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#include <linux/nstree.h>
  49
  50#include <linux/uaccess.h>
  51
  52#include "util.h"
  53
  54struct shmid_kernel /* private to the kernel */
  55{
  56	struct kern_ipc_perm	shm_perm;
  57	struct file		*shm_file;
  58	unsigned long		shm_nattch;
  59	unsigned long		shm_segsz;
  60	time64_t		shm_atim;
  61	time64_t		shm_dtim;
  62	time64_t		shm_ctim;
  63	struct pid		*shm_cprid;
  64	struct pid		*shm_lprid;
  65	struct ucounts		*mlock_ucounts;
  66
  67	/*
  68	 * The task created the shm object, for
  69	 * task_lock(shp->shm_creator)
  70	 */
  71	struct task_struct	*shm_creator;
  72
  73	/*
  74	 * List by creator. task_lock(->shm_creator) required for read/write.
  75	 * If list_empty(), then the creator is dead already.
  76	 */
  77	struct list_head	shm_clist;
  78	struct ipc_namespace	*ns;
  79} __randomize_layout;
  80
  81/* shm_mode upper byte flags */
  82#define SHM_DEST	01000	/* segment will be destroyed on last detach */
  83#define SHM_LOCKED	02000   /* segment will not be swapped */
  84
  85struct shm_file_data {
  86	int id;
  87	struct ipc_namespace *ns;
  88	struct file *file;
  89	const struct vm_operations_struct *vm_ops;
  90};
  91
  92#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  93
  94static const struct file_operations shm_file_operations;
  95static const struct vm_operations_struct shm_vm_ops;
  96
  97#define shm_ids(ns)	((ns)->ids[IPC_SHM_IDS])
  98
  99#define shm_unlock(shp)			\
 100	ipc_unlock(&(shp)->shm_perm)
 101
 102static int newseg(struct ipc_namespace *, struct ipc_params *);
 103static void shm_open(struct vm_area_struct *vma);
 104static void shm_close(struct vm_area_struct *vma);
 105static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
 106#ifdef CONFIG_PROC_FS
 107static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
 108#endif
 109
 110void shm_init_ns(struct ipc_namespace *ns)
 111{
 112	ns->shm_ctlmax = SHMMAX;
 113	ns->shm_ctlall = SHMALL;
 114	ns->shm_ctlmni = SHMMNI;
 115	ns->shm_rmid_forced = 0;
 116	ns->shm_tot = 0;
 117	ipc_init_ids(&shm_ids(ns));
 118}
 119
 120/*
 121 * Called with shm_ids.rwsem (writer) and the shp structure locked.
 122 * Only shm_ids.rwsem remains locked on exit.
 123 */
 124static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 125{
 126	struct shmid_kernel *shp;
 127
 128	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 129	WARN_ON(ns != shp->ns);
 130
 131	if (shp->shm_nattch) {
 132		shp->shm_perm.mode |= SHM_DEST;
 133		/* Do not find it any more */
 134		ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
 135		shm_unlock(shp);
 136	} else
 137		shm_destroy(ns, shp);
 138}
 139
 140#ifdef CONFIG_IPC_NS
 141void shm_exit_ns(struct ipc_namespace *ns)
 142{
 143	free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 144	idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 145	rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
 146}
 147#endif
 148
 149static int __init ipc_ns_init(void)
 150{
 151	shm_init_ns(&init_ipc_ns);
 152	ns_tree_add(&init_ipc_ns);
 153	return 0;
 154}
 155
 156pure_initcall(ipc_ns_init);
 157
 158void __init shm_init(void)
 159{
 160	ipc_init_proc_interface("sysvipc/shm",
 161#if BITS_PER_LONG <= 32
 162				"       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 163#else
 164				"       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 165#endif
 166				IPC_SHM_IDS, sysvipc_shm_proc_show);
 167}
 168
 169static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 170{
 171	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 172
 173	if (IS_ERR(ipcp))
 174		return ERR_CAST(ipcp);
 175
 176	return container_of(ipcp, struct shmid_kernel, shm_perm);
 177}
 178
 179static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 180{
 181	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 182
 183	if (IS_ERR(ipcp))
 184		return ERR_CAST(ipcp);
 185
 186	return container_of(ipcp, struct shmid_kernel, shm_perm);
 187}
 188
 189/*
 190 * shm_lock_(check_) routines are called in the paths where the rwsem
 191 * is not necessarily held.
 192 */
 193static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 194{
 195	struct kern_ipc_perm *ipcp;
 196
 197	rcu_read_lock();
 198	ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 199	if (IS_ERR(ipcp))
 200		goto err;
 201
 202	ipc_lock_object(ipcp);
 203	/*
 204	 * ipc_rmid() may have already freed the ID while ipc_lock_object()
 205	 * was spinning: here verify that the structure is still valid.
 206	 * Upon races with RMID, return -EIDRM, thus indicating that
 207	 * the ID points to a removed identifier.
 208	 */
 209	if (ipc_valid_object(ipcp)) {
 210		/* return a locked ipc object upon success */
 211		return container_of(ipcp, struct shmid_kernel, shm_perm);
 212	}
 213
 214	ipc_unlock_object(ipcp);
 215	ipcp = ERR_PTR(-EIDRM);
 216err:
 217	rcu_read_unlock();
 218	/*
 219	 * Callers of shm_lock() must validate the status of the returned ipc
 220	 * object pointer and error out as appropriate.
 221	 */
 222	return ERR_CAST(ipcp);
 223}
 224
 225static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 226{
 227	rcu_read_lock();
 228	ipc_lock_object(&ipcp->shm_perm);
 229}
 230
 231static void shm_rcu_free(struct rcu_head *head)
 232{
 233	struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 234							rcu);
 235	struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 236							shm_perm);
 237	security_shm_free(&shp->shm_perm);
 238	kfree(shp);
 239}
 240
 241/*
 242 * It has to be called with shp locked.
 243 * It must be called before ipc_rmid()
 244 */
 245static inline void shm_clist_rm(struct shmid_kernel *shp)
 246{
 247	struct task_struct *creator;
 248
 249	/* ensure that shm_creator does not disappear */
 250	rcu_read_lock();
 251
 252	/*
 253	 * A concurrent exit_shm may do a list_del_init() as well.
 254	 * Just do nothing if exit_shm already did the work
 255	 */
 256	if (!list_empty(&shp->shm_clist)) {
 257		/*
 258		 * shp->shm_creator is guaranteed to be valid *only*
 259		 * if shp->shm_clist is not empty.
 260		 */
 261		creator = shp->shm_creator;
 262
 263		task_lock(creator);
 264		/*
 265		 * list_del_init() is a nop if the entry was already removed
 266		 * from the list.
 267		 */
 268		list_del_init(&shp->shm_clist);
 269		task_unlock(creator);
 270	}
 271	rcu_read_unlock();
 272}
 273
 274static inline void shm_rmid(struct shmid_kernel *s)
 275{
 276	shm_clist_rm(s);
 277	ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
 278}
 279
 280
 281static int __shm_open(struct shm_file_data *sfd)
 282{
 283	struct shmid_kernel *shp;
 284
 285	shp = shm_lock(sfd->ns, sfd->id);
 286
 287	if (IS_ERR(shp))
 288		return PTR_ERR(shp);
 289
 290	if (shp->shm_file != sfd->file) {
 291		/* ID was reused */
 292		shm_unlock(shp);
 293		return -EINVAL;
 294	}
 295
 296	shp->shm_atim = ktime_get_real_seconds();
 297	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 298	shp->shm_nattch++;
 299	shm_unlock(shp);
 300	return 0;
 301}
 302
 303/* This is called by fork, once for every shm attach. */
 304static void shm_open(struct vm_area_struct *vma)
 305{
 306	struct file *file = vma->vm_file;
 307	struct shm_file_data *sfd = shm_file_data(file);
 308	int err;
 309
 310	/* Always call underlying open if present */
 311	if (sfd->vm_ops->open)
 312		sfd->vm_ops->open(vma);
 313
 314	err = __shm_open(sfd);
 315	/*
 316	 * We raced in the idr lookup or with shm_destroy().
 317	 * Either way, the ID is busted.
 318	 */
 319	WARN_ON_ONCE(err);
 320}
 321
 322/*
 323 * shm_destroy - free the struct shmid_kernel
 324 *
 325 * @ns: namespace
 326 * @shp: struct to free
 327 *
 328 * It has to be called with shp and shm_ids.rwsem (writer) locked,
 329 * but returns with shp unlocked and freed.
 330 */
 331static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 332{
 333	struct file *shm_file;
 334
 335	shm_file = shp->shm_file;
 336	shp->shm_file = NULL;
 337	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 338	shm_rmid(shp);
 339	shm_unlock(shp);
 340	if (!is_file_hugepages(shm_file))
 341		shmem_lock(shm_file, 0, shp->mlock_ucounts);
 342	fput(shm_file);
 343	ipc_update_pid(&shp->shm_cprid, NULL);
 344	ipc_update_pid(&shp->shm_lprid, NULL);
 345	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 346}
 347
 348/*
 349 * shm_may_destroy - identifies whether shm segment should be destroyed now
 350 *
 351 * Returns true if and only if there are no active users of the segment and
 352 * one of the following is true:
 353 *
 354 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 355 *
 356 * 2) sysctl kernel.shm_rmid_forced is set to 1.
 357 */
 358static bool shm_may_destroy(struct shmid_kernel *shp)
 359{
 360	return (shp->shm_nattch == 0) &&
 361	       (shp->ns->shm_rmid_forced ||
 362		(shp->shm_perm.mode & SHM_DEST));
 363}
 364
 365/*
 366 * remove the attach descriptor vma.
 367 * free memory for segment if it is marked destroyed.
 368 * The descriptor has already been removed from the current->mm->mmap list
 369 * and will later be kfree()d.
 370 */
 371static void __shm_close(struct shm_file_data *sfd)
 372{
 373	struct shmid_kernel *shp;
 374	struct ipc_namespace *ns = sfd->ns;
 375
 376	down_write(&shm_ids(ns).rwsem);
 377	/* remove from the list of attaches of the shm segment */
 378	shp = shm_lock(ns, sfd->id);
 379
 380	/*
 381	 * We raced in the idr lookup or with shm_destroy().
 382	 * Either way, the ID is busted.
 383	 */
 384	if (WARN_ON_ONCE(IS_ERR(shp)))
 385		goto done; /* no-op */
 386
 387	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 388	shp->shm_dtim = ktime_get_real_seconds();
 389	shp->shm_nattch--;
 390	if (shm_may_destroy(shp))
 391		shm_destroy(ns, shp);
 392	else
 393		shm_unlock(shp);
 394done:
 395	up_write(&shm_ids(ns).rwsem);
 396}
 397
 398static void shm_close(struct vm_area_struct *vma)
 399{
 400	struct file *file = vma->vm_file;
 401	struct shm_file_data *sfd = shm_file_data(file);
 402
 403	/* Always call underlying close if present */
 404	if (sfd->vm_ops->close)
 405		sfd->vm_ops->close(vma);
 406
 407	__shm_close(sfd);
 408}
 409
 410/* Called with ns->shm_ids(ns).rwsem locked */
 411static int shm_try_destroy_orphaned(int id, void *p, void *data)
 412{
 413	struct ipc_namespace *ns = data;
 414	struct kern_ipc_perm *ipcp = p;
 415	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 416
 417	/*
 418	 * We want to destroy segments without users and with already
 419	 * exit'ed originating process.
 420	 *
 421	 * As shp->* are changed under rwsem, it's safe to skip shp locking.
 422	 */
 423	if (!list_empty(&shp->shm_clist))
 424		return 0;
 425
 426	if (shm_may_destroy(shp)) {
 427		shm_lock_by_ptr(shp);
 428		shm_destroy(ns, shp);
 429	}
 430	return 0;
 431}
 432
 433void shm_destroy_orphaned(struct ipc_namespace *ns)
 434{
 435	down_write(&shm_ids(ns).rwsem);
 436	if (shm_ids(ns).in_use) {
 437		rcu_read_lock();
 438		idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 439		rcu_read_unlock();
 440	}
 441	up_write(&shm_ids(ns).rwsem);
 442}
 443
 444/* Locking assumes this will only be called with task == current */
 445void exit_shm(struct task_struct *task)
 446{
 447	for (;;) {
 448		struct shmid_kernel *shp;
 449		struct ipc_namespace *ns;
 450
 451		task_lock(task);
 452
 453		if (list_empty(&task->sysvshm.shm_clist)) {
 454			task_unlock(task);
 455			break;
 456		}
 457
 458		shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
 459				shm_clist);
 460
 461		/*
 462		 * 1) Get pointer to the ipc namespace. It is worth to say
 463		 * that this pointer is guaranteed to be valid because
 464		 * shp lifetime is always shorter than namespace lifetime
 465		 * in which shp lives.
 466		 * We taken task_lock it means that shp won't be freed.
 467		 */
 468		ns = shp->ns;
 469
 470		/*
 471		 * 2) If kernel.shm_rmid_forced is not set then only keep track of
 472		 * which shmids are orphaned, so that a later set of the sysctl
 473		 * can clean them up.
 474		 */
 475		if (!ns->shm_rmid_forced)
 476			goto unlink_continue;
 477
 478		/*
 479		 * 3) get a reference to the namespace.
 480		 *    The refcount could be already 0. If it is 0, then
 481		 *    the shm objects will be free by free_ipc_work().
 482		 */
 483		ns = get_ipc_ns_not_zero(ns);
 484		if (!ns) {
 485unlink_continue:
 486			list_del_init(&shp->shm_clist);
 487			task_unlock(task);
 488			continue;
 489		}
 490
 491		/*
 492		 * 4) get a reference to shp.
 493		 *   This cannot fail: shm_clist_rm() is called before
 494		 *   ipc_rmid(), thus the refcount cannot be 0.
 495		 */
 496		WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
 497
 498		/*
 499		 * 5) unlink the shm segment from the list of segments
 500		 *    created by current.
 501		 *    This must be done last. After unlinking,
 502		 *    only the refcounts obtained above prevent IPC_RMID
 503		 *    from destroying the segment or the namespace.
 504		 */
 505		list_del_init(&shp->shm_clist);
 506
 507		task_unlock(task);
 508
 509		/*
 510		 * 6) we have all references
 511		 *    Thus lock & if needed destroy shp.
 512		 */
 513		down_write(&shm_ids(ns).rwsem);
 514		shm_lock_by_ptr(shp);
 515		/*
 516		 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
 517		 * safe to call ipc_rcu_putref here
 518		 */
 519		ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 520
 521		if (ipc_valid_object(&shp->shm_perm)) {
 522			if (shm_may_destroy(shp))
 523				shm_destroy(ns, shp);
 524			else
 525				shm_unlock(shp);
 526		} else {
 527			/*
 528			 * Someone else deleted the shp from namespace
 529			 * idr/kht while we have waited.
 530			 * Just unlock and continue.
 531			 */
 532			shm_unlock(shp);
 533		}
 534
 535		up_write(&shm_ids(ns).rwsem);
 536		put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
 537	}
 538}
 539
 540static vm_fault_t shm_fault(struct vm_fault *vmf)
 541{
 542	struct file *file = vmf->vma->vm_file;
 543	struct shm_file_data *sfd = shm_file_data(file);
 544
 545	return sfd->vm_ops->fault(vmf);
 546}
 547
 548static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
 549{
 550	struct file *file = vma->vm_file;
 551	struct shm_file_data *sfd = shm_file_data(file);
 552
 553	if (sfd->vm_ops->may_split)
 554		return sfd->vm_ops->may_split(vma, addr);
 555
 556	return 0;
 557}
 558
 559static unsigned long shm_pagesize(struct vm_area_struct *vma)
 560{
 561	struct file *file = vma->vm_file;
 562	struct shm_file_data *sfd = shm_file_data(file);
 563
 564	if (sfd->vm_ops->pagesize)
 565		return sfd->vm_ops->pagesize(vma);
 566
 567	return PAGE_SIZE;
 568}
 569
 570#ifdef CONFIG_NUMA
 571static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
 572{
 573	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
 574	int err = 0;
 575
 576	if (sfd->vm_ops->set_policy)
 577		err = sfd->vm_ops->set_policy(vma, mpol);
 578	return err;
 579}
 580
 581static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 582					unsigned long addr, pgoff_t *ilx)
 583{
 584	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
 585	struct mempolicy *mpol = vma->vm_policy;
 586
 587	if (sfd->vm_ops->get_policy)
 588		mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
 589	return mpol;
 590}
 591#endif
 592
 593static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 594{
 595	struct shm_file_data *sfd = shm_file_data(file);
 596	int ret;
 597
 598	/*
 599	 * In case of remap_file_pages() emulation, the file can represent an
 600	 * IPC ID that was removed, and possibly even reused by another shm
 601	 * segment already.  Propagate this case as an error to caller.
 602	 */
 603	ret = __shm_open(sfd);
 604	if (ret)
 605		return ret;
 606
 607	ret = vfs_mmap(sfd->file, vma);
 608	if (ret) {
 609		__shm_close(sfd);
 610		return ret;
 611	}
 612	sfd->vm_ops = vma->vm_ops;
 613#ifdef CONFIG_MMU
 614	WARN_ON(!sfd->vm_ops->fault);
 615#endif
 616	vma->vm_ops = &shm_vm_ops;
 617	return 0;
 618}
 619
 620static int shm_release(struct inode *ino, struct file *file)
 621{
 622	struct shm_file_data *sfd = shm_file_data(file);
 623
 624	put_ipc_ns(sfd->ns);
 625	fput(sfd->file);
 626	shm_file_data(file) = NULL;
 627	kfree(sfd);
 628	return 0;
 629}
 630
 631static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 632{
 633	struct shm_file_data *sfd = shm_file_data(file);
 634
 635	if (!sfd->file->f_op->fsync)
 636		return -EINVAL;
 637	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 638}
 639
 640static long shm_fallocate(struct file *file, int mode, loff_t offset,
 641			  loff_t len)
 642{
 643	struct shm_file_data *sfd = shm_file_data(file);
 644
 645	if (!sfd->file->f_op->fallocate)
 646		return -EOPNOTSUPP;
 647	return sfd->file->f_op->fallocate(file, mode, offset, len);
 648}
 649
 650static unsigned long shm_get_unmapped_area(struct file *file,
 651	unsigned long addr, unsigned long len, unsigned long pgoff,
 652	unsigned long flags)
 653{
 654	struct shm_file_data *sfd = shm_file_data(file);
 655
 656	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 657						pgoff, flags);
 658}
 659
 660static const struct file_operations shm_file_operations = {
 661	.mmap		= shm_mmap,
 662	.fsync		= shm_fsync,
 663	.release	= shm_release,
 664	.get_unmapped_area	= shm_get_unmapped_area,
 665	.llseek		= noop_llseek,
 666	.fallocate	= shm_fallocate,
 667};
 668
 669/*
 670 * shm_file_operations_huge is now identical to shm_file_operations
 671 * except for fop_flags
 672 */
 673static const struct file_operations shm_file_operations_huge = {
 674	.mmap		= shm_mmap,
 675	.fsync		= shm_fsync,
 676	.release	= shm_release,
 677	.get_unmapped_area	= shm_get_unmapped_area,
 678	.llseek		= noop_llseek,
 679	.fallocate	= shm_fallocate,
 680	.fop_flags	= FOP_HUGE_PAGES,
 681};
 682
 683static const struct vm_operations_struct shm_vm_ops = {
 684	.open	= shm_open,	/* callback for a new vm-area open */
 685	.close	= shm_close,	/* callback for when the vm-area is released */
 686	.fault	= shm_fault,
 687	.may_split = shm_may_split,
 688	.pagesize = shm_pagesize,
 689#if defined(CONFIG_NUMA)
 690	.set_policy = shm_set_policy,
 691	.get_policy = shm_get_policy,
 692#endif
 693};
 694
 695/**
 696 * newseg - Create a new shared memory segment
 697 * @ns: namespace
 698 * @params: ptr to the structure that contains key, size and shmflg
 699 *
 700 * Called with shm_ids.rwsem held as a writer.
 701 */
 702static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 703{
 704	key_t key = params->key;
 705	int shmflg = params->flg;
 706	size_t size = params->u.size;
 707	int error;
 708	struct shmid_kernel *shp;
 709	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 710	struct file *file;
 711	char name[13];
 712	vm_flags_t acctflag = 0;
 713
 714	if (size < SHMMIN || size > ns->shm_ctlmax)
 715		return -EINVAL;
 716
 717	if (numpages << PAGE_SHIFT < size)
 718		return -ENOSPC;
 719
 720	if (ns->shm_tot + numpages < ns->shm_tot ||
 721			ns->shm_tot + numpages > ns->shm_ctlall)
 722		return -ENOSPC;
 723
 724	shp = kmalloc(sizeof(*shp), GFP_KERNEL_ACCOUNT);
 725	if (unlikely(!shp))
 726		return -ENOMEM;
 727
 728	shp->shm_perm.key = key;
 729	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 730	shp->mlock_ucounts = NULL;
 731
 732	shp->shm_perm.security = NULL;
 733	error = security_shm_alloc(&shp->shm_perm);
 734	if (error) {
 735		kfree(shp);
 736		return error;
 737	}
 738
 739	sprintf(name, "SYSV%08x", key);
 740	if (shmflg & SHM_HUGETLB) {
 741		struct hstate *hs;
 742		size_t hugesize;
 743
 744		hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 745		if (!hs) {
 746			error = -EINVAL;
 747			goto no_file;
 748		}
 749		hugesize = ALIGN(size, huge_page_size(hs));
 750
 751		/* hugetlb_file_setup applies strict accounting */
 752		if (shmflg & SHM_NORESERVE)
 753			acctflag = VM_NORESERVE;
 754		file = hugetlb_file_setup(name, hugesize, acctflag,
 755				HUGETLB_SHMFS_INODE, (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 756	} else {
 757		/*
 758		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 759		 * if it's asked for.
 760		 */
 761		if  ((shmflg & SHM_NORESERVE) &&
 762				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 763			acctflag = VM_NORESERVE;
 764		file = shmem_kernel_file_setup(name, size, acctflag);
 765	}
 766	error = PTR_ERR(file);
 767	if (IS_ERR(file))
 768		goto no_file;
 769
 770	shp->shm_cprid = get_pid(task_tgid(current));
 771	shp->shm_lprid = NULL;
 772	shp->shm_atim = shp->shm_dtim = 0;
 773	shp->shm_ctim = ktime_get_real_seconds();
 774	shp->shm_segsz = size;
 775	shp->shm_nattch = 0;
 776	shp->shm_file = file;
 777	shp->shm_creator = current;
 778
 779	/* ipc_addid() locks shp upon success. */
 780	error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 781	if (error < 0)
 782		goto no_id;
 783
 784	shp->ns = ns;
 785
 786	task_lock(current);
 787	list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 788	task_unlock(current);
 789
 790	/*
 791	 * shmid gets reported as "inode#" in /proc/pid/maps.
 792	 * proc-ps tools use this. Changing this will break them.
 793	 */
 794	file_inode(file)->i_ino = shp->shm_perm.id;
 795
 796	ns->shm_tot += numpages;
 797	error = shp->shm_perm.id;
 798
 799	ipc_unlock_object(&shp->shm_perm);
 800	rcu_read_unlock();
 801	return error;
 802
 803no_id:
 804	ipc_update_pid(&shp->shm_cprid, NULL);
 805	ipc_update_pid(&shp->shm_lprid, NULL);
 806	fput(file);
 807	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 808	return error;
 809no_file:
 810	call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 811	return error;
 812}
 813
 814/*
 815 * Called with shm_ids.rwsem and ipcp locked.
 816 */
 817static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
 818{
 819	struct shmid_kernel *shp;
 820
 821	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 822	if (shp->shm_segsz < params->u.size)
 823		return -EINVAL;
 824
 825	return 0;
 826}
 827
 828long ksys_shmget(key_t key, size_t size, int shmflg)
 829{
 830	struct ipc_namespace *ns;
 831	static const struct ipc_ops shm_ops = {
 832		.getnew = newseg,
 833		.associate = security_shm_associate,
 834		.more_checks = shm_more_checks,
 835	};
 836	struct ipc_params shm_params;
 837
 838	ns = current->nsproxy->ipc_ns;
 839
 840	shm_params.key = key;
 841	shm_params.flg = shmflg;
 842	shm_params.u.size = size;
 843
 844	return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 845}
 846
 847SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 848{
 849	return ksys_shmget(key, size, shmflg);
 850}
 851
 852static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 853{
 854	switch (version) {
 855	case IPC_64:
 856		return copy_to_user(buf, in, sizeof(*in));
 857	case IPC_OLD:
 858	    {
 859		struct shmid_ds out;
 860
 861		memset(&out, 0, sizeof(out));
 862		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 863		out.shm_segsz	= in->shm_segsz;
 864		out.shm_atime	= in->shm_atime;
 865		out.shm_dtime	= in->shm_dtime;
 866		out.shm_ctime	= in->shm_ctime;
 867		out.shm_cpid	= in->shm_cpid;
 868		out.shm_lpid	= in->shm_lpid;
 869		out.shm_nattch	= in->shm_nattch;
 870
 871		return copy_to_user(buf, &out, sizeof(out));
 872	    }
 873	default:
 874		return -EINVAL;
 875	}
 876}
 877
 878static inline unsigned long
 879copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 880{
 881	switch (version) {
 882	case IPC_64:
 883		if (copy_from_user(out, buf, sizeof(*out)))
 884			return -EFAULT;
 885		return 0;
 886	case IPC_OLD:
 887	    {
 888		struct shmid_ds tbuf_old;
 889
 890		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 891			return -EFAULT;
 892
 893		out->shm_perm.uid	= tbuf_old.shm_perm.uid;
 894		out->shm_perm.gid	= tbuf_old.shm_perm.gid;
 895		out->shm_perm.mode	= tbuf_old.shm_perm.mode;
 896
 897		return 0;
 898	    }
 899	default:
 900		return -EINVAL;
 901	}
 902}
 903
 904static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 905{
 906	switch (version) {
 907	case IPC_64:
 908		return copy_to_user(buf, in, sizeof(*in));
 909	case IPC_OLD:
 910	    {
 911		struct shminfo out;
 912
 913		if (in->shmmax > INT_MAX)
 914			out.shmmax = INT_MAX;
 915		else
 916			out.shmmax = (int)in->shmmax;
 917
 918		out.shmmin	= in->shmmin;
 919		out.shmmni	= in->shmmni;
 920		out.shmseg	= in->shmseg;
 921		out.shmall	= in->shmall;
 922
 923		return copy_to_user(buf, &out, sizeof(out));
 924	    }
 925	default:
 926		return -EINVAL;
 927	}
 928}
 929
 930/*
 931 * Calculate and add used RSS and swap pages of a shm.
 932 * Called with shm_ids.rwsem held as a reader
 933 */
 934static void shm_add_rss_swap(struct shmid_kernel *shp,
 935	unsigned long *rss_add, unsigned long *swp_add)
 936{
 937	struct inode *inode;
 938
 939	inode = file_inode(shp->shm_file);
 940
 941	if (is_file_hugepages(shp->shm_file)) {
 942		struct address_space *mapping = inode->i_mapping;
 943		struct hstate *h = hstate_file(shp->shm_file);
 944		*rss_add += pages_per_huge_page(h) * mapping->nrpages;
 945	} else {
 946#ifdef CONFIG_SHMEM
 947		struct shmem_inode_info *info = SHMEM_I(inode);
 948
 949		spin_lock_irq(&info->lock);
 950		*rss_add += inode->i_mapping->nrpages;
 951		*swp_add += info->swapped;
 952		spin_unlock_irq(&info->lock);
 953#else
 954		*rss_add += inode->i_mapping->nrpages;
 955#endif
 956	}
 957}
 958
 959/*
 960 * Called with shm_ids.rwsem held as a reader
 961 */
 962static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 963		unsigned long *swp)
 964{
 965	int next_id;
 966	int total, in_use;
 967
 968	*rss = 0;
 969	*swp = 0;
 970
 971	in_use = shm_ids(ns).in_use;
 972
 973	for (total = 0, next_id = 0; total < in_use; next_id++) {
 974		struct kern_ipc_perm *ipc;
 975		struct shmid_kernel *shp;
 976
 977		ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 978		if (ipc == NULL)
 979			continue;
 980		shp = container_of(ipc, struct shmid_kernel, shm_perm);
 981
 982		shm_add_rss_swap(shp, rss, swp);
 983
 984		total++;
 985	}
 986}
 987
 988/*
 989 * This function handles some shmctl commands which require the rwsem
 990 * to be held in write mode.
 991 * NOTE: no locks must be held, the rwsem is taken inside this function.
 992 */
 993static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 994		       struct shmid64_ds *shmid64)
 995{
 996	struct kern_ipc_perm *ipcp;
 997	struct shmid_kernel *shp;
 998	int err;
 999
1000	down_write(&shm_ids(ns).rwsem);
1001	rcu_read_lock();
1002
1003	ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
1004				      &shmid64->shm_perm, 0);
1005	if (IS_ERR(ipcp)) {
1006		err = PTR_ERR(ipcp);
1007		goto out_unlock1;
1008	}
1009
1010	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1011
1012	err = security_shm_shmctl(&shp->shm_perm, cmd);
1013	if (err)
1014		goto out_unlock1;
1015
1016	switch (cmd) {
1017	case IPC_RMID:
1018		ipc_lock_object(&shp->shm_perm);
1019		/* do_shm_rmid unlocks the ipc object and rcu */
1020		do_shm_rmid(ns, ipcp);
1021		goto out_up;
1022	case IPC_SET:
1023		ipc_lock_object(&shp->shm_perm);
1024		err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1025		if (err)
1026			goto out_unlock0;
1027		shp->shm_ctim = ktime_get_real_seconds();
1028		break;
1029	default:
1030		err = -EINVAL;
1031		goto out_unlock1;
1032	}
1033
1034out_unlock0:
1035	ipc_unlock_object(&shp->shm_perm);
1036out_unlock1:
1037	rcu_read_unlock();
1038out_up:
1039	up_write(&shm_ids(ns).rwsem);
1040	return err;
1041}
1042
1043static int shmctl_ipc_info(struct ipc_namespace *ns,
1044			   struct shminfo64 *shminfo)
1045{
1046	int err = security_shm_shmctl(NULL, IPC_INFO);
1047	if (!err) {
1048		memset(shminfo, 0, sizeof(*shminfo));
1049		shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1050		shminfo->shmmax = ns->shm_ctlmax;
1051		shminfo->shmall = ns->shm_ctlall;
1052		shminfo->shmmin = SHMMIN;
1053		down_read(&shm_ids(ns).rwsem);
1054		err = ipc_get_maxidx(&shm_ids(ns));
1055		up_read(&shm_ids(ns).rwsem);
1056		if (err < 0)
1057			err = 0;
1058	}
1059	return err;
1060}
1061
1062static int shmctl_shm_info(struct ipc_namespace *ns,
1063			   struct shm_info *shm_info)
1064{
1065	int err = security_shm_shmctl(NULL, SHM_INFO);
1066	if (!err) {
1067		memset(shm_info, 0, sizeof(*shm_info));
1068		down_read(&shm_ids(ns).rwsem);
1069		shm_info->used_ids = shm_ids(ns).in_use;
1070		shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1071		shm_info->shm_tot = ns->shm_tot;
1072		shm_info->swap_attempts = 0;
1073		shm_info->swap_successes = 0;
1074		err = ipc_get_maxidx(&shm_ids(ns));
1075		up_read(&shm_ids(ns).rwsem);
1076		if (err < 0)
1077			err = 0;
1078	}
1079	return err;
1080}
1081
1082static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1083			int cmd, struct shmid64_ds *tbuf)
1084{
1085	struct shmid_kernel *shp;
1086	int err;
1087
1088	memset(tbuf, 0, sizeof(*tbuf));
1089
1090	rcu_read_lock();
1091	if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1092		shp = shm_obtain_object(ns, shmid);
1093		if (IS_ERR(shp)) {
1094			err = PTR_ERR(shp);
1095			goto out_unlock;
1096		}
1097	} else { /* IPC_STAT */
1098		shp = shm_obtain_object_check(ns, shmid);
1099		if (IS_ERR(shp)) {
1100			err = PTR_ERR(shp);
1101			goto out_unlock;
1102		}
1103	}
1104
1105	/*
1106	 * Semantically SHM_STAT_ANY ought to be identical to
1107	 * that functionality provided by the /proc/sysvipc/
1108	 * interface. As such, only audit these calls and
1109	 * do not do traditional S_IRUGO permission checks on
1110	 * the ipc object.
1111	 */
1112	if (cmd == SHM_STAT_ANY)
1113		audit_ipc_obj(&shp->shm_perm);
1114	else {
1115		err = -EACCES;
1116		if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1117			goto out_unlock;
1118	}
1119
1120	err = security_shm_shmctl(&shp->shm_perm, cmd);
1121	if (err)
1122		goto out_unlock;
1123
1124	ipc_lock_object(&shp->shm_perm);
1125
1126	if (!ipc_valid_object(&shp->shm_perm)) {
1127		ipc_unlock_object(&shp->shm_perm);
1128		err = -EIDRM;
1129		goto out_unlock;
1130	}
1131
1132	kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1133	tbuf->shm_segsz	= shp->shm_segsz;
1134	tbuf->shm_atime	= shp->shm_atim;
1135	tbuf->shm_dtime	= shp->shm_dtim;
1136	tbuf->shm_ctime	= shp->shm_ctim;
1137#ifndef CONFIG_64BIT
1138	tbuf->shm_atime_high = shp->shm_atim >> 32;
1139	tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1140	tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1141#endif
1142	tbuf->shm_cpid	= pid_vnr(shp->shm_cprid);
1143	tbuf->shm_lpid	= pid_vnr(shp->shm_lprid);
1144	tbuf->shm_nattch = shp->shm_nattch;
1145
1146	if (cmd == IPC_STAT) {
1147		/*
1148		 * As defined in SUS:
1149		 * Return 0 on success
1150		 */
1151		err = 0;
1152	} else {
1153		/*
1154		 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1155		 * Return the full id, including the sequence number
1156		 */
1157		err = shp->shm_perm.id;
1158	}
1159
1160	ipc_unlock_object(&shp->shm_perm);
1161out_unlock:
1162	rcu_read_unlock();
1163	return err;
1164}
1165
1166static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1167{
1168	struct shmid_kernel *shp;
1169	struct file *shm_file;
1170	int err;
1171
1172	rcu_read_lock();
1173	shp = shm_obtain_object_check(ns, shmid);
1174	if (IS_ERR(shp)) {
1175		err = PTR_ERR(shp);
1176		goto out_unlock1;
1177	}
1178
1179	audit_ipc_obj(&(shp->shm_perm));
1180	err = security_shm_shmctl(&shp->shm_perm, cmd);
1181	if (err)
1182		goto out_unlock1;
1183
1184	ipc_lock_object(&shp->shm_perm);
1185
1186	/* check if shm_destroy() is tearing down shp */
1187	if (!ipc_valid_object(&shp->shm_perm)) {
1188		err = -EIDRM;
1189		goto out_unlock0;
1190	}
1191
1192	if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1193		kuid_t euid = current_euid();
1194
1195		if (!uid_eq(euid, shp->shm_perm.uid) &&
1196		    !uid_eq(euid, shp->shm_perm.cuid)) {
1197			err = -EPERM;
1198			goto out_unlock0;
1199		}
1200		if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1201			err = -EPERM;
1202			goto out_unlock0;
1203		}
1204	}
1205
1206	shm_file = shp->shm_file;
1207	if (is_file_hugepages(shm_file))
1208		goto out_unlock0;
1209
1210	if (cmd == SHM_LOCK) {
1211		struct ucounts *ucounts = current_ucounts();
1212
1213		err = shmem_lock(shm_file, 1, ucounts);
1214		if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1215			shp->shm_perm.mode |= SHM_LOCKED;
1216			shp->mlock_ucounts = ucounts;
1217		}
1218		goto out_unlock0;
1219	}
1220
1221	/* SHM_UNLOCK */
1222	if (!(shp->shm_perm.mode & SHM_LOCKED))
1223		goto out_unlock0;
1224	shmem_lock(shm_file, 0, shp->mlock_ucounts);
1225	shp->shm_perm.mode &= ~SHM_LOCKED;
1226	shp->mlock_ucounts = NULL;
1227	get_file(shm_file);
1228	ipc_unlock_object(&shp->shm_perm);
1229	rcu_read_unlock();
1230	shmem_unlock_mapping(shm_file->f_mapping);
1231
1232	fput(shm_file);
1233	return err;
1234
1235out_unlock0:
1236	ipc_unlock_object(&shp->shm_perm);
1237out_unlock1:
1238	rcu_read_unlock();
1239	return err;
1240}
1241
1242static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1243{
1244	int err;
1245	struct ipc_namespace *ns;
1246	struct shmid64_ds sem64;
1247
1248	if (cmd < 0 || shmid < 0)
1249		return -EINVAL;
1250
1251	ns = current->nsproxy->ipc_ns;
1252
1253	switch (cmd) {
1254	case IPC_INFO: {
1255		struct shminfo64 shminfo;
1256		err = shmctl_ipc_info(ns, &shminfo);
1257		if (err < 0)
1258			return err;
1259		if (copy_shminfo_to_user(buf, &shminfo, version))
1260			err = -EFAULT;
1261		return err;
1262	}
1263	case SHM_INFO: {
1264		struct shm_info shm_info;
1265		err = shmctl_shm_info(ns, &shm_info);
1266		if (err < 0)
1267			return err;
1268		if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1269			err = -EFAULT;
1270		return err;
1271	}
1272	case SHM_STAT:
1273	case SHM_STAT_ANY:
1274	case IPC_STAT: {
1275		err = shmctl_stat(ns, shmid, cmd, &sem64);
1276		if (err < 0)
1277			return err;
1278		if (copy_shmid_to_user(buf, &sem64, version))
1279			err = -EFAULT;
1280		return err;
1281	}
1282	case IPC_SET:
1283		if (copy_shmid_from_user(&sem64, buf, version))
1284			return -EFAULT;
1285		fallthrough;
1286	case IPC_RMID:
1287		return shmctl_down(ns, shmid, cmd, &sem64);
1288	case SHM_LOCK:
1289	case SHM_UNLOCK:
1290		return shmctl_do_lock(ns, shmid, cmd);
1291	default:
1292		return -EINVAL;
1293	}
1294}
1295
1296SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1297{
1298	return ksys_shmctl(shmid, cmd, buf, IPC_64);
1299}
1300
1301#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1302long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1303{
1304	int version = ipc_parse_version(&cmd);
1305
1306	return ksys_shmctl(shmid, cmd, buf, version);
1307}
1308
1309SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1310{
1311	return ksys_old_shmctl(shmid, cmd, buf);
1312}
1313#endif
1314
1315#ifdef CONFIG_COMPAT
1316
1317struct compat_shmid_ds {
1318	struct compat_ipc_perm shm_perm;
1319	int shm_segsz;
1320	old_time32_t shm_atime;
1321	old_time32_t shm_dtime;
1322	old_time32_t shm_ctime;
1323	compat_ipc_pid_t shm_cpid;
1324	compat_ipc_pid_t shm_lpid;
1325	unsigned short shm_nattch;
1326	unsigned short shm_unused;
1327	compat_uptr_t shm_unused2;
1328	compat_uptr_t shm_unused3;
1329};
1330
1331struct compat_shminfo64 {
1332	compat_ulong_t shmmax;
1333	compat_ulong_t shmmin;
1334	compat_ulong_t shmmni;
1335	compat_ulong_t shmseg;
1336	compat_ulong_t shmall;
1337	compat_ulong_t __unused1;
1338	compat_ulong_t __unused2;
1339	compat_ulong_t __unused3;
1340	compat_ulong_t __unused4;
1341};
1342
1343struct compat_shm_info {
1344	compat_int_t used_ids;
1345	compat_ulong_t shm_tot, shm_rss, shm_swp;
1346	compat_ulong_t swap_attempts, swap_successes;
1347};
1348
1349static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1350					int version)
1351{
1352	if (in->shmmax > INT_MAX)
1353		in->shmmax = INT_MAX;
1354	if (version == IPC_64) {
1355		struct compat_shminfo64 info;
1356		memset(&info, 0, sizeof(info));
1357		info.shmmax = in->shmmax;
1358		info.shmmin = in->shmmin;
1359		info.shmmni = in->shmmni;
1360		info.shmseg = in->shmseg;
1361		info.shmall = in->shmall;
1362		return copy_to_user(buf, &info, sizeof(info));
1363	} else {
1364		struct shminfo info;
1365		memset(&info, 0, sizeof(info));
1366		info.shmmax = in->shmmax;
1367		info.shmmin = in->shmmin;
1368		info.shmmni = in->shmmni;
1369		info.shmseg = in->shmseg;
1370		info.shmall = in->shmall;
1371		return copy_to_user(buf, &info, sizeof(info));
1372	}
1373}
1374
1375static int put_compat_shm_info(struct shm_info *ip,
1376				struct compat_shm_info __user *uip)
1377{
1378	struct compat_shm_info info;
1379
1380	memset(&info, 0, sizeof(info));
1381	info.used_ids = ip->used_ids;
1382	info.shm_tot = ip->shm_tot;
1383	info.shm_rss = ip->shm_rss;
1384	info.shm_swp = ip->shm_swp;
1385	info.swap_attempts = ip->swap_attempts;
1386	info.swap_successes = ip->swap_successes;
1387	return copy_to_user(uip, &info, sizeof(info));
1388}
1389
1390static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1391					int version)
1392{
1393	if (version == IPC_64) {
1394		struct compat_shmid64_ds v;
1395		memset(&v, 0, sizeof(v));
1396		to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1397		v.shm_atime	 = lower_32_bits(in->shm_atime);
1398		v.shm_atime_high = upper_32_bits(in->shm_atime);
1399		v.shm_dtime	 = lower_32_bits(in->shm_dtime);
1400		v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1401		v.shm_ctime	 = lower_32_bits(in->shm_ctime);
1402		v.shm_ctime_high = upper_32_bits(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	} else {
1409		struct compat_shmid_ds v;
1410		memset(&v, 0, sizeof(v));
1411		to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1412		v.shm_perm.key = in->shm_perm.key;
1413		v.shm_atime = in->shm_atime;
1414		v.shm_dtime = in->shm_dtime;
1415		v.shm_ctime = in->shm_ctime;
1416		v.shm_segsz = in->shm_segsz;
1417		v.shm_nattch = in->shm_nattch;
1418		v.shm_cpid = in->shm_cpid;
1419		v.shm_lpid = in->shm_lpid;
1420		return copy_to_user(buf, &v, sizeof(v));
1421	}
1422}
1423
1424static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1425					int version)
1426{
1427	memset(out, 0, sizeof(*out));
1428	if (version == IPC_64) {
1429		struct compat_shmid64_ds __user *p = buf;
1430		return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1431	} else {
1432		struct compat_shmid_ds __user *p = buf;
1433		return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1434	}
1435}
1436
1437static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1438{
1439	struct ipc_namespace *ns;
1440	struct shmid64_ds sem64;
1441	int err;
1442
1443	ns = current->nsproxy->ipc_ns;
1444
1445	if (cmd < 0 || shmid < 0)
1446		return -EINVAL;
1447
1448	switch (cmd) {
1449	case IPC_INFO: {
1450		struct shminfo64 shminfo;
1451		err = shmctl_ipc_info(ns, &shminfo);
1452		if (err < 0)
1453			return err;
1454		if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1455			err = -EFAULT;
1456		return err;
1457	}
1458	case SHM_INFO: {
1459		struct shm_info shm_info;
1460		err = shmctl_shm_info(ns, &shm_info);
1461		if (err < 0)
1462			return err;
1463		if (put_compat_shm_info(&shm_info, uptr))
1464			err = -EFAULT;
1465		return err;
1466	}
1467	case IPC_STAT:
1468	case SHM_STAT_ANY:
1469	case SHM_STAT:
1470		err = shmctl_stat(ns, shmid, cmd, &sem64);
1471		if (err < 0)
1472			return err;
1473		if (copy_compat_shmid_to_user(uptr, &sem64, version))
1474			err = -EFAULT;
1475		return err;
1476
1477	case IPC_SET:
1478		if (copy_compat_shmid_from_user(&sem64, uptr, version))
1479			return -EFAULT;
1480		fallthrough;
1481	case IPC_RMID:
1482		return shmctl_down(ns, shmid, cmd, &sem64);
1483	case SHM_LOCK:
1484	case SHM_UNLOCK:
1485		return shmctl_do_lock(ns, shmid, cmd);
1486	default:
1487		return -EINVAL;
1488	}
1489	return err;
1490}
1491
1492COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1493{
1494	return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1495}
1496
1497#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1498long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1499{
1500	int version = compat_ipc_parse_version(&cmd);
1501
1502	return compat_ksys_shmctl(shmid, cmd, uptr, version);
1503}
1504
1505COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1506{
1507	return compat_ksys_old_shmctl(shmid, cmd, uptr);
1508}
1509#endif
1510#endif
1511
1512/*
1513 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1514 *
1515 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1516 * "raddr" thing points to kernel space, and there has to be a wrapper around
1517 * this.
1518 */
1519long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1520	      ulong *raddr, unsigned long shmlba)
1521{
1522	struct shmid_kernel *shp;
1523	unsigned long addr = (unsigned long)shmaddr;
1524	unsigned long size;
1525	struct file *file, *base;
1526	int    err;
1527	unsigned long flags = MAP_SHARED;
1528	unsigned long prot;
1529	int acc_mode;
1530	struct ipc_namespace *ns;
1531	struct shm_file_data *sfd;
1532	int f_flags;
1533	unsigned long populate = 0;
1534
1535	err = -EINVAL;
1536	if (shmid < 0)
1537		goto out;
1538
1539	if (addr) {
1540		if (addr & (shmlba - 1)) {
1541			if (shmflg & SHM_RND) {
1542				addr &= ~(shmlba - 1);  /* round down */
1543
1544				/*
1545				 * Ensure that the round-down is non-nil
1546				 * when remapping. This can happen for
1547				 * cases when addr < shmlba.
1548				 */
1549				if (!addr && (shmflg & SHM_REMAP))
1550					goto out;
1551			} else
1552#ifndef __ARCH_FORCE_SHMLBA
1553				if (addr & ~PAGE_MASK)
1554#endif
1555					goto out;
1556		}
1557
1558		flags |= MAP_FIXED;
1559	} else if ((shmflg & SHM_REMAP))
1560		goto out;
1561
1562	if (shmflg & SHM_RDONLY) {
1563		prot = PROT_READ;
1564		acc_mode = S_IRUGO;
1565		f_flags = O_RDONLY;
1566	} else {
1567		prot = PROT_READ | PROT_WRITE;
1568		acc_mode = S_IRUGO | S_IWUGO;
1569		f_flags = O_RDWR;
1570	}
1571	if (shmflg & SHM_EXEC) {
1572		prot |= PROT_EXEC;
1573		acc_mode |= S_IXUGO;
1574	}
1575
1576	/*
1577	 * We cannot rely on the fs check since SYSV IPC does have an
1578	 * additional creator id...
1579	 */
1580	ns = current->nsproxy->ipc_ns;
1581	rcu_read_lock();
1582	shp = shm_obtain_object_check(ns, shmid);
1583	if (IS_ERR(shp)) {
1584		err = PTR_ERR(shp);
1585		goto out_unlock;
1586	}
1587
1588	err = -EACCES;
1589	if (ipcperms(ns, &shp->shm_perm, acc_mode))
1590		goto out_unlock;
1591
1592	err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1593	if (err)
1594		goto out_unlock;
1595
1596	ipc_lock_object(&shp->shm_perm);
1597
1598	/* check if shm_destroy() is tearing down shp */
1599	if (!ipc_valid_object(&shp->shm_perm)) {
1600		ipc_unlock_object(&shp->shm_perm);
1601		err = -EIDRM;
1602		goto out_unlock;
1603	}
1604
1605	/*
1606	 * We need to take a reference to the real shm file to prevent the
1607	 * pointer from becoming stale in cases where the lifetime of the outer
1608	 * file extends beyond that of the shm segment.  It's not usually
1609	 * possible, but it can happen during remap_file_pages() emulation as
1610	 * that unmaps the memory, then does ->mmap() via file reference only.
1611	 * We'll deny the ->mmap() if the shm segment was since removed, but to
1612	 * detect shm ID reuse we need to compare the file pointers.
1613	 */
1614	base = get_file(shp->shm_file);
1615	shp->shm_nattch++;
1616	size = i_size_read(file_inode(base));
1617	ipc_unlock_object(&shp->shm_perm);
1618	rcu_read_unlock();
1619
1620	err = -ENOMEM;
1621	sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1622	if (!sfd) {
1623		fput(base);
1624		goto out_nattch;
1625	}
1626
1627	file = alloc_file_clone(base, f_flags,
1628			  is_file_hugepages(base) ?
1629				&shm_file_operations_huge :
1630				&shm_file_operations);
1631	err = PTR_ERR(file);
1632	if (IS_ERR(file)) {
1633		kfree(sfd);
1634		fput(base);
1635		goto out_nattch;
1636	}
1637
1638	sfd->id = shp->shm_perm.id;
1639	sfd->ns = get_ipc_ns(ns);
1640	sfd->file = base;
1641	sfd->vm_ops = NULL;
1642	file->private_data = sfd;
1643
1644	err = security_mmap_file(file, prot, flags);
1645	if (err)
1646		goto out_fput;
1647
1648	if (mmap_write_lock_killable(current->mm)) {
1649		err = -EINTR;
1650		goto out_fput;
1651	}
1652
1653	if (addr && !(shmflg & SHM_REMAP)) {
1654		err = -EINVAL;
1655		if (addr + size < addr)
1656			goto invalid;
1657
1658		if (find_vma_intersection(current->mm, addr, addr + size))
1659			goto invalid;
1660	}
1661
1662	addr = do_mmap(file, addr, size, prot, flags, 0, 0, &populate, NULL);
1663	*raddr = addr;
1664	err = 0;
1665	if (IS_ERR_VALUE(addr))
1666		err = (long)addr;
1667invalid:
1668	mmap_write_unlock(current->mm);
1669	if (populate)
1670		mm_populate(addr, populate);
1671
1672out_fput:
1673	fput(file);
1674
1675out_nattch:
1676	down_write(&shm_ids(ns).rwsem);
1677	shp = shm_lock(ns, shmid);
1678	shp->shm_nattch--;
1679
1680	if (shm_may_destroy(shp))
1681		shm_destroy(ns, shp);
1682	else
1683		shm_unlock(shp);
1684	up_write(&shm_ids(ns).rwsem);
1685	return err;
1686
1687out_unlock:
1688	rcu_read_unlock();
1689out:
1690	return err;
1691}
1692
1693SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1694{
1695	unsigned long ret;
1696	long err;
1697
1698	err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1699	if (err)
1700		return err;
1701	force_successful_syscall_return();
1702	return (long)ret;
1703}
1704
1705#ifdef CONFIG_COMPAT
1706
1707#ifndef COMPAT_SHMLBA
1708#define COMPAT_SHMLBA	SHMLBA
1709#endif
1710
1711COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1712{
1713	unsigned long ret;
1714	long err;
1715
1716	err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1717	if (err)
1718		return err;
1719	force_successful_syscall_return();
1720	return (long)ret;
1721}
1722#endif
1723
1724/*
1725 * detach and kill segment if marked destroyed.
1726 * The work is done in shm_close.
1727 */
1728long ksys_shmdt(char __user *shmaddr)
1729{
1730	struct mm_struct *mm = current->mm;
1731	struct vm_area_struct *vma;
1732	unsigned long addr = (unsigned long)shmaddr;
1733	int retval = -EINVAL;
1734#ifdef CONFIG_MMU
1735	loff_t size = 0;
1736	struct file *file;
1737	VMA_ITERATOR(vmi, mm, addr);
1738#endif
1739
1740	if (addr & ~PAGE_MASK)
1741		return retval;
1742
1743	if (mmap_write_lock_killable(mm))
1744		return -EINTR;
1745
1746	/*
1747	 * This function tries to be smart and unmap shm segments that
1748	 * were modified by partial mlock or munmap calls:
1749	 * - It first determines the size of the shm segment that should be
1750	 *   unmapped: It searches for a vma that is backed by shm and that
1751	 *   started at address shmaddr. It records it's size and then unmaps
1752	 *   it.
1753	 * - Then it unmaps all shm vmas that started at shmaddr and that
1754	 *   are within the initially determined size and that are from the
1755	 *   same shm segment from which we determined the size.
1756	 * Errors from do_munmap are ignored: the function only fails if
1757	 * it's called with invalid parameters or if it's called to unmap
1758	 * a part of a vma. Both calls in this function are for full vmas,
1759	 * the parameters are directly copied from the vma itself and always
1760	 * valid - therefore do_munmap cannot fail. (famous last words?)
1761	 */
1762	/*
1763	 * If it had been mremap()'d, the starting address would not
1764	 * match the usual checks anyway. So assume all vma's are
1765	 * above the starting address given.
1766	 */
1767
1768#ifdef CONFIG_MMU
1769	for_each_vma(vmi, vma) {
1770		/*
1771		 * Check if the starting address would match, i.e. it's
1772		 * a fragment created by mprotect() and/or munmap(), or it
1773		 * otherwise it starts at this address with no hassles.
1774		 */
1775		if ((vma->vm_ops == &shm_vm_ops) &&
1776			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1777
1778			/*
1779			 * Record the file of the shm segment being
1780			 * unmapped.  With mremap(), someone could place
1781			 * page from another segment but with equal offsets
1782			 * in the range we are unmapping.
1783			 */
1784			file = vma->vm_file;
1785			size = i_size_read(file_inode(vma->vm_file));
1786			do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1787					    vma->vm_end, NULL, false);
1788			/*
1789			 * We discovered the size of the shm segment, so
1790			 * break out of here and fall through to the next
1791			 * loop that uses the size information to stop
1792			 * searching for matching vma's.
1793			 */
1794			retval = 0;
1795			vma = vma_next(&vmi);
1796			break;
1797		}
1798	}
1799
1800	/*
1801	 * We need look no further than the maximum address a fragment
1802	 * could possibly have landed at. Also cast things to loff_t to
1803	 * prevent overflows and make comparisons vs. equal-width types.
1804	 */
1805	size = PAGE_ALIGN(size);
1806	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1807		/* finding a matching vma now does not alter retval */
1808		if ((vma->vm_ops == &shm_vm_ops) &&
1809		    ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1810		    (vma->vm_file == file)) {
1811			do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1812					    vma->vm_end, NULL, false);
1813		}
1814
1815		vma = vma_next(&vmi);
1816	}
1817
1818#else	/* CONFIG_MMU */
1819	vma = vma_lookup(mm, addr);
1820	/* under NOMMU conditions, the exact address to be destroyed must be
1821	 * given
1822	 */
1823	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1824		do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1825		retval = 0;
1826	}
1827
1828#endif
1829
1830	mmap_write_unlock(mm);
1831	return retval;
1832}
1833
1834SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1835{
1836	return ksys_shmdt(shmaddr);
1837}
1838
1839#ifdef CONFIG_PROC_FS
1840static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1841{
1842	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1843	struct user_namespace *user_ns = seq_user_ns(s);
1844	struct kern_ipc_perm *ipcp = it;
1845	struct shmid_kernel *shp;
1846	unsigned long rss = 0, swp = 0;
1847
1848	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1849	shm_add_rss_swap(shp, &rss, &swp);
1850
1851#if BITS_PER_LONG <= 32
1852#define SIZE_SPEC "%10lu"
1853#else
1854#define SIZE_SPEC "%21lu"
1855#endif
1856
1857	seq_printf(s,
1858		   "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1859		   "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1860		   SIZE_SPEC " " SIZE_SPEC "\n",
1861		   shp->shm_perm.key,
1862		   shp->shm_perm.id,
1863		   shp->shm_perm.mode,
1864		   shp->shm_segsz,
1865		   pid_nr_ns(shp->shm_cprid, pid_ns),
1866		   pid_nr_ns(shp->shm_lprid, pid_ns),
1867		   shp->shm_nattch,
1868		   from_kuid_munged(user_ns, shp->shm_perm.uid),
1869		   from_kgid_munged(user_ns, shp->shm_perm.gid),
1870		   from_kuid_munged(user_ns, shp->shm_perm.cuid),
1871		   from_kgid_munged(user_ns, shp->shm_perm.cgid),
1872		   shp->shm_atim,
1873		   shp->shm_dtim,
1874		   shp->shm_ctim,
1875		   rss * PAGE_SIZE,
1876		   swp * PAGE_SIZE);
1877
1878	return 0;
1879}
1880#endif