linux/mman.h>
  36 #include <linux/shmem_fs.h>
  37 #include <linux/security.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/audit.h>
  40 #include <linux/capability.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/rwsem.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/mount.h>
  46 #include <linux/ipc_namespace.h>
  47 #include <linux/rhashtable.h>
  48
  49 #include <linux/uaccess.h>
  50
  51 #include "util.h"
  52
  53 struct shmid_kernel /* private to the kernel */
  54 {
  55         struct kern_ipc_perm    shm_perm;
  56         struct file             *shm_file;
  57         unsigned long           shm_nattch;
  58         unsigned long           shm_segsz;
  59         time64_t                shm_atim;
  60         time64_t                shm_dtim;
  61         time64_t                shm_ctim;
  62         struct pid              *shm_cprid;
  63         struct pid              *shm_lprid;
  64         struct ucounts          *mlock_ucounts;
  65
  66         /*
  67          * The task created the shm object, for
  68          * task_lock(shp->shm_creator)
  69          */
  70         struct task_struct      *shm_creator;
  71
  72         /*
  73          * List by creator. task_lock(->shm_creator) required for read/write.
  74          * If list_empty(), then the creator is dead already.
  75          */
  76         struct list_head        shm_clist;
  77         struct ipc_namespace    *ns;
  78 } __randomize_layout;
  79
  80 /* shm_mode upper byte flags */
  81 #define SHM_DEST        01000   /* segment will be destroyed on last detach */
  82 #define SHM_LOCKED      02000   /* segment will not be swapped */
  83
  84 struct shm_file_data {
  85         int id;
  86         struct ipc_namespace *ns;
  87         struct file *file;
  88         const struct vm_operations_struct *vm_ops;
  89 };
  90
  91 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  92
  93 static const struct file_operations shm_file_operations;
  94 static const struct vm_operations_struct shm_vm_ops;
  95
  96 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  97
  98 #define shm_unlock(shp)                 \
  99         ipc_unlock(&(shp)->shm_perm)
 100
 101 static int newseg(struct ipc_namespace *, struct ipc_params *);
 102 static void shm_open(struct vm_area_struct *vma);
 103 static void shm_close(struct vm_area_struct *vma);
 104 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
 105 #ifdef CONFIG_PROC_FS
 106 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
 107 #endif
 108
 109 void shm_init_ns(struct ipc_namespace *ns)
 110 {
 111         ns->shm_ctlmax = SHMMAX;
 112         ns->shm_ctlall = SHMALL;
 113         ns->shm_ctlmni = SHMMNI;
 114         ns->shm_rmid_forced = 0;
 115         ns->shm_tot = 0;
 116         ipc_init_ids(&shm_ids(ns));
 117 }
 118
 119 /*
 120  * Called with shm_ids.rwsem (writer) and the shp structure locked.
 121  * Only shm_ids.rwsem remains locked on exit.
 122  */
 123 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 124 {
 125         struct shmid_kernel *shp;
 126
 127         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 128         WARN_ON(ns != shp->ns);
 129
 130         if (shp->shm_nattch) {
 131                 shp->shm_perm.mode |= SHM_DEST;
 132                 /* Do not find it any more */
 133                 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
 134                 shm_unlock(shp);
 135         } else
 136                 shm_destroy(ns, shp);
 137 }
 138
 139 #ifdef CONFIG_IPC_NS
 140 void shm_exit_ns(struct ipc_namespace *ns)
 141 {
 142         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 143         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 144         rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
 145 }
 146 #endif
 147
 148 static int __init ipc_ns_init(void)
 149 {
 150         shm_init_ns(&init_ipc_ns);
 151         return 0;
 152 }
 153
 154 pure_initcall(ipc_ns_init);
 155
 156 void __init shm_init(void)
 157 {
 158         ipc_init_proc_interface("sysvipc/shm",
 159 #if BITS_PER_LONG <= 32
 160                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 161 #else
 162                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 163 #endif
 164                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 165 }
 166
 167 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 168 {
 169         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 170
 171         if (IS_ERR(ipcp))
 172                 return ERR_CAST(ipcp);
 173
 174         return container_of(ipcp, struct shmid_kernel, shm_perm);
 175 }
 176
 177 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 178 {
 179         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 180
 181         if (IS_ERR(ipcp))
 182                 return ERR_CAST(ipcp);
 183
 184         return container_of(ipcp, struct shmid_kernel, shm_perm);
 185 }
 186
 187 /*
 188  * shm_lock_(check_) routines are called in the paths where the rwsem
 189  * is not necessarily held.
 190  */
 191 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 192 {
 193         struct kern_ipc_perm *ipcp;
 194
 195         rcu_read_lock();
 196         ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 197         if (IS_ERR(ipcp))
 198                 goto err;
 199
 200         ipc_lock_object(ipcp);
 201         /*
 202          * ipc_rmid() may have already freed the ID while ipc_lock_object()
 203          * was spinning: here verify that the structure is still valid.
 204          * Upon races with RMID, return -EIDRM, thus indicating that
 205          * the ID points to a removed identifier.
 206          */
 207         if (ipc_valid_object(ipcp)) {
 208                 /* return a locked ipc object upon success */
 209                 return container_of(ipcp, struct shmid_kernel, shm_perm);
 210         }
 211
 212         ipc_unlock_object(ipcp);
 213         ipcp = ERR_PTR(-EIDRM);
 214 err:
 215         rcu_read_unlock();
 216         /*
 217          * Callers of shm_lock() must validate the status of the returned ipc
 218          * object pointer and error out as appropriate.
 219          */
 220         return ERR_CAST(ipcp);
 221 }
 222
 223 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 224 {
 225         rcu_read_lock();
 226         ipc_lock_object(&ipcp->shm_perm);
 227 }
 228
 229 static void shm_rcu_free(struct rcu_head *head)
 230 {
 231         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 232                                                         rcu);
 233         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 234                                                         shm_perm);
 235         security_shm_free(&shp->shm_perm);
 236         kfree(shp);
 237 }
 238
 239 /*
 240  * It has to be called with shp locked.
 241  * It must be called before ipc_rmid()
 242  */
 243 static inline void shm_clist_rm(struct shmid_kernel *shp)
 244 {
 245         struct task_struct *creator;
 246
 247         /* ensure that shm_creator does not disappear */
 248         rcu_read_lock();
 249
 250         /*
 251          * A concurrent exit_shm may do a list_del_init() as well.
 252          * Just do nothing if exit_shm already did the work
 253          */
 254         if (!list_empty(&shp->shm_clist)) {
 255                 /*
 256                  * shp->shm_creator is guaranteed to be valid *only*
 257                  * if shp->shm_clist is not empty.
 258                  */
 259                 creator = shp->shm_creator;
 260
 261                 task_lock(creator);
 262                 /*
 263                  * list_del_init() is a nop if the entry was already removed
 264                  * from the list.
 265                  */
 266                 list_del_init(&shp->shm_clist);
 267                 task_unlock(creator);
 268         }
 269         rcu_read_unlock();
 270 }
 271
 272 static inline void shm_rmid(struct shmid_kernel *s)
 273 {
 274         shm_clist_rm(s);
 275         ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
 276 }
 277
 278
 279 static int __shm_open(struct shm_file_data *sfd)
 280 {
 281         struct shmid_kernel *shp;
 282
 283         shp = shm_lock(sfd->ns, sfd->id);
 284
 285         if (IS_ERR(shp))
 286                 return PTR_ERR(shp);
 287
 288         if (shp->shm_file != sfd->file) {
 289                 /* ID was reused */
 290                 shm_unlock(shp);
 291                 return -EINVAL;
 292         }
 293
 294         shp->shm_atim = ktime_get_real_seconds();
 295         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 296         shp->shm_nattch++;
 297         shm_unlock(shp);
 298         return 0;
 299 }
 300
 301 /* This is called by fork, once for every shm attach. */
 302 static void shm_open(struct vm_area_struct *vma)
 303 {
 304         struct file *file = vma->vm_file;
 305         struct shm_file_data *sfd = shm_file_data(file);
 306         int err;
 307
 308         /* Always call underlying open if present */
 309         if (sfd->vm_ops->open)
 310                 sfd->vm_ops->open(vma);
 311
 312         err = __shm_open(sfd);
 313         /*
 314          * We raced in the idr lookup or with shm_destroy().
 315          * Either way, the ID is busted.
 316          */
 317         WARN_ON_ONCE(err);
 318 }
 319
 320 /*
 321  * shm_destroy - free the struct shmid_kernel
 322  *
 323  * @ns: namespace
 324  * @shp: struct to free
 325  *
 326  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 327  * but returns with shp unlocked and freed.
 328  */
 329 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 330 {
 331         struct file *shm_file;
 332
 333         shm_file = shp->shm_file;
 334         shp->shm_file = NULL;
 335         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 336         shm_rmid(shp);
 337         shm_unlock(shp);
 338         if (!is_file_hugepages(shm_file))
 339                 shmem_lock(shm_file, 0, shp->mlock_ucounts);
 340         fput(shm_file);
 341         ipc_update_pid(&shp->shm_cprid, NULL);
 342         ipc_update_pid(&shp->shm_lprid, NULL);
 343         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 344 }
 345
 346 /*
 347  * shm_may_destroy - identifies whether shm segment should be destroyed now
 348  *
 349  * Returns true if and only if there are no active users of the segment and
 350  * one of the following is true:
 351  *
 352  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 353  *
 354  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 355  */
 356 static bool shm_may_destroy(struct shmid_kernel *shp)
 357 {
 358         return (shp->shm_nattch == 0) &&
 359                (shp->ns->shm_rmid_forced ||
 360                 (shp->shm_perm.mode & SHM_DEST));
 361 }
 362
 363 /*
 364  * remove the attach descriptor vma.
 365  * free memory for segment if it is marked destroyed.
 366  * The descriptor has already been removed from the current->mm->mmap list
 367  * and will later be kfree()d.
 368  */
 369 static void __shm_close(struct shm_file_data *sfd)
 370 {
 371         struct shmid_kernel *shp;
 372         struct ipc_namespace *ns = sfd->ns;
 373
 374         down_write(&shm_ids(ns).rwsem);
 375         /* remove from the list of attaches of the shm segment */
 376         shp = shm_lock(ns, sfd->id);
 377
 378         /*
 379          * We raced in the idr lookup or with shm_destroy().
 380          * Either way, the ID is busted.
 381          */
 382         if (WARN_ON_ONCE(IS_ERR(shp)))
 383                 goto done; /* no-op */
 384
 385         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 386         shp->shm_dtim = ktime_get_real_seconds();
 387         shp->shm_nattch--;
 388         if (shm_may_destroy(shp))
 389                 shm_destroy(ns, shp);
 390         else
 391                 shm_unlock(shp);
 392 done:
 393         up_write(&shm_ids(ns).rwsem);
 394 }
 395
 396 static void shm_close(struct vm_area_struct *vma)
 397 {
 398         struct file *file = vma->vm_file;
 399         struct shm_file_data *sfd = shm_file_data(file);
 400
 401         /* Always call underlying close if present */
 402         if (sfd->vm_ops->close)
 403                 sfd->vm_ops->close(vma);
 404
 405         __shm_close(sfd);
 406 }
 407
 408 /* Called with ns->shm_ids(ns).rwsem locked */
 409 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 410 {
 411         struct ipc_namespace *ns = data;
 412         struct kern_ipc_perm *ipcp = p;
 413         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 414
 415         /*
 416          * We want to destroy segments without users and with already
 417          * exit'ed originating process.
 418          *
 419          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 420          */
 421         if (!list_empty(&shp->shm_clist))
 422                 return 0;
 423
 424         if (shm_may_destroy(shp)) {
 425                 shm_lock_by_ptr(shp);
 426                 shm_destroy(ns, shp);
 427         }
 428         return 0;
 429 }
 430
 431 void shm_destroy_orphaned(struct ipc_namespace *ns)
 432 {
 433         down_write(&shm_ids(ns).rwsem);
 434         if (shm_ids(ns).in_use)
 435                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 436         up_write(&shm_ids(ns).rwsem);
 437 }
 438
 439 /* Locking assumes this will only be called with task == current */
 440 void exit_shm(struct task_struct *task)
 441 {
 442         for (;;) {
 443                 struct shmid_kernel *shp;
 444                 struct ipc_namespace *ns;
 445
 446                 task_lock(task);
 447
 448                 if (list_empty(&task->sysvshm.shm_clist)) {
 449                         task_unlock(task);
 450                         break;
 451                 }
 452
 453                 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
 454                                 shm_clist);
 455
 456                 /*
 457                  * 1) Get pointer to the ipc namespace. It is worth to say
 458                  * that this pointer is guaranteed to be valid because
 459                  * shp lifetime is always shorter than namespace lifetime
 460                  * in which shp lives.
 461                  * We taken task_lock it means that shp won't be freed.
 462                  */
 463                 ns = shp->ns;
 464
 465                 /*
 466                  * 2) If kernel.shm_rmid_forced is not set then only keep track of
 467                  * which shmids are orphaned, so that a later set of the sysctl
 468                  * can clean them up.
 469                  */
 470                 if (!ns->shm_rmid_forced)
 471                         goto unlink_continue;
 472
 473                 /*
 474                  * 3) get a reference to the namespace.
 475                  *    The refcount could be already 0. If it is 0, then
 476                  *    the shm objects will be free by free_ipc_work().
 477                  */
 478                 ns = get_ipc_ns_not_zero(ns);
 479                 if (!ns) {
 480 unlink_continue:
 481                         list_del_init(&shp->shm_clist);
 482                         task_unlock(task);
 483                         continue;
 484                 }
 485
 486                 /*
 487                  * 4) get a reference to shp.
 488                  *   This cannot fail: shm_clist_rm() is called before
 489                  *   ipc_rmid(), thus the refcount cannot be 0.
 490                  */
 491                 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
 492
 493                 /*
 494                  * 5) unlink the shm segment from the list of segments
 495                  *    created by current.
 496                  *    This must be done last. After unlinking,
 497                  *    only the refcounts obtained above prevent IPC_RMID
 498                  *    from destroying the segment or the namespace.
 499                  */
 500                 list_del_init(&shp->shm_clist);
 501
 502                 task_unlock(task);
 503
 504                 /*
 505                  * 6) we have all references
 506                  *    Thus lock & if needed destroy shp.
 507                  */
 508                 down_write(&shm_ids(ns).rwsem);
 509                 shm_lock_by_ptr(shp);
 510                 /*
 511                  * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
 512                  * safe to call ipc_rcu_putref here
 513                  */
 514                 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 515
 516                 if (ipc_valid_object(&shp->shm_perm)) {
 517                         if (shm_may_destroy(shp))
 518                                 shm_destroy(ns, shp);
 519                         else
 520                                 shm_unlock(shp);
 521                 } else {
 522                         /*
 523                          * Someone else deleted the shp from namespace
 524                          * idr/kht while we have waited.
 525                          * Just unlock and continue.
 526                          */
 527                         shm_unlock(shp);
 528                 }
 529
 530                 up_write(&shm_ids(ns).rwsem);
 531                 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
 532         }
 533 }
 534
 535 static vm_fault_t shm_fault(struct vm_fault *vmf)
 536 {
 537         struct file *file = vmf->vma->vm_file;
 538         struct shm_file_data *sfd = shm_file_data(file);
 539
 540         return sfd->vm_ops->fault(vmf);
 541 }
 542
 543 static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
 544 {
 545         struct file *file = vma->vm_file;
 546         struct shm_file_data *sfd = shm_file_data(file);
 547
 548         if (sfd->vm_ops->may_split)
 549                 return sfd->vm_ops->may_split(vma, addr);
 550
 551         return 0;
 552 }
 553
 554 static unsigned long shm_pagesize(struct vm_area_struct *vma)
 555 {
 556         struct file *file = vma->vm_file;
 557         struct shm_file_data *sfd = shm_file_data(file);
 558
 559         if (sfd->vm_ops->pagesize)
 560                 return sfd->vm_ops->pagesize(vma);
 561
 562         return PAGE_SIZE;
 563 }
 564
 565 #ifdef CONFIG_NUMA
 566 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
 567 {
 568         struct shm_file_data *sfd = shm_file_data(vma->vm_file);
 569         int err = 0;
 570
 571         if (sfd->vm_ops->set_policy)
 572                 err = sfd->vm_ops->set_policy(vma, mpol);
 573         return err;
 574 }
 575
 576 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 577                                         unsigned long addr, pgoff_t *ilx)
 578 {
 579         struct shm_file_data *sfd = shm_file_data(vma->vm_file);
 580         struct mempolicy *mpol = vma->vm_policy;
 581
 582         if (sfd->vm_ops->get_policy)
 583                 mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
 584         return mpol;
 585 }
 586 #endif
 587
 588 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 589 {
 590         struct shm_file_data *sfd = shm_file_data(file);
 591         int ret;
 592
 593         /*
 594          * In case of remap_file_pages() emulation, the file can represent an
 595          * IPC ID that was removed, and possibly even reused by another shm
 596          * segment already.  Propagate this case as an error to caller.
 597          */
 598         ret = __shm_open(sfd);
 599         if (ret)
 600                 return ret;
 601
 602         ret = call_mmap(sfd->file, vma);
 603         if (ret) {
 604                 __shm_close(sfd);
 605                 return ret;
 606         }
 607         sfd->vm_ops = vma->vm_ops;
 608 #ifdef CONFIG_MMU
 609         WARN_ON(!sfd->vm_ops->fault);
 610 #endif
 611         vma->vm_ops = &shm_vm_ops;
 612         return 0;
 613 }
 614
 615 static int shm_release(struct inode *ino, struct file *file)
 616 {
 617         struct shm_file_data *sfd = shm_file_data(file);
 618
 619         put_ipc_ns(sfd->ns);
 620         fput(sfd->file);
 621         shm_file_data(file) = NULL;
 622         kfree(sfd);
 623         return 0;
 624 }
 625
 626 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 627 {
 628         struct shm_file_data *sfd = shm_file_data(file);
 629
 630         if (!sfd->file->f_op->fsync)
 631                 return -EINVAL;
 632         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 633 }
 634
 635 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 636                           loff_t len)
 637 {
 638         struct shm_file_data *sfd = shm_file_data(file);
 639
 640         if (!sfd->file->f_op->fallocate)
 641                 return -EOPNOTSUPP;
 642         return sfd->file->f_op->fallocate(file, mode, offset, len);
 643 }
 644
 645 static unsigned long shm_get_unmapped_area(struct file *file,
 646         unsigned long addr, unsigned long len, unsigned long pgoff,
 647         unsigned long flags)
 648 {
 649         struct shm_file_data *sfd = shm_file_data(file);
 650
 651         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 652                                                 pgoff, flags);
 653 }
 654
 655 static const struct file_operations shm_file_operations = {
 656         .mmap           = shm_mmap,
 657         .fsync          = shm_fsync,
 658         .release        = shm_release,
 659         .get_unmapped_area      = shm_get_unmapped_area,
 660         .llseek         = noop_llseek,
 661         .fallocate      = shm_fallocate,
 662 };
 663
 664 /*
 665  * shm_file_operations_huge is now identical to shm_file_operations,
 666  * but we keep it distinct for the sake of is_file_shm_hugepages().
 667  */
 668 static const struct file_operations shm_file_operations_huge = {
 669         .mmap           = shm_mmap,
 670         .fsync          = shm_fsync,
 671         .release        = shm_release,
 672         .get_unmapped_area      = shm_get_unmapped_area,
 673         .llseek         = noop_llseek,
 674         .fallocate      = shm_fallocate,
 675 };
 676
 677 bool is_file_shm_hugepages(struct file *file)
 678 {
 679         return file->f_op == &shm_file_operations_huge;
 680 }
 681
 682 static const struct vm_operations_struct shm_vm_ops = {
 683         .open   = shm_open,     /* callback for a new vm-area open */
 684         .close  = shm_close,    /* callback for when the vm-area is released */
 685         .fault  = shm_fault,
 686         .may_split = shm_may_split,
 687         .pagesize = shm_pagesize,
 688 #if defined(CONFIG_NUMA)
 689         .set_policy = shm_set_policy,
 690         .get_policy = shm_get_policy,
 691 #endif
 692 };
 693
 694 /**
 695  * newseg - Create a new shared memory segment
 696  * @ns: namespace
 697  * @params: ptr to the structure that contains key, size and shmflg
 698  *
 699  * Called with shm_ids.rwsem held as a writer.
 700  */
 701 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 702 {
 703         key_t key = params->key;
 704         int shmflg = params->flg;
 705         size_t size = params->u.size;
 706         int error;
 707         struct shmid_kernel *shp;
 708         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 709         struct file *file;
 710         char name[13];
 711         vm_flags_t acctflag = 0;
 712
 713         if (size < SHMMIN || size > ns->shm_ctlmax)
 714                 return -EINVAL;
 715
 716         if (numpages << PAGE_SHIFT < size)
 717                 return -ENOSPC;
 718
 719         if (ns->shm_tot + numpages < ns->shm_tot ||
 720                         ns->shm_tot + numpages > ns->shm_ctlall)
 721                 return -ENOSPC;
 722
 723         shp = kmalloc(sizeof(*shp), GFP_KERNEL_ACCOUNT);
 724         if (unlikely(!shp))
 725                 return -ENOMEM;
 726
 727         shp->shm_perm.key = key;
 728         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 729         shp->mlock_ucounts = NULL;
 730
 731         shp->shm_perm.security = NULL;
 732         error = security_shm_alloc(&shp->shm_perm);
 733         if (error) {
 734                 kfree(shp);
 735                 return error;
 736         }
 737
 738         sprintf(name, "SYSV%08x", key);
 739         if (shmflg & SHM_HUGETLB) {
 740                 struct hstate *hs;
 741                 size_t hugesize;
 742
 743                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 744                 if (!hs) {
 745                         error = -EINVAL;
 746                         goto no_file;
 747                 }
 748                 hugesize = ALIGN(size, huge_page_size(hs));
 749
 750                 /* hugetlb_file_setup applies strict accounting */
 751                 if (shmflg & SHM_NORESERVE)
 752                         acctflag = VM_NORESERVE;
 753                 file = hugetlb_file_setup(name, hugesize, acctflag,
 754                                 HUGETLB_SHMFS_INODE, (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 755         } else {
 756                 /*
 757                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 758                  * if it's asked for.
 759                  */
 760                 if  ((shmflg & SHM_NORESERVE) &&
 761                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 762                         acctflag = VM_NORESERVE;
 763                 file = shmem_kernel_file_setup(name, size, acctflag);
 764         }
 765         error = PTR_ERR(file);
 766         if (IS_ERR(file))
 767                 goto no_file;
 768
 769         shp->shm_cprid = get_pid(task_tgid(current));
 770         shp->shm_lprid = NULL;
 771         shp->shm_atim = shp->shm_dtim = 0;
 772         shp->shm_ctim = ktime_get_real_seconds();
 773         shp->shm_segsz = size;
 774         shp->shm_nattch = 0;
 775         shp->shm_file = file;
 776         shp->shm_creator = current;
 777
 778         /* ipc_addid() locks shp upon success. */
 779         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 780         if (error < 0)
 781                 goto no_id;
 782
 783         shp->ns = ns;
 784
 785         task_lock(current);
 786         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 787         task_unlock(current);
 788
 789         /*
 790          * shmid gets reported as "inode#" in /proc/pid/maps.
 791          * proc-ps tools use this. Changing this will break them.
 792          */
 793         file_inode(file)->i_ino = shp->shm_perm.id;
 794
 795         ns->shm_tot += numpages;
 796         error = shp->shm_perm.id;
 797
 798         ipc_unlock_object(&shp->shm_perm);
 799         rcu_read_unlock();
 800         return error;
 801
 802 no_id:
 803         ipc_update_pid(&shp->shm_cprid, NULL);
 804         ipc_update_pid(&shp->shm_lprid, NULL);
 805         fput(file);
 806         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 807         return error;
 808 no_file:
 809         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 810         return error;
 811 }
 812
 813 /*
 814  * Called with shm_ids.rwsem and ipcp locked.
 815  */
 816 static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
 817 {
 818         struct shmid_kernel *shp;
 819
 820         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 821         if (shp->shm_segsz < params->u.size)
 822                 return -EINVAL;
 823
 824         return 0;
 825 }
 826
 827 long ksys_shmget(key_t key, size_t size, int shmflg)
 828 {
 829         struct ipc_namespace *ns;
 830         static const struct ipc_ops shm_ops = {
 831                 .getnew = newseg,
 832                 .associate = security_shm_associate,
 833                 .more_checks = shm_more_checks,
 834         };
 835         struct ipc_params shm_params;
 836
 837         ns = current->nsproxy->ipc_ns;
 838
 839         shm_params.key = key;
 840         shm_params.flg = shmflg;
 841         shm_params.u.size = size;
 842
 843         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 844 }
 845
 846 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 847 {
 848         return ksys_shmget(key, size, shmflg);
 849 }
 850
 851 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 852 {
 853         switch (version) {
 854         case IPC_64:
 855                 return copy_to_user(buf, in, sizeof(*in));
 856         case IPC_OLD:
 857             {
 858                 struct shmid_ds out;
 859
 860                 memset(&out, 0, sizeof(out));
 861                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 862                 out.shm_segsz   = in->shm_segsz;
 863                 out.shm_atime   = in->shm_atime;
 864                 out.shm_dtime   = in->shm_dtime;
 865                 out.shm_ctime   = in->shm_ctime;
 866                 out.shm_cpid    = in->shm_cpid;
 867                 out.shm_lpid    = in->shm_lpid;
 868                 out.shm_nattch  = in->shm_nattch;
 869
 870                 return copy_to_user(buf, &out, sizeof(out));
 871             }
 872         default:
 873                 return -EINVAL;
 874         }
 875 }
 876
 877 static inline unsigned long
 878 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 879 {
 880         switch (version) {
 881         case IPC_64:
 882                 if (copy_from_user(out, buf, sizeof(*out)))
 883                         return -EFAULT;
 884                 return 0;
 885         case IPC_OLD:
 886             {
 887                 struct shmid_ds tbuf_old;
 888
 889                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 890                         return -EFAULT;
 891
 892                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 893                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 894                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 895
 896                 return 0;
 897             }
 898         default:
 899                 return -EINVAL;
 900         }
 901 }
 902
 903 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 904 {
 905         switch (version) {
 906         case IPC_64:
 907                 return copy_to_user(buf, in, sizeof(*in));
 908         case IPC_OLD:
 909             {
 910                 struct shminfo out;
 911
 912                 if (in->shmmax > INT_MAX)
 913                         out.shmmax = INT_MAX;
 914                 else
 915                         out.shmmax = (int)in->shmmax;
 916
 917                 out.shmmin      = in->shmmin;
 918                 out.shmmni      = in->shmmni;
 919                 out.shmseg      = in->shmseg;
 920                 out.shmall      = in->shmall;
 921
 922                 return copy_to_user(buf, &out, sizeof(out));
 923             }
 924         default:
 925                 return -EINVAL;
 926         }
 927 }
 928
 929 /*
 930  * Calculate and add used RSS and swap pages of a shm.
 931  * Called with shm_ids.rwsem held as a reader
 932  */
 933 static void shm_add_rss_swap(struct shmid_kernel *shp,
 934         unsigned long *rss_add, unsigned long *swp_add)
 935 {
 936         struct inode *inode;
 937
 938         inode = file_inode(shp->shm_file);
 939
 940         if (is_file_hugepages(shp->shm_file)) {
 941                 struct address_space *mapping = inode->i_mapping;
 942                 struct hstate *h = hstate_file(shp->shm_file);
 943                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 944         } else {
 945 #ifdef CONFIG_SHMEM
 946                 struct shmem_inode_info *info = SHMEM_I(inode);
 947
 948                 spin_lock_irq(&info->lock);
 949                 *rss_add += inode->i_mapping->nrpages;
 950                 *swp_add += info->swapped;
 951                 spin_unlock_irq(&info->lock);
 952 #else
 953                 *rss_add += inode->i_mapping->nrpages;
 954 #endif
 955         }
 956 }
 957
 958 /*
 959  * Called with shm_ids.rwsem held as a reader
 960  */
 961 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 962                 unsigned long *swp)
 963 {
 964         int next_id;
 965         int total, in_use;
 966
 967         *rss = 0;
 968         *swp = 0;
 969
 970         in_use = shm_ids(ns).in_use;
 971
 972         for (total = 0, next_id = 0; total < in_use; next_id++) {
 973                 struct kern_ipc_perm *ipc;
 974                 struct shmid_kernel *shp;
 975
 976                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 977                 if (ipc == NULL)
 978                         continue;
 979                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 980
 981                 shm_add_rss_swap(shp, rss, swp);
 982
 983                 total++;
 984         }
 985 }
 986
 987 /*
 988  * This function handles some shmctl commands which require the rwsem
 989  * to be held in write mode.
 990  * NOTE: no locks must be held, the rwsem is taken inside this function.
 991  */
 992 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 993                        struct shmid64_ds *shmid64)
 994 {
 995         struct kern_ipc_perm *ipcp;
 996         struct shmid_kernel *shp;
 997         int err;
 998
 999         down_write(&shm_ids(ns).rwsem);
1000         rcu_read_lock();
1001
1002         ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
1003                                       &shmid64->shm_perm, 0);
1004         if (IS_ERR(ipcp)) {
1005                 err = PTR_ERR(ipcp);
1006                 goto out_unlock1;
1007         }
1008
1009         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1010
1011         err = security_shm_shmctl(&shp->shm_perm, cmd);
1012         if (err)
1013                 goto out_unlock1;
1014
1015         switch (cmd) {
1016         case IPC_RMID:
1017                 ipc_lock_object(&shp->shm_perm);
1018                 /* do_shm_rmid unlocks the ipc object and rcu */
1019                 do_shm_rmid(ns, ipcp);
1020                 goto out_up;
1021         case IPC_SET:
1022                 ipc_lock_object(&shp->shm_perm);
1023                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1024                 if (err)
1025                         goto out_unlock0;
1026                 shp->shm_ctim = ktime_get_real_seconds();
1027                 break;
1028         default:
1029                 err = -EINVAL;
1030                 goto out_unlock1;
1031         }
1032
1033 out_unlock0:
1034         ipc_unlock_object(&shp->shm_perm);
1035 out_unlock1:
1036         rcu_read_unlock();
1037 out_up:
1038         up_write(&shm_ids(ns).rwsem);
1039         return err;
1040 }
1041
1042 static int shmctl_ipc_info(struct ipc_namespace *ns,
1043                            struct shminfo64 *shminfo)
1044 {
1045         int err = security_shm_shmctl(NULL, IPC_INFO);
1046         if (!err) {
1047                 memset(shminfo, 0, sizeof(*shminfo));
1048                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1049                 shminfo->shmmax = ns->shm_ctlmax;
1050                 shminfo->shmall = ns->shm_ctlall;
1051                 shminfo->shmmin = SHMMIN;
1052                 down_read(&shm_ids(ns).rwsem);
1053                 err = ipc_get_maxidx(&shm_ids(ns));
1054                 up_read(&shm_ids(ns).rwsem);
1055                 if (err < 0)
1056                         err = 0;
1057         }
1058         return err;
1059 }
1060
1061 static int shmctl_shm_info(struct ipc_namespace *ns,
1062                            struct shm_info *shm_info)
1063 {
1064         int err = security_shm_shmctl(NULL, SHM_INFO);
1065         if (!err) {
1066                 memset(shm_info, 0, sizeof(*shm_info));
1067                 down_read(&shm_ids(ns).rwsem);
1068                 shm_info->used_ids = shm_ids(ns).in_use;
1069                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1070                 shm_info->shm_tot = ns->shm_tot;
1071                 shm_info->swap_attempts = 0;
1072                 shm_info->swap_successes = 0;
1073                 err = ipc_get_maxidx(&shm_ids(ns));
1074                 up_read(&shm_ids(ns).rwsem);
1075                 if (err < 0)
1076                         err = 0;
1077         }
1078         return err;
1079 }
1080
1081 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1082                         int cmd, struct shmid64_ds *tbuf)
1083 {
1084         struct shmid_kernel *shp;
1085         int err;
1086
1087         memset(tbuf, 0, sizeof(*tbuf));
1088
1089         rcu_read_lock();
1090         if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1091                 shp = shm_obtain_object(ns, shmid);
1092                 if (IS_ERR(shp)) {
1093                         err = PTR_ERR(shp);
1094                         goto out_unlock;
1095                 }
1096         } else { /* IPC_STAT */
1097                 shp = shm_obtain_object_check(ns, shmid);
1098                 if (IS_ERR(shp)) {
1099                         err = PTR_ERR(shp);
1100                         goto out_unlock;
1101                 }
1102         }
1103
1104         /*
1105          * Semantically SHM_STAT_ANY ought to be identical to
1106          * that functionality provided by the /proc/sysvipc/
1107          * interface. As such, only audit these calls and
1108          * do not do traditional S_IRUGO permission checks on
1109          * the ipc object.
1110          */
1111         if (cmd == SHM_STAT_ANY)
1112                 audit_ipc_obj(&shp->shm_perm);
1113         else {
1114                 err = -EACCES;
1115                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1116                         goto out_unlock;
1117         }
1118
1119         err = security_shm_shmctl(&shp->shm_perm, cmd);
1120         if (err)
1121                 goto out_unlock;
1122
1123         ipc_lock_object(&shp->shm_perm);
1124
1125         if (!ipc_valid_object(&shp->shm_perm)) {
1126                 ipc_unlock_object(&shp->shm_perm);
1127                 err = -EIDRM;
1128                 goto out_unlock;
1129         }
1130
1131         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1132         tbuf->shm_segsz = shp->shm_segsz;
1133         tbuf->shm_atime = shp->shm_atim;
1134         tbuf->shm_dtime = shp->shm_dtim;
1135         tbuf->shm_ctime = shp->shm_ctim;
1136 #ifndef CONFIG_64BIT
1137         tbuf->shm_atime_high = shp->shm_atim >> 32;
1138         tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1139         tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1140 #endif
1141         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
1142         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
1143         tbuf->shm_nattch = shp->shm_nattch;
1144
1145         if (cmd == IPC_STAT) {
1146                 /*
1147                  * As defined in SUS:
1148                  * Return 0 on success
1149                  */
1150                 err = 0;
1151         } else {
1152                 /*
1153                  * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1154                  * Return the full id, including the sequence number
1155                  */
1156                 err = shp->shm_perm.id;
1157         }
1158
1159         ipc_unlock_object(&shp->shm_perm);
1160 out_unlock:
1161         rcu_read_unlock();
1162         return err;
1163 }
1164
1165 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1166 {
1167         struct shmid_kernel *shp;
1168         struct file *shm_file;
1169         int err;
1170
1171         rcu_read_lock();
1172         shp = shm_obtain_object_check(ns, shmid);
1173         if (IS_ERR(shp)) {
1174                 err = PTR_ERR(shp);
1175                 goto out_unlock1;
1176         }
1177
1178         audit_ipc_obj(&(shp->shm_perm));
1179         err = security_shm_shmctl(&shp->shm_perm, cmd);
1180         if (err)
1181                 goto out_unlock1;
1182
1183         ipc_lock_object(&shp->shm_perm);
1184
1185         /* check if shm_destroy() is tearing down shp */
1186         if (!ipc_valid_object(&shp->shm_perm)) {
1187                 err = -EIDRM;
1188                 goto out_unlock0;
1189         }
1190
1191         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1192                 kuid_t euid = current_euid();
1193
1194                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1195                     !uid_eq(euid, shp->shm_perm.cuid)) {
1196                         err = -EPERM;
1197                         goto out_unlock0;
1198                 }
1199                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1200                         err = -EPERM;
1201                         goto out_unlock0;
1202                 }
1203         }
1204
1205         shm_file = shp->shm_file;
1206         if (is_file_hugepages(shm_file))
1207                 goto out_unlock0;
1208
1209         if (cmd == SHM_LOCK) {
1210                 struct ucounts *ucounts = current_ucounts();
1211
1212                 err = shmem_lock(shm_file, 1, ucounts);
1213                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1214                         shp->shm_perm.mode |= SHM_LOCKED;
1215                         shp->mlock_ucounts = ucounts;
1216                 }
1217                 goto out_unlock0;
1218         }
1219
1220         /* SHM_UNLOCK */
1221         if (!(shp->shm_perm.mode & SHM_LOCKED))
1222                 goto out_unlock0;
1223         shmem_lock(shm_file, 0, shp->mlock_ucounts);
1224         shp->shm_perm.mode &= ~SHM_LOCKED;
1225         shp->mlock_ucounts = NULL;
1226         get_file(shm_file);
1227         ipc_unlock_object(&shp->shm_perm);
1228         rcu_read_unlock();
1229         shmem_unlock_mapping(shm_file->f_mapping);
1230
1231         fput(shm_file);
1232         return err;
1233
1234 out_unlock0:
1235         ipc_unlock_object(&shp->shm_perm);
1236 out_unlock1:
1237         rcu_read_unlock();
1238         return err;
1239 }
1240
1241 static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1242 {
1243         int err;
1244         struct ipc_namespace *ns;
1245         struct shmid64_ds sem64;
1246
1247         if (cmd < 0 || shmid < 0)
1248                 return -EINVAL;
1249
1250         ns = current->nsproxy->ipc_ns;
1251
1252         switch (cmd) {
1253         case IPC_INFO: {
1254                 struct shminfo64 shminfo;
1255                 err = shmctl_ipc_info(ns, &shminfo);
1256                 if (err < 0)
1257                         return err;
1258                 if (copy_shminfo_to_user(buf, &shminfo, version))
1259                         err = -EFAULT;
1260                 return err;
1261         }
1262         case SHM_INFO: {
1263                 struct shm_info shm_info;
1264                 err = shmctl_shm_info(ns, &shm_info);
1265                 if (err < 0)
1266                         return err;
1267                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1268                         err = -EFAULT;
1269                 return err;
1270         }
1271         case SHM_STAT:
1272         case SHM_STAT_ANY:
1273         case IPC_STAT: {
1274                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1275                 if (err < 0)
1276                         return err;
1277                 if (copy_shmid_to_user(buf, &sem64, version))
1278                         err = -EFAULT;
1279                 return err;
1280         }
1281         case IPC_SET:
1282                 if (copy_shmid_from_user(&sem64, buf, version))
1283                         return -EFAULT;
1284                 fallthrough;
1285         case IPC_RMID:
1286                 return shmctl_down(ns, shmid, cmd, &sem64);
1287         case SHM_LOCK:
1288         case SHM_UNLOCK:
1289                 return shmctl_do_lock(ns, shmid, cmd);
1290         default:
1291                 return -EINVAL;
1292         }
1293 }
1294
1295 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1296 {
1297         return ksys_shmctl(shmid, cmd, buf, IPC_64);
1298 }
1299
1300 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1301 long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1302 {
1303         int version = ipc_parse_version(&cmd);
1304
1305         return ksys_shmctl(shmid, cmd, buf, version);
1306 }
1307
1308 SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1309 {
1310         return ksys_old_shmctl(shmid, cmd, buf);
1311 }
1312 #endif
1313
1314 #ifdef CONFIG_COMPAT
1315
1316 struct compat_shmid_ds {
1317         struct compat_ipc_perm shm_perm;
1318         int shm_segsz;
1319         old_time32_t shm_atime;
1320         old_time32_t shm_dtime;
1321         old_time32_t shm_ctime;
1322         compat_ipc_pid_t shm_cpid;
1323         compat_ipc_pid_t shm_lpid;
1324         unsigned short shm_nattch;
1325         unsigned short shm_unused;
1326         compat_uptr_t shm_unused2;
1327         compat_uptr_t shm_unused3;
1328 };
1329
1330 struct compat_shminfo64 {
1331         compat_ulong_t shmmax;
1332         compat_ulong_t shmmin;
1333         compat_ulong_t shmmni;
1334         compat_ulong_t shmseg;
1335         compat_ulong_t shmall;
1336         compat_ulong_t __unused1;
1337         compat_ulong_t __unused2;
1338         compat_ulong_t __unused3;
1339         compat_ulong_t __unused4;
1340 };
1341
1342 struct compat_shm_info {
1343         compat_int_t used_ids;
1344         compat_ulong_t shm_tot, shm_rss, shm_swp;
1345         compat_ulong_t swap_attempts, swap_successes;
1346 };
1347
1348 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1349                                         int version)
1350 {
1351         if (in->shmmax > INT_MAX)
1352                 in->shmmax = INT_MAX;
1353         if (version == IPC_64) {
1354                 struct compat_shminfo64 info;
1355                 memset(&info, 0, sizeof(info));
1356                 info.shmmax = in->shmmax;
1357                 info.shmmin = in->shmmin;
1358                 info.shmmni = in->shmmni;
1359                 info.shmseg = in->shmseg;
1360                 info.shmall = in->shmall;
1361                 return copy_to_user(buf, &info, sizeof(info));
1362         } else {
1363                 struct shminfo info;
1364                 memset(&info, 0, sizeof(info));
1365                 info.shmmax = in->shmmax;
1366                 info.shmmin = in->shmmin;
1367                 info.shmmni = in->shmmni;
1368                 info.shmseg = in->shmseg;
1369                 info.shmall = in->shmall;
1370                 return copy_to_user(buf, &info, sizeof(info));
1371         }
1372 }
1373
1374 static int put_compat_shm_info(struct shm_info *ip,
1375                                 struct compat_shm_info __user *uip)
1376 {
1377         struct compat_shm_info info;
1378
1379         memset(&info, 0, sizeof(info));
1380         info.used_ids = ip->used_ids;
1381         info.shm_tot = ip->shm_tot;
1382         info.shm_rss = ip->shm_rss;
1383         info.shm_swp = ip->shm_swp;
1384         info.swap_attempts = ip->swap_attempts;
1385         info.swap_successes = ip->swap_successes;
1386         return copy_to_user(uip, &info, sizeof(info));
1387 }
1388
1389 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1390                                         int version)
1391 {
1392         if (version == IPC_64) {
1393                 struct compat_shmid64_ds v;
1394                 memset(&v, 0, sizeof(v));
1395                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1396                 v.shm_atime      = lower_32_bits(in->shm_atime);
1397                 v.shm_atime_high = upper_32_bits(in->shm_atime);
1398                 v.shm_dtime      = lower_32_bits(in->shm_dtime);
1399                 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1400                 v.shm_ctime      = lower_32_bits(in->shm_ctime);
1401                 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1402                 v.shm_segsz = in->shm_segsz;
1403                 v.shm_nattch = in->shm_nattch;
1404                 v.shm_cpid = in->shm_cpid;
1405                 v.shm_lpid = in->shm_lpid;
1406                 return copy_to_user(buf, &v, sizeof(v));
1407         } else {
1408                 struct compat_shmid_ds v;
1409                 memset(&v, 0, sizeof(v));
1410                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1411                 v.shm_perm.key = in->shm_perm.key;
1412                 v.shm_atime = in->shm_atime;
1413                 v.shm_dtime = in->shm_dtime;
1414                 v.shm_ctime = in->shm_ctime;
1415                 v.shm_segsz = in->shm_segsz;
1416                 v.shm_nattch = in->shm_nattch;
1417                 v.shm_cpid = in->shm_cpid;
1418                 v.shm_lpid = in->shm_lpid;
1419                 return copy_to_user(buf, &v, sizeof(v));
1420         }
1421 }
1422
1423 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1424                                         int version)
1425 {
1426         memset(out, 0, sizeof(*out));
1427         if (version == IPC_64) {
1428                 struct compat_shmid64_ds __user *p = buf;
1429                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1430         } else {
1431                 struct compat_shmid_ds __user *p = buf;
1432                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1433         }
1434 }
1435
1436 static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1437 {
1438         struct ipc_namespace *ns;
1439         struct shmid64_ds sem64;
1440         int err;
1441
1442         ns = current->nsproxy->ipc_ns;
1443
1444         if (cmd < 0 || shmid < 0)
1445                 return -EINVAL;
1446
1447         switch (cmd) {
1448         case IPC_INFO: {
1449                 struct shminfo64 shminfo;
1450                 err = shmctl_ipc_info(ns, &shminfo);
1451                 if (err < 0)
1452                         return err;
1453                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1454                         err = -EFAULT;
1455                 return err;
1456         }
1457         case SHM_INFO: {
1458                 struct shm_info shm_info;
1459                 err = shmctl_shm_info(ns, &shm_info);
1460                 if (err < 0)
1461                         return err;
1462                 if (put_compat_shm_info(&shm_info, uptr))
1463                         err = -EFAULT;
1464                 return err;
1465         }
1466         case IPC_STAT:
1467         case SHM_STAT_ANY:
1468         case SHM_STAT:
1469                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1470                 if (err < 0)
1471                         return err;
1472                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1473                         err = -EFAULT;
1474                 return err;
1475
1476         case IPC_SET:
1477                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1478                         return -EFAULT;
1479                 fallthrough;
1480         case IPC_RMID:
1481                 return shmctl_down(ns, shmid, cmd, &sem64);
1482         case SHM_LOCK:
1483         case SHM_UNLOCK:
1484                 return shmctl_do_lock(ns, shmid, cmd);
1485         default:
1486                 return -EINVAL;
1487         }
1488         return err;
1489 }
1490
1491 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1492 {
1493         return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1494 }
1495
1496 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1497 long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1498 {
1499         int version = compat_ipc_parse_version(&cmd);
1500
1501         return compat_ksys_shmctl(shmid, cmd, uptr, version);
1502 }
1503
1504 COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1505 {
1506         return compat_ksys_old_shmctl(shmid, cmd, uptr);
1507 }
1508 #endif
1509 #endif
1510
1511 /*
1512  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1513  *
1514  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1515  * "raddr" thing points to kernel space, and there has to be a wrapper around
1516  * this.
1517  */
1518 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1519               ulong *raddr, unsigned long shmlba)
1520 {
1521         struct shmid_kernel *shp;
1522         unsigned long addr = (unsigned long)shmaddr;
1523         unsigned long size;
1524         struct file *file, *base;
1525         int    err;
1526         unsigned long flags = MAP_SHARED;
1527         unsigned long prot;
1528         int acc_mode;
1529         struct ipc_namespace *ns;
1530         struct shm_file_data *sfd;
1531         int f_flags;
1532         unsigned long populate = 0;
1533
1534         err = -EINVAL;
1535         if (shmid < 0)
1536                 goto out;
1537
1538         if (addr) {
1539                 if (addr & (shmlba - 1)) {
1540                         if (shmflg & SHM_RND) {
1541                                 addr &= ~(shmlba - 1);  /* round down */
1542
1543                                 /*
1544                                  * Ensure that the round-down is non-nil
1545                                  * when remapping. This can happen for
1546                                  * cases when addr < shmlba.
1547                                  */
1548                                 if (!addr && (shmflg & SHM_REMAP))
1549                                         goto out;
1550                         } else
1551 #ifndef __ARCH_FORCE_SHMLBA
1552                                 if (addr & ~PAGE_MASK)
1553 #endif
1554                                         goto out;
1555                 }
1556
1557                 flags |= MAP_FIXED;
1558         } else if ((shmflg & SHM_REMAP))
1559                 goto out;
1560
1561         if (shmflg & SHM_RDONLY) {
1562                 prot = PROT_READ;
1563                 acc_mode = S_IRUGO;
1564                 f_flags = O_RDONLY;
1565         } else {
1566                 prot = PROT_READ | PROT_WRITE;
1567                 acc_mode = S_IRUGO | S_IWUGO;
1568                 f_flags = O_RDWR;
1569         }
1570         if (shmflg & SHM_EXEC) {
1571                 prot |= PROT_EXEC;
1572                 acc_mode |= S_IXUGO;
1573         }
1574
1575         /*
1576          * We cannot rely on the fs check since SYSV IPC does have an
1577          * additional creator id...
1578          */
1579         ns = current->nsproxy->ipc_ns;
1580         rcu_read_lock();
1581         shp = shm_obtain_object_check(ns, shmid);
1582         if (IS_ERR(shp)) {
1583                 err = PTR_ERR(shp);
1584                 goto out_unlock;
1585         }
1586
1587         err = -EACCES;
1588         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1589                 goto out_unlock;
1590
1591         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1592         if (err)
1593                 goto out_unlock;
1594
1595         ipc_lock_object(&shp->shm_perm);
1596
1597         /* check if shm_destroy() is tearing down shp */
1598         if (!ipc_valid_object(&shp->shm_perm)) {
1599                 ipc_unlock_object(&shp->shm_perm);
1600                 err = -EIDRM;
1601                 goto out_unlock;
1602         }
1603
1604         /*
1605          * We need to take a reference to the real shm file to prevent the
1606          * pointer from becoming stale in cases where the lifetime of the outer
1607          * file extends beyond that of the shm segment.  It's not usually
1608          * possible, but it can happen during remap_file_pages() emulation as
1609          * that unmaps the memory, then does ->mmap() via file reference only.
1610          * We'll deny the ->mmap() if the shm segment was since removed, but to
1611          * detect shm ID reuse we need to compare the file pointers.
1612          */
1613         base = get_file(shp->shm_file);
1614         shp->shm_nattch++;
1615         size = i_size_read(file_inode(base));
1616         ipc_unlock_object(&shp->shm_perm);
1617         rcu_read_unlock();
1618
1619         err = -ENOMEM;
1620         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1621         if (!sfd) {
1622                 fput(base);
1623                 goto out_nattch;
1624         }
1625
1626         file = alloc_file_clone(base, f_flags,
1627                           is_file_hugepages(base) ?
1628                                 &shm_file_operations_huge :
1629                                 &shm_file_operations);
1630         err = PTR_ERR(file);
1631         if (IS_ERR(file)) {
1632                 kfree(sfd);
1633                 fput(base);
1634                 goto out_nattch;
1635         }
1636
1637         sfd->id = shp->shm_perm.id;
1638         sfd->ns = get_ipc_ns(ns);
1639         sfd->file = base;
1640         sfd->vm_ops = NULL;
1641         file->private_data = sfd;
1642
1643         err = security_mmap_file(file, prot, flags);
1644         if (err)
1645                 goto out_fput;
1646
1647         if (mmap_write_lock_killable(current->mm)) {
1648                 err = -EINTR;
1649                 goto out_fput;
1650         }
1651
1652         if (addr && !(shmflg & SHM_REMAP)) {
1653                 err = -EINVAL;
1654                 if (addr + size < addr)
1655                         goto invalid;
1656
1657                 if (find_vma_intersection(current->mm, addr, addr + size))
1658                         goto invalid;
1659         }
1660
1661         addr = do_mmap(file, addr, size, prot, flags, 0, 0, &populate, NULL);
1662         *raddr = addr;
1663         err = 0;
1664         if (IS_ERR_VALUE(addr))
1665                 err = (long)addr;
1666 invalid:
1667         mmap_write_unlock(current->mm);
1668         if (populate)
1669                 mm_populate(addr, populate);
1670
1671 out_fput:
1672         fput(file);
1673
1674 out_nattch:
1675         down_write(&shm_ids(ns).rwsem);
1676         shp = shm_lock(ns, shmid);
1677         shp->shm_nattch--;
1678
1679         if (shm_may_destroy(shp))
1680                 shm_destroy(ns, shp);
1681         else
1682                 shm_unlock(shp);
1683         up_write(&shm_ids(ns).rwsem);
1684         return err;
1685
1686 out_unlock:
1687         rcu_read_unlock();
1688 out:
1689         return err;
1690 }
1691
1692 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1693 {
1694         unsigned long ret;
1695         long err;
1696
1697         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1698         if (err)
1699                 return err;
1700         force_successful_syscall_return();
1701         return (long)ret;
1702 }
1703
1704 #ifdef CONFIG_COMPAT
1705
1706 #ifndef COMPAT_SHMLBA
1707 #define COMPAT_SHMLBA   SHMLBA
1708 #endif
1709
1710 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1711 {
1712         unsigned long ret;
1713         long err;
1714
1715         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1716         if (err)
1717                 return err;
1718         force_successful_syscall_return();
1719         return (long)ret;
1720 }
1721 #endif
1722
1723 /*
1724  * detach and kill segment if marked destroyed.
1725  * The work is done in shm_close.
1726  */
1727 long ksys_shmdt(char __user *shmaddr)
1728 {
1729         struct mm_struct *mm = current->mm;
1730         struct vm_area_struct *vma;
1731         unsigned long addr = (unsigned long)shmaddr;
1732         int retval = -EINVAL;
1733 #ifdef CONFIG_MMU
1734         loff_t size = 0;
1735         struct file *file;
1736         VMA_ITERATOR(vmi, mm, addr);
1737 #endif
1738
1739         if (addr & ~PAGE_MASK)
1740                 return retval;
1741
1742         if (mmap_write_lock_killable(mm))
1743                 return -EINTR;
1744
1745         /*
1746          * This function tries to be smart and unmap shm segments that
1747          * were modified by partial mlock or munmap calls:
1748          * - It first determines the size of the shm segment that should be
1749          *   unmapped: It searches for a vma that is backed by shm and that
1750          *   started at address shmaddr. It records it's size and then unmaps
1751          *   it.
1752          * - Then it unmaps all shm vmas that started at shmaddr and that
1753          *   are within the initially determined size and that are from the
1754          *   same shm segment from which we determined the size.
1755          * Errors from do_munmap are ignored: the function only fails if
1756          * it's called with invalid parameters or if it's called to unmap
1757          * a part of a vma. Both calls in this function are for full vmas,
1758          * the parameters are directly copied from the vma itself and always
1759          * valid - therefore do_munmap cannot fail. (famous last words?)
1760          */
1761         /*
1762          * If it had been mremap()'d, the starting address would not
1763          * match the usual checks anyway. So assume all vma's are
1764          * above the starting address given.
1765          */
1766
1767 #ifdef CONFIG_MMU
1768         for_each_vma(vmi, vma) {
1769                 /*
1770                  * Check if the starting address would match, i.e. it's
1771                  * a fragment created by mprotect() and/or munmap(), or it
1772                  * otherwise it starts at this address with no hassles.
1773                  */
1774                 if ((vma->vm_ops == &shm_vm_ops) &&
1775                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1776
1777                         /*
1778                          * Record the file of the shm segment being
1779                          * unmapped.  With mremap(), someone could place
1780                          * page from another segment but with equal offsets
1781                          * in the range we are unmapping.
1782                          */
1783                         file = vma->vm_file;
1784                         size = i_size_read(file_inode(vma->vm_file));
1785                         do_vma_munmap(&vmi, vma, vma->vm_start, vma->vm_end,
1786                                       NULL, false);
1787                         /*
1788                          * We discovered the size of the shm segment, so
1789                          * break out of here and fall through to the next
1790                          * loop that uses the size information to stop
1791                          * searching for matching vma's.
1792                          */
1793                         retval = 0;
1794                         vma = vma_next(&vmi);
1795                         break;
1796                 }
1797         }
1798
1799         /*
1800          * We need look no further than the maximum address a fragment
1801          * could possibly have landed at. Also cast things to loff_t to
1802          * prevent overflows and make comparisons vs. equal-width types.
1803          */
1804         size = PAGE_ALIGN(size);
1805         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1806                 /* finding a matching vma now does not alter retval */
1807                 if ((vma->vm_ops == &shm_vm_ops) &&
1808                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1809                     (vma->vm_file == file)) {
1810                         do_vma_munmap(&vmi, vma, vma->vm_start, vma->vm_end,
1811                                       NULL, false);
1812                 }
1813
1814                 vma = vma_next(&vmi);
1815         }
1816
1817 #else   /* CONFIG_MMU */
1818         vma = vma_lookup(mm, addr);
1819         /* under NOMMU conditions, the exact address to be destroyed must be
1820          * given
1821          */
1822         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1823                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1824                 retval = 0;
1825         }
1826
1827 #endif
1828
1829         mmap_write_unlock(mm);
1830         return retval;
1831 }
1832
1833 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1834 {
1835         return ksys_shmdt(shmaddr);
1836 }
1837
1838 #ifdef CONFIG_PROC_FS
1839 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1840 {
1841         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1842         struct user_namespace *user_ns = seq_user_ns(s);
1843         struct kern_ipc_perm *ipcp = it;
1844         struct shmid_kernel *shp;
1845         unsigned long rss = 0, swp = 0;
1846
1847         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1848         shm_add_rss_swap(shp, &rss, &swp);
1849
1850 #if BITS_PER_LONG <= 32
1851 #define SIZE_SPEC "%10lu"
1852 #else
1853 #define SIZE_SPEC "%21lu"
1854 #endif
1855
1856         seq_printf(s,
1857                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1858                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1859                    SIZE_SPEC " " SIZE_SPEC "\n",
1860                    shp->shm_perm.key,
1861                    shp->shm_perm.id,
1862                    shp->shm_perm.mode,
1863                    shp->shm_segsz,
1864                    pid_nr_ns(shp->shm_cprid, pid_ns),
1865                    pid_nr_ns(shp->shm_lprid, pid_ns),
1866                    shp->shm_nattch,
1867                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1868                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1869                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1870                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1871                    shp->shm_atim,
1872                    shp->shm_dtim,
1873                    shp->shm_ctim,
1874                    rss * PAGE_SIZE,
1875                    swp * PAGE_SIZE);
1876
1877         return 0;
1878 }
1879 #endif