ipc/shm.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/ipc/shm.c
   4  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   5  *       Many improvements/fixes by Bruno Haible.
   6  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   7  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   8  *
   9  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  10  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  11  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  12  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  13  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  14  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  15  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  *
  24  * Better ipc lock (kern_ipc_perm.lock) handling
  25  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
  26  */
  27
  28 #include <linux/slab.h>
  29 #include <linux/mm.h>
  30 #include <linux/hugetlb.h>
  31 #include <linux/shm.h>
  32 #include <linux/init.h>
  33 #include <linux/file.h>
  34 #include <linux/mman.h>
  35 #include <linux/shmem_fs.h>
  36 #include <linux/security.h>
  37 #include <linux/syscalls.h>
  38 #include <linux/audit.h>
  39 #include <linux/capability.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/seq_file.h>
  42 #include <linux/rwsem.h>
  43 #include <linux/nsproxy.h>
  44 #include <linux/mount.h>
  45 #include <linux/ipc_namespace.h>
  46 #include <linux/rhashtable.h>
  47
  48 #include <linux/uaccess.h>
  49
  50 #include "util.h"
  51
  52 struct shmid_kernel /* private to the kernel */
  53 {
  54         struct kern_ipc_perm    shm_perm;
  55         struct file             *shm_file;
  56         unsigned long           shm_nattch;
  57         unsigned long           shm_segsz;
  58         time64_t                shm_atim;
  59         time64_t                shm_dtim;
  60         time64_t                shm_ctim;
  61         struct pid              *shm_cprid;
  62         struct pid              *shm_lprid;
  63         struct ucounts          *mlock_ucounts;
  64
  65         /*
  66          * The task created the shm object, for
  67          * task_lock(shp->shm_creator)
  68          */
  69         struct task_struct      *shm_creator;
  70
  71         /*
  72          * List by creator. task_lock(->shm_creator) required for read/write.
  73          * If list_empty(), then the creator is dead already.
  74          */
  75         struct list_head        shm_clist;
  76         struct ipc_namespace    *ns;
  77 } __randomize_layout;
  78
  79 /* shm_mode upper byte flags */
  80 #define SHM_DEST        01000   /* segment will be destroyed on last detach */
  81 #define SHM_LOCKED      02000   /* segment will not be swapped */
  82
  83 struct shm_file_data {
  84         int id;
  85         struct ipc_namespace *ns;
  86         struct file *file;
  87         const struct vm_operations_struct *vm_ops;
  88 };
  89
  90 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  91
  92 static const struct file_operations shm_file_operations;
  93 static const struct vm_operations_struct shm_vm_ops;
  94
  95 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  96
  97 #define shm_unlock(shp)                 \
  98         ipc_unlock(&(shp)->shm_perm)
  99
 100 static int newseg(struct ipc_namespace *, struct ipc_params *);
 101 static void shm_open(struct vm_area_struct *vma);
 102 static void shm_close(struct vm_area_struct *vma);
 103 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
 104 #ifdef CONFIG_PROC_FS
 105 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
 106 #endif
 107
 108 void shm_init_ns(struct ipc_namespace *ns)
 109 {
 110         ns->shm_ctlmax = SHMMAX;
 111         ns->shm_ctlall = SHMALL;
 112         ns->shm_ctlmni = SHMMNI;
 113         ns->shm_rmid_forced = 0;
 114         ns->shm_tot = 0;
 115         ipc_init_ids(&shm_ids(ns));
 116 }
 117
 118 /*
 119  * Called with shm_ids.rwsem (writer) and the shp structure locked.
 120  * Only shm_ids.rwsem remains locked on exit.
 121  */
 122 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 123 {
 124         struct shmid_kernel *shp;
 125
 126         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 127         WARN_ON(ns != shp->ns);
 128
 129         if (shp->shm_nattch) {
 130                 shp->shm_perm.mode |= SHM_DEST;
 131                 /* Do not find it any more */
 132                 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
 133                 shm_unlock(shp);
 134         } else
 135                 shm_destroy(ns, shp);
 136 }
 137
 138 #ifdef CONFIG_IPC_NS
 139 void shm_exit_ns(struct ipc_namespace *ns)
 140 {
 141         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 142         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 143         rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
 144 }
 145 #endif
 146
 147 static int __init ipc_ns_init(void)
 148 {
 149         shm_init_ns(&init_ipc_ns);
 150         return 0;
 151 }
 152
 153 pure_initcall(ipc_ns_init);
 154
 155 void __init shm_init(void)
 156 {
 157         ipc_init_proc_interface("sysvipc/shm",
 158 #if BITS_PER_LONG <= 32
 159                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 160 #else
 161                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 162 #endif
 163                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 164 }
 165
 166 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 167 {
 168         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 169
 170         if (IS_ERR(ipcp))
 171                 return ERR_CAST(ipcp);
 172
 173         return container_of(ipcp, struct shmid_kernel, shm_perm);
 174 }
 175
 176 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 177 {
 178         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 179
 180         if (IS_ERR(ipcp))
 181                 return ERR_CAST(ipcp);
 182
 183         return container_of(ipcp, struct shmid_kernel, shm_perm);
 184 }
 185
 186 /*
 187  * shm_lock_(check_) routines are called in the paths where the rwsem
 188  * is not necessarily held.
 189  */
 190 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 191 {
 192         struct kern_ipc_perm *ipcp;
 193
 194         rcu_read_lock();
 195         ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 196         if (IS_ERR(ipcp))
 197                 goto err;
 198
 199         ipc_lock_object(ipcp);
 200         /*
 201          * ipc_rmid() may have already freed the ID while ipc_lock_object()
 202          * was spinning: here verify that the structure is still valid.
 203          * Upon races with RMID, return -EIDRM, thus indicating that
 204          * the ID points to a removed identifier.
 205          */
 206         if (ipc_valid_object(ipcp)) {
 207                 /* return a locked ipc object upon success */
 208                 return container_of(ipcp, struct shmid_kernel, shm_perm);
 209         }
 210
 211         ipc_unlock_object(ipcp);
 212         ipcp = ERR_PTR(-EIDRM);
 213 err:
 214         rcu_read_unlock();
 215         /*
 216          * Callers of shm_lock() must validate the status of the returned ipc
 217          * object pointer and error out as appropriate.
 218          */
 219         return ERR_CAST(ipcp);
 220 }
 221
 222 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 223 {
 224         rcu_read_lock();
 225         ipc_lock_object(&ipcp->shm_perm);
 226 }
 227
 228 static void shm_rcu_free(struct rcu_head *head)
 229 {
 230         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 231                                                         rcu);
 232         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 233                                                         shm_perm);
 234         security_shm_free(&shp->shm_perm);
 235         kfree(shp);
 236 }
 237
 238 /*
 239  * It has to be called with shp locked.
 240  * It must be called before ipc_rmid()
 241  */
 242 static inline void shm_clist_rm(struct shmid_kernel *shp)
 243 {
 244         struct task_struct *creator;
 245
 246         /* ensure that shm_creator does not disappear */
 247         rcu_read_lock();
 248
 249         /*
 250          * A concurrent exit_shm may do a list_del_init() as well.
 251          * Just do nothing if exit_shm already did the work
 252          */
 253         if (!list_empty(&shp->shm_clist)) {
 254                 /*
 255                  * shp->shm_creator is guaranteed to be valid *only*
 256                  * if shp->shm_clist is not empty.
 257                  */
 258                 creator = shp->shm_creator;
 259
 260                 task_lock(creator);
 261                 /*
 262                  * list_del_init() is a nop if the entry was already removed
 263                  * from the list.
 264                  */
 265                 list_del_init(&shp->shm_clist);
 266                 task_unlock(creator);
 267         }
 268         rcu_read_unlock();
 269 }
 270
 271 static inline void shm_rmid(struct shmid_kernel *s)
 272 {
 273         shm_clist_rm(s);
 274         ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
 275 }
 276
 277
 278 static int __shm_open(struct vm_area_struct *vma)
 279 {
 280         struct file *file = vma->vm_file;
 281         struct shm_file_data *sfd = shm_file_data(file);
 282         struct shmid_kernel *shp;
 283
 284         shp = shm_lock(sfd->ns, sfd->id);
 285
 286         if (IS_ERR(shp))
 287                 return PTR_ERR(shp);
 288
 289         if (shp->shm_file != sfd->file) {
 290                 /* ID was reused */
 291                 shm_unlock(shp);
 292                 return -EINVAL;
 293         }
 294
 295         shp->shm_atim = ktime_get_real_seconds();
 296         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 297         shp->shm_nattch++;
 298         shm_unlock(shp);
 299         return 0;
 300 }
 301
 302 /* This is called by fork, once for every shm attach. */
 303 static void shm_open(struct vm_area_struct *vma)
 304 {
 305         int err = __shm_open(vma);
 306         /*
 307          * We raced in the idr lookup or with shm_destroy().
 308          * Either way, the ID is busted.
 309          */
 310         WARN_ON_ONCE(err);
 311 }
 312
 313 /*
 314  * shm_destroy - free the struct shmid_kernel
 315  *
 316  * @ns: namespace
 317  * @shp: struct to free
 318  *
 319  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 320  * but returns with shp unlocked and freed.
 321  */
 322 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 323 {
 324         struct file *shm_file;
 325
 326         shm_file = shp->shm_file;
 327         shp->shm_file = NULL;
 328         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 329         shm_rmid(shp);
 330         shm_unlock(shp);
 331         if (!is_file_hugepages(shm_file))
 332                 shmem_lock(shm_file, 0, shp->mlock_ucounts);
 333         else if (shp->mlock_ucounts)
 334                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 335                                 shp->mlock_ucounts);
 336         fput(shm_file);
 337         ipc_update_pid(&shp->shm_cprid, NULL);
 338         ipc_update_pid(&shp->shm_lprid, NULL);
 339         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 340 }
 341
 342 /*
 343  * shm_may_destroy - identifies whether shm segment should be destroyed now
 344  *
 345  * Returns true if and only if there are no active users of the segment and
 346  * one of the following is true:
 347  *
 348  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 349  *
 350  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 351  */
 352 static bool shm_may_destroy(struct shmid_kernel *shp)
 353 {
 354         return (shp->shm_nattch == 0) &&
 355                (shp->ns->shm_rmid_forced ||
 356                 (shp->shm_perm.mode & SHM_DEST));
 357 }
 358
 359 /*
 360  * remove the attach descriptor vma.
 361  * free memory for segment if it is marked destroyed.
 362  * The descriptor has already been removed from the current->mm->mmap list
 363  * and will later be kfree()d.
 364  */
 365 static void shm_close(struct vm_area_struct *vma)
 366 {
 367         struct file *file = vma->vm_file;
 368         struct shm_file_data *sfd = shm_file_data(file);
 369         struct shmid_kernel *shp;
 370         struct ipc_namespace *ns = sfd->ns;
 371
 372         down_write(&shm_ids(ns).rwsem);
 373         /* remove from the list of attaches of the shm segment */
 374         shp = shm_lock(ns, sfd->id);
 375
 376         /*
 377          * We raced in the idr lookup or with shm_destroy().
 378          * Either way, the ID is busted.
 379          */
 380         if (WARN_ON_ONCE(IS_ERR(shp)))
 381                 goto done; /* no-op */
 382
 383         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 384         shp->shm_dtim = ktime_get_real_seconds();
 385         shp->shm_nattch--;
 386         if (shm_may_destroy(shp))
 387                 shm_destroy(ns, shp);
 388         else
 389                 shm_unlock(shp);
 390 done:
 391         up_write(&shm_ids(ns).rwsem);
 392 }
 393
 394 /* Called with ns->shm_ids(ns).rwsem locked */
 395 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 396 {
 397         struct ipc_namespace *ns = data;
 398         struct kern_ipc_perm *ipcp = p;
 399         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 400
 401         /*
 402          * We want to destroy segments without users and with already
 403          * exit'ed originating process.
 404          *
 405          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 406          */
 407         if (!list_empty(&shp->shm_clist))
 408                 return 0;
 409
 410         if (shm_may_destroy(shp)) {
 411                 shm_lock_by_ptr(shp);
 412                 shm_destroy(ns, shp);
 413         }
 414         return 0;
 415 }
 416
 417 void shm_destroy_orphaned(struct ipc_namespace *ns)
 418 {
 419         down_write(&shm_ids(ns).rwsem);
 420         if (shm_ids(ns).in_use)
 421                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 422         up_write(&shm_ids(ns).rwsem);
 423 }
 424
 425 /* Locking assumes this will only be called with task == current */
 426 void exit_shm(struct task_struct *task)
 427 {
 428         for (;;) {
 429                 struct shmid_kernel *shp;
 430                 struct ipc_namespace *ns;
 431
 432                 task_lock(task);
 433
 434                 if (list_empty(&task->sysvshm.shm_clist)) {
 435                         task_unlock(task);
 436                         break;
 437                 }
 438
 439                 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
 440                                 shm_clist);
 441
 442                 /*
 443                  * 1) Get pointer to the ipc namespace. It is worth to say
 444                  * that this pointer is guaranteed to be valid because
 445                  * shp lifetime is always shorter than namespace lifetime
 446                  * in which shp lives.
 447                  * We taken task_lock it means that shp won't be freed.
 448                  */
 449                 ns = shp->ns;
 450
 451                 /*
 452                  * 2) If kernel.shm_rmid_forced is not set then only keep track of
 453                  * which shmids are orphaned, so that a later set of the sysctl
 454                  * can clean them up.
 455                  */
 456                 if (!ns->shm_rmid_forced)
 457                         goto unlink_continue;
 458
 459                 /*
 460                  * 3) get a reference to the namespace.
 461                  *    The refcount could be already 0. If it is 0, then
 462                  *    the shm objects will be free by free_ipc_work().
 463                  */
 464                 ns = get_ipc_ns_not_zero(ns);
 465                 if (!ns) {
 466 unlink_continue:
 467                         list_del_init(&shp->shm_clist);
 468                         task_unlock(task);
 469                         continue;
 470                 }
 471
 472                 /*
 473                  * 4) get a reference to shp.
 474                  *   This cannot fail: shm_clist_rm() is called before
 475                  *   ipc_rmid(), thus the refcount cannot be 0.
 476                  */
 477                 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
 478
 479                 /*
 480                  * 5) unlink the shm segment from the list of segments
 481                  *    created by current.
 482                  *    This must be done last. After unlinking,
 483                  *    only the refcounts obtained above prevent IPC_RMID
 484                  *    from destroying the segment or the namespace.
 485                  */
 486                 list_del_init(&shp->shm_clist);
 487
 488                 task_unlock(task);
 489
 490                 /*
 491                  * 6) we have all references
 492                  *    Thus lock & if needed destroy shp.
 493                  */
 494                 down_write(&shm_ids(ns).rwsem);
 495                 shm_lock_by_ptr(shp);
 496                 /*
 497                  * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
 498                  * safe to call ipc_rcu_putref here
 499                  */
 500                 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 501
 502                 if (ipc_valid_object(&shp->shm_perm)) {
 503                         if (shm_may_destroy(shp))
 504                                 shm_destroy(ns, shp);
 505                         else
 506                                 shm_unlock(shp);
 507                 } else {
 508                         /*
 509                          * Someone else deleted the shp from namespace
 510                          * idr/kht while we have waited.
 511                          * Just unlock and continue.
 512                          */
 513                         shm_unlock(shp);
 514                 }
 515
 516                 up_write(&shm_ids(ns).rwsem);
 517                 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
 518         }
 519 }
 520
 521 static vm_fault_t shm_fault(struct vm_fault *vmf)
 522 {
 523         struct file *file = vmf->vma->vm_file;
 524         struct shm_file_data *sfd = shm_file_data(file);
 525
 526         return sfd->vm_ops->fault(vmf);
 527 }
 528
 529 static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
 530 {
 531         struct file *file = vma->vm_file;
 532         struct shm_file_data *sfd = shm_file_data(file);
 533
 534         if (sfd->vm_ops->may_split)
 535                 return sfd->vm_ops->may_split(vma, addr);
 536
 537         return 0;
 538 }
 539
 540 static unsigned long shm_pagesize(struct vm_area_struct *vma)
 541 {
 542         struct file *file = vma->vm_file;
 543         struct shm_file_data *sfd = shm_file_data(file);
 544
 545         if (sfd->vm_ops->pagesize)
 546                 return sfd->vm_ops->pagesize(vma);
 547
 548         return PAGE_SIZE;
 549 }
 550
 551 #ifdef CONFIG_NUMA
 552 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 553 {
 554         struct file *file = vma->vm_file;
 555         struct shm_file_data *sfd = shm_file_data(file);
 556         int err = 0;
 557
 558         if (sfd->vm_ops->set_policy)
 559                 err = sfd->vm_ops->set_policy(vma, new);
 560         return err;
 561 }
 562
 563 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 564                                         unsigned long addr)
 565 {
 566         struct file *file = vma->vm_file;
 567         struct shm_file_data *sfd = shm_file_data(file);
 568         struct mempolicy *pol = NULL;
 569
 570         if (sfd->vm_ops->get_policy)
 571                 pol = sfd->vm_ops->get_policy(vma, addr);
 572         else if (vma->vm_policy)
 573                 pol = vma->vm_policy;
 574
 575         return pol;
 576 }
 577 #endif
 578
 579 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 580 {
 581         struct shm_file_data *sfd = shm_file_data(file);
 582         int ret;
 583
 584         /*
 585          * In case of remap_file_pages() emulation, the file can represent an
 586          * IPC ID that was removed, and possibly even reused by another shm
 587          * segment already.  Propagate this case as an error to caller.
 588          */
 589         ret = __shm_open(vma);
 590         if (ret)
 591                 return ret;
 592
 593         ret = call_mmap(sfd->file, vma);
 594         if (ret) {
 595                 shm_close(vma);
 596                 return ret;
 597         }
 598         sfd->vm_ops = vma->vm_ops;
 599 #ifdef CONFIG_MMU
 600         WARN_ON(!sfd->vm_ops->fault);
 601 #endif
 602         vma->vm_ops = &shm_vm_ops;
 603         return 0;
 604 }
 605
 606 static int shm_release(struct inode *ino, struct file *file)
 607 {
 608         struct shm_file_data *sfd = shm_file_data(file);
 609
 610         put_ipc_ns(sfd->ns);
 611         fput(sfd->file);
 612         shm_file_data(file) = NULL;
 613         kfree(sfd);
 614         return 0;
 615 }
 616
 617 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 618 {
 619         struct shm_file_data *sfd = shm_file_data(file);
 620
 621         if (!sfd->file->f_op->fsync)
 622                 return -EINVAL;
 623         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 624 }
 625
 626 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 627                           loff_t len)
 628 {
 629         struct shm_file_data *sfd = shm_file_data(file);
 630
 631         if (!sfd->file->f_op->fallocate)
 632                 return -EOPNOTSUPP;
 633         return sfd->file->f_op->fallocate(file, mode, offset, len);
 634 }
 635
 636 static unsigned long shm_get_unmapped_area(struct file *file,
 637         unsigned long addr, unsigned long len, unsigned long pgoff,
 638         unsigned long flags)
 639 {
 640         struct shm_file_data *sfd = shm_file_data(file);
 641
 642         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 643                                                 pgoff, flags);
 644 }
 645
 646 static const struct file_operations shm_file_operations = {
 647         .mmap           = shm_mmap,
 648         .fsync          = shm_fsync,
 649         .release        = shm_release,
 650         .get_unmapped_area      = shm_get_unmapped_area,
 651         .llseek         = noop_llseek,
 652         .fallocate      = shm_fallocate,
 653 };
 654
 655 /*
 656  * shm_file_operations_huge is now identical to shm_file_operations,
 657  * but we keep it distinct for the sake of is_file_shm_hugepages().
 658  */
 659 static const struct file_operations shm_file_operations_huge = {
 660         .mmap           = shm_mmap,
 661         .fsync          = shm_fsync,
 662         .release        = shm_release,
 663         .get_unmapped_area      = shm_get_unmapped_area,
 664         .llseek         = noop_llseek,
 665         .fallocate      = shm_fallocate,
 666 };
 667
 668 bool is_file_shm_hugepages(struct file *file)
 669 {
 670         return file->f_op == &shm_file_operations_huge;
 671 }
 672
 673 static const struct vm_operations_struct shm_vm_ops = {
 674         .open   = shm_open,     /* callback for a new vm-area open */
 675         .close  = shm_close,    /* callback for when the vm-area is released */
 676         .fault  = shm_fault,
 677         .may_split = shm_may_split,
 678         .pagesize = shm_pagesize,
 679 #if defined(CONFIG_NUMA)
 680         .set_policy = shm_set_policy,
 681         .get_policy = shm_get_policy,
 682 #endif
 683 };
 684
 685 /**
 686  * newseg - Create a new shared memory segment
 687  * @ns: namespace
 688  * @params: ptr to the structure that contains key, size and shmflg
 689  *
 690  * Called with shm_ids.rwsem held as a writer.
 691  */
 692 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 693 {
 694         key_t key = params->key;
 695         int shmflg = params->flg;
 696         size_t size = params->u.size;
 697         int error;
 698         struct shmid_kernel *shp;
 699         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 700         struct file *file;
 701         char name[13];
 702         vm_flags_t acctflag = 0;
 703
 704         if (size < SHMMIN || size > ns->shm_ctlmax)
 705                 return -EINVAL;
 706
 707         if (numpages << PAGE_SHIFT < size)
 708                 return -ENOSPC;
 709
 710         if (ns->shm_tot + numpages < ns->shm_tot ||
 711                         ns->shm_tot + numpages > ns->shm_ctlall)
 712                 return -ENOSPC;
 713
 714         shp = kmalloc(sizeof(*shp), GFP_KERNEL_ACCOUNT);
 715         if (unlikely(!shp))
 716                 return -ENOMEM;
 717
 718         shp->shm_perm.key = key;
 719         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 720         shp->mlock_ucounts = NULL;
 721
 722         shp->shm_perm.security = NULL;
 723         error = security_shm_alloc(&shp->shm_perm);
 724         if (error) {
 725                 kfree(shp);
 726                 return error;
 727         }
 728
 729         sprintf(name, "SYSV%08x", key);
 730         if (shmflg & SHM_HUGETLB) {
 731                 struct hstate *hs;
 732                 size_t hugesize;
 733
 734                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 735                 if (!hs) {
 736                         error = -EINVAL;
 737                         goto no_file;
 738                 }
 739                 hugesize = ALIGN(size, huge_page_size(hs));
 740
 741                 /* hugetlb_file_setup applies strict accounting */
 742                 if (shmflg & SHM_NORESERVE)
 743                         acctflag = VM_NORESERVE;
 744                 file = hugetlb_file_setup(name, hugesize, acctflag,
 745                                   &shp->mlock_ucounts, HUGETLB_SHMFS_INODE,
 746                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 747         } else {
 748                 /*
 749                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 750                  * if it's asked for.
 751                  */
 752                 if  ((shmflg & SHM_NORESERVE) &&
 753                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 754                         acctflag = VM_NORESERVE;
 755                 file = shmem_kernel_file_setup(name, size, acctflag);
 756         }
 757         error = PTR_ERR(file);
 758         if (IS_ERR(file))
 759                 goto no_file;
 760
 761         shp->shm_cprid = get_pid(task_tgid(current));
 762         shp->shm_lprid = NULL;
 763         shp->shm_atim = shp->shm_dtim = 0;
 764         shp->shm_ctim = ktime_get_real_seconds();
 765         shp->shm_segsz = size;
 766         shp->shm_nattch = 0;
 767         shp->shm_file = file;
 768         shp->shm_creator = current;
 769
 770         /* ipc_addid() locks shp upon success. */
 771         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 772         if (error < 0)
 773                 goto no_id;
 774
 775         shp->ns = ns;
 776
 777         task_lock(current);
 778         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 779         task_unlock(current);
 780
 781         /*
 782          * shmid gets reported as "inode#" in /proc/pid/maps.
 783          * proc-ps tools use this. Changing this will break them.
 784          */
 785         file_inode(file)->i_ino = shp->shm_perm.id;
 786
 787         ns->shm_tot += numpages;
 788         error = shp->shm_perm.id;
 789
 790         ipc_unlock_object(&shp->shm_perm);
 791         rcu_read_unlock();
 792         return error;
 793
 794 no_id:
 795         ipc_update_pid(&shp->shm_cprid, NULL);
 796         ipc_update_pid(&shp->shm_lprid, NULL);
 797         if (is_file_hugepages(file) && shp->mlock_ucounts)
 798                 user_shm_unlock(size, shp->mlock_ucounts);
 799         fput(file);
 800         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 801         return error;
 802 no_file:
 803         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 804         return error;
 805 }
 806
 807 /*
 808  * Called with shm_ids.rwsem and ipcp locked.
 809  */
 810 static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
 811 {
 812         struct shmid_kernel *shp;
 813
 814         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 815         if (shp->shm_segsz < params->u.size)
 816                 return -EINVAL;
 817
 818         return 0;
 819 }
 820
 821 long ksys_shmget(key_t key, size_t size, int shmflg)
 822 {
 823         struct ipc_namespace *ns;
 824         static const struct ipc_ops shm_ops = {
 825                 .getnew = newseg,
 826                 .associate = security_shm_associate,
 827                 .more_checks = shm_more_checks,
 828         };
 829         struct ipc_params shm_params;
 830
 831         ns = current->nsproxy->ipc_ns;
 832
 833         shm_params.key = key;
 834         shm_params.flg = shmflg;
 835         shm_params.u.size = size;
 836
 837         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 838 }
 839
 840 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 841 {
 842         return ksys_shmget(key, size, shmflg);
 843 }
 844
 845 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 846 {
 847         switch (version) {
 848         case IPC_64:
 849                 return copy_to_user(buf, in, sizeof(*in));
 850         case IPC_OLD:
 851             {
 852                 struct shmid_ds out;
 853
 854                 memset(&out, 0, sizeof(out));
 855                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 856                 out.shm_segsz   = in->shm_segsz;
 857                 out.shm_atime   = in->shm_atime;
 858                 out.shm_dtime   = in->shm_dtime;
 859                 out.shm_ctime   = in->shm_ctime;
 860                 out.shm_cpid    = in->shm_cpid;
 861                 out.shm_lpid    = in->shm_lpid;
 862                 out.shm_nattch  = in->shm_nattch;
 863
 864                 return copy_to_user(buf, &out, sizeof(out));
 865             }
 866         default:
 867                 return -EINVAL;
 868         }
 869 }
 870
 871 static inline unsigned long
 872 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 873 {
 874         switch (version) {
 875         case IPC_64:
 876                 if (copy_from_user(out, buf, sizeof(*out)))
 877                         return -EFAULT;
 878                 return 0;
 879         case IPC_OLD:
 880             {
 881                 struct shmid_ds tbuf_old;
 882
 883                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 884                         return -EFAULT;
 885
 886                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 887                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 888                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 889
 890                 return 0;
 891             }
 892         default:
 893                 return -EINVAL;
 894         }
 895 }
 896
 897 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 898 {
 899         switch (version) {
 900         case IPC_64:
 901                 return copy_to_user(buf, in, sizeof(*in));
 902         case IPC_OLD:
 903             {
 904                 struct shminfo out;
 905
 906                 if (in->shmmax > INT_MAX)
 907                         out.shmmax = INT_MAX;
 908                 else
 909                         out.shmmax = (int)in->shmmax;
 910
 911                 out.shmmin      = in->shmmin;
 912                 out.shmmni      = in->shmmni;
 913                 out.shmseg      = in->shmseg;
 914                 out.shmall      = in->shmall;
 915
 916                 return copy_to_user(buf, &out, sizeof(out));
 917             }
 918         default:
 919                 return -EINVAL;
 920         }
 921 }
 922
 923 /*
 924  * Calculate and add used RSS and swap pages of a shm.
 925  * Called with shm_ids.rwsem held as a reader
 926  */
 927 static void shm_add_rss_swap(struct shmid_kernel *shp,
 928         unsigned long *rss_add, unsigned long *swp_add)
 929 {
 930         struct inode *inode;
 931
 932         inode = file_inode(shp->shm_file);
 933
 934         if (is_file_hugepages(shp->shm_file)) {
 935                 struct address_space *mapping = inode->i_mapping;
 936                 struct hstate *h = hstate_file(shp->shm_file);
 937                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 938         } else {
 939 #ifdef CONFIG_SHMEM
 940                 struct shmem_inode_info *info = SHMEM_I(inode);
 941
 942                 spin_lock_irq(&info->lock);
 943                 *rss_add += inode->i_mapping->nrpages;
 944                 *swp_add += info->swapped;
 945                 spin_unlock_irq(&info->lock);
 946 #else
 947                 *rss_add += inode->i_mapping->nrpages;
 948 #endif
 949         }
 950 }
 951
 952 /*
 953  * Called with shm_ids.rwsem held as a reader
 954  */
 955 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 956                 unsigned long *swp)
 957 {
 958         int next_id;
 959         int total, in_use;
 960
 961         *rss = 0;
 962         *swp = 0;
 963
 964         in_use = shm_ids(ns).in_use;
 965
 966         for (total = 0, next_id = 0; total < in_use; next_id++) {
 967                 struct kern_ipc_perm *ipc;
 968                 struct shmid_kernel *shp;
 969
 970                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 971                 if (ipc == NULL)
 972                         continue;
 973                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 974
 975                 shm_add_rss_swap(shp, rss, swp);
 976
 977                 total++;
 978         }
 979 }
 980
 981 /*
 982  * This function handles some shmctl commands which require the rwsem
 983  * to be held in write mode.
 984  * NOTE: no locks must be held, the rwsem is taken inside this function.
 985  */
 986 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 987                        struct shmid64_ds *shmid64)
 988 {
 989         struct kern_ipc_perm *ipcp;
 990         struct shmid_kernel *shp;
 991         int err;
 992
 993         down_write(&shm_ids(ns).rwsem);
 994         rcu_read_lock();
 995
 996         ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
 997                                       &shmid64->shm_perm, 0);
 998         if (IS_ERR(ipcp)) {
 999                 err = PTR_ERR(ipcp);
1000                 goto out_unlock1;
1001         }
1002
1003         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1004
1005         err = security_shm_shmctl(&shp->shm_perm, cmd);
1006         if (err)
1007                 goto out_unlock1;
1008
1009         switch (cmd) {
1010         case IPC_RMID:
1011                 ipc_lock_object(&shp->shm_perm);
1012                 /* do_shm_rmid unlocks the ipc object and rcu */
1013                 do_shm_rmid(ns, ipcp);
1014                 goto out_up;
1015         case IPC_SET:
1016                 ipc_lock_object(&shp->shm_perm);
1017                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1018                 if (err)
1019                         goto out_unlock0;
1020                 shp->shm_ctim = ktime_get_real_seconds();
1021                 break;
1022         default:
1023                 err = -EINVAL;
1024                 goto out_unlock1;
1025         }
1026
1027 out_unlock0:
1028         ipc_unlock_object(&shp->shm_perm);
1029 out_unlock1:
1030         rcu_read_unlock();
1031 out_up:
1032         up_write(&shm_ids(ns).rwsem);
1033         return err;
1034 }
1035
1036 static int shmctl_ipc_info(struct ipc_namespace *ns,
1037                            struct shminfo64 *shminfo)
1038 {
1039         int err = security_shm_shmctl(NULL, IPC_INFO);
1040         if (!err) {
1041                 memset(shminfo, 0, sizeof(*shminfo));
1042                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1043                 shminfo->shmmax = ns->shm_ctlmax;
1044                 shminfo->shmall = ns->shm_ctlall;
1045                 shminfo->shmmin = SHMMIN;
1046                 down_read(&shm_ids(ns).rwsem);
1047                 err = ipc_get_maxidx(&shm_ids(ns));
1048                 up_read(&shm_ids(ns).rwsem);
1049                 if (err < 0)
1050                         err = 0;
1051         }
1052         return err;
1053 }
1054
1055 static int shmctl_shm_info(struct ipc_namespace *ns,
1056                            struct shm_info *shm_info)
1057 {
1058         int err = security_shm_shmctl(NULL, SHM_INFO);
1059         if (!err) {
1060                 memset(shm_info, 0, sizeof(*shm_info));
1061                 down_read(&shm_ids(ns).rwsem);
1062                 shm_info->used_ids = shm_ids(ns).in_use;
1063                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1064                 shm_info->shm_tot = ns->shm_tot;
1065                 shm_info->swap_attempts = 0;
1066                 shm_info->swap_successes = 0;
1067                 err = ipc_get_maxidx(&shm_ids(ns));
1068                 up_read(&shm_ids(ns).rwsem);
1069                 if (err < 0)
1070                         err = 0;
1071         }
1072         return err;
1073 }
1074
1075 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1076                         int cmd, struct shmid64_ds *tbuf)
1077 {
1078         struct shmid_kernel *shp;
1079         int err;
1080
1081         memset(tbuf, 0, sizeof(*tbuf));
1082
1083         rcu_read_lock();
1084         if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1085                 shp = shm_obtain_object(ns, shmid);
1086                 if (IS_ERR(shp)) {
1087                         err = PTR_ERR(shp);
1088                         goto out_unlock;
1089                 }
1090         } else { /* IPC_STAT */
1091                 shp = shm_obtain_object_check(ns, shmid);
1092                 if (IS_ERR(shp)) {
1093                         err = PTR_ERR(shp);
1094                         goto out_unlock;
1095                 }
1096         }
1097
1098         /*
1099          * Semantically SHM_STAT_ANY ought to be identical to
1100          * that functionality provided by the /proc/sysvipc/
1101          * interface. As such, only audit these calls and
1102          * do not do traditional S_IRUGO permission checks on
1103          * the ipc object.
1104          */
1105         if (cmd == SHM_STAT_ANY)
1106                 audit_ipc_obj(&shp->shm_perm);
1107         else {
1108                 err = -EACCES;
1109                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1110                         goto out_unlock;
1111         }
1112
1113         err = security_shm_shmctl(&shp->shm_perm, cmd);
1114         if (err)
1115                 goto out_unlock;
1116
1117         ipc_lock_object(&shp->shm_perm);
1118
1119         if (!ipc_valid_object(&shp->shm_perm)) {
1120                 ipc_unlock_object(&shp->shm_perm);
1121                 err = -EIDRM;
1122                 goto out_unlock;
1123         }
1124
1125         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1126         tbuf->shm_segsz = shp->shm_segsz;
1127         tbuf->shm_atime = shp->shm_atim;
1128         tbuf->shm_dtime = shp->shm_dtim;
1129         tbuf->shm_ctime = shp->shm_ctim;
1130 #ifndef CONFIG_64BIT
1131         tbuf->shm_atime_high = shp->shm_atim >> 32;
1132         tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1133         tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1134 #endif
1135         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
1136         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
1137         tbuf->shm_nattch = shp->shm_nattch;
1138
1139         if (cmd == IPC_STAT) {
1140                 /*
1141                  * As defined in SUS:
1142                  * Return 0 on success
1143                  */
1144                 err = 0;
1145         } else {
1146                 /*
1147                  * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1148                  * Return the full id, including the sequence number
1149                  */
1150                 err = shp->shm_perm.id;
1151         }
1152
1153         ipc_unlock_object(&shp->shm_perm);
1154 out_unlock:
1155         rcu_read_unlock();
1156         return err;
1157 }
1158
1159 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1160 {
1161         struct shmid_kernel *shp;
1162         struct file *shm_file;
1163         int err;
1164
1165         rcu_read_lock();
1166         shp = shm_obtain_object_check(ns, shmid);
1167         if (IS_ERR(shp)) {
1168                 err = PTR_ERR(shp);
1169                 goto out_unlock1;
1170         }
1171
1172         audit_ipc_obj(&(shp->shm_perm));
1173         err = security_shm_shmctl(&shp->shm_perm, cmd);
1174         if (err)
1175                 goto out_unlock1;
1176
1177         ipc_lock_object(&shp->shm_perm);
1178
1179         /* check if shm_destroy() is tearing down shp */
1180         if (!ipc_valid_object(&shp->shm_perm)) {
1181                 err = -EIDRM;
1182                 goto out_unlock0;
1183         }
1184
1185         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1186                 kuid_t euid = current_euid();
1187
1188                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1189                     !uid_eq(euid, shp->shm_perm.cuid)) {
1190                         err = -EPERM;
1191                         goto out_unlock0;
1192                 }
1193                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1194                         err = -EPERM;
1195                         goto out_unlock0;
1196                 }
1197         }
1198
1199         shm_file = shp->shm_file;
1200         if (is_file_hugepages(shm_file))
1201                 goto out_unlock0;
1202
1203         if (cmd == SHM_LOCK) {
1204                 struct ucounts *ucounts = current_ucounts();
1205
1206                 err = shmem_lock(shm_file, 1, ucounts);
1207                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1208                         shp->shm_perm.mode |= SHM_LOCKED;
1209                         shp->mlock_ucounts = ucounts;
1210                 }
1211                 goto out_unlock0;
1212         }
1213
1214         /* SHM_UNLOCK */
1215         if (!(shp->shm_perm.mode & SHM_LOCKED))
1216                 goto out_unlock0;
1217         shmem_lock(shm_file, 0, shp->mlock_ucounts);
1218         shp->shm_perm.mode &= ~SHM_LOCKED;
1219         shp->mlock_ucounts = NULL;
1220         get_file(shm_file);
1221         ipc_unlock_object(&shp->shm_perm);
1222         rcu_read_unlock();
1223         shmem_unlock_mapping(shm_file->f_mapping);
1224
1225         fput(shm_file);
1226         return err;
1227
1228 out_unlock0:
1229         ipc_unlock_object(&shp->shm_perm);
1230 out_unlock1:
1231         rcu_read_unlock();
1232         return err;
1233 }
1234
1235 static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1236 {
1237         int err;
1238         struct ipc_namespace *ns;
1239         struct shmid64_ds sem64;
1240
1241         if (cmd < 0 || shmid < 0)
1242                 return -EINVAL;
1243
1244         ns = current->nsproxy->ipc_ns;
1245
1246         switch (cmd) {
1247         case IPC_INFO: {
1248                 struct shminfo64 shminfo;
1249                 err = shmctl_ipc_info(ns, &shminfo);
1250                 if (err < 0)
1251                         return err;
1252                 if (copy_shminfo_to_user(buf, &shminfo, version))
1253                         err = -EFAULT;
1254                 return err;
1255         }
1256         case SHM_INFO: {
1257                 struct shm_info shm_info;
1258                 err = shmctl_shm_info(ns, &shm_info);
1259                 if (err < 0)
1260                         return err;
1261                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1262                         err = -EFAULT;
1263                 return err;
1264         }
1265         case SHM_STAT:
1266         case SHM_STAT_ANY:
1267         case IPC_STAT: {
1268                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1269                 if (err < 0)
1270                         return err;
1271                 if (copy_shmid_to_user(buf, &sem64, version))
1272                         err = -EFAULT;
1273                 return err;
1274         }
1275         case IPC_SET:
1276                 if (copy_shmid_from_user(&sem64, buf, version))
1277                         return -EFAULT;
1278                 fallthrough;
1279         case IPC_RMID:
1280                 return shmctl_down(ns, shmid, cmd, &sem64);
1281         case SHM_LOCK:
1282         case SHM_UNLOCK:
1283                 return shmctl_do_lock(ns, shmid, cmd);
1284         default:
1285                 return -EINVAL;
1286         }
1287 }
1288
1289 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1290 {
1291         return ksys_shmctl(shmid, cmd, buf, IPC_64);
1292 }
1293
1294 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1295 long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1296 {
1297         int version = ipc_parse_version(&cmd);
1298
1299         return ksys_shmctl(shmid, cmd, buf, version);
1300 }
1301
1302 SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1303 {
1304         return ksys_old_shmctl(shmid, cmd, buf);
1305 }
1306 #endif
1307
1308 #ifdef CONFIG_COMPAT
1309
1310 struct compat_shmid_ds {
1311         struct compat_ipc_perm shm_perm;
1312         int shm_segsz;
1313         old_time32_t shm_atime;
1314         old_time32_t shm_dtime;
1315         old_time32_t shm_ctime;
1316         compat_ipc_pid_t shm_cpid;
1317         compat_ipc_pid_t shm_lpid;
1318         unsigned short shm_nattch;
1319         unsigned short shm_unused;
1320         compat_uptr_t shm_unused2;
1321         compat_uptr_t shm_unused3;
1322 };
1323
1324 struct compat_shminfo64 {
1325         compat_ulong_t shmmax;
1326         compat_ulong_t shmmin;
1327         compat_ulong_t shmmni;
1328         compat_ulong_t shmseg;
1329         compat_ulong_t shmall;
1330         compat_ulong_t __unused1;
1331         compat_ulong_t __unused2;
1332         compat_ulong_t __unused3;
1333         compat_ulong_t __unused4;
1334 };
1335
1336 struct compat_shm_info {
1337         compat_int_t used_ids;
1338         compat_ulong_t shm_tot, shm_rss, shm_swp;
1339         compat_ulong_t swap_attempts, swap_successes;
1340 };
1341
1342 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1343                                         int version)
1344 {
1345         if (in->shmmax > INT_MAX)
1346                 in->shmmax = INT_MAX;
1347         if (version == IPC_64) {
1348                 struct compat_shminfo64 info;
1349                 memset(&info, 0, sizeof(info));
1350                 info.shmmax = in->shmmax;
1351                 info.shmmin = in->shmmin;
1352                 info.shmmni = in->shmmni;
1353                 info.shmseg = in->shmseg;
1354                 info.shmall = in->shmall;
1355                 return copy_to_user(buf, &info, sizeof(info));
1356         } else {
1357                 struct shminfo info;
1358                 memset(&info, 0, sizeof(info));
1359                 info.shmmax = in->shmmax;
1360                 info.shmmin = in->shmmin;
1361                 info.shmmni = in->shmmni;
1362                 info.shmseg = in->shmseg;
1363                 info.shmall = in->shmall;
1364                 return copy_to_user(buf, &info, sizeof(info));
1365         }
1366 }
1367
1368 static int put_compat_shm_info(struct shm_info *ip,
1369                                 struct compat_shm_info __user *uip)
1370 {
1371         struct compat_shm_info info;
1372
1373         memset(&info, 0, sizeof(info));
1374         info.used_ids = ip->used_ids;
1375         info.shm_tot = ip->shm_tot;
1376         info.shm_rss = ip->shm_rss;
1377         info.shm_swp = ip->shm_swp;
1378         info.swap_attempts = ip->swap_attempts;
1379         info.swap_successes = ip->swap_successes;
1380         return copy_to_user(uip, &info, sizeof(info));
1381 }
1382
1383 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1384                                         int version)
1385 {
1386         if (version == IPC_64) {
1387                 struct compat_shmid64_ds v;
1388                 memset(&v, 0, sizeof(v));
1389                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1390                 v.shm_atime      = lower_32_bits(in->shm_atime);
1391                 v.shm_atime_high = upper_32_bits(in->shm_atime);
1392                 v.shm_dtime      = lower_32_bits(in->shm_dtime);
1393                 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1394                 v.shm_ctime      = lower_32_bits(in->shm_ctime);
1395                 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1396                 v.shm_segsz = in->shm_segsz;
1397                 v.shm_nattch = in->shm_nattch;
1398                 v.shm_cpid = in->shm_cpid;
1399                 v.shm_lpid = in->shm_lpid;
1400                 return copy_to_user(buf, &v, sizeof(v));
1401         } else {
1402                 struct compat_shmid_ds v;
1403                 memset(&v, 0, sizeof(v));
1404                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1405                 v.shm_perm.key = in->shm_perm.key;
1406                 v.shm_atime = in->shm_atime;
1407                 v.shm_dtime = in->shm_dtime;
1408                 v.shm_ctime = in->shm_ctime;
1409                 v.shm_segsz = in->shm_segsz;
1410                 v.shm_nattch = in->shm_nattch;
1411                 v.shm_cpid = in->shm_cpid;
1412                 v.shm_lpid = in->shm_lpid;
1413                 return copy_to_user(buf, &v, sizeof(v));
1414         }
1415 }
1416
1417 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1418                                         int version)
1419 {
1420         memset(out, 0, sizeof(*out));
1421         if (version == IPC_64) {
1422                 struct compat_shmid64_ds __user *p = buf;
1423                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1424         } else {
1425                 struct compat_shmid_ds __user *p = buf;
1426                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1427         }
1428 }
1429
1430 static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1431 {
1432         struct ipc_namespace *ns;
1433         struct shmid64_ds sem64;
1434         int err;
1435
1436         ns = current->nsproxy->ipc_ns;
1437
1438         if (cmd < 0 || shmid < 0)
1439                 return -EINVAL;
1440
1441         switch (cmd) {
1442         case IPC_INFO: {
1443                 struct shminfo64 shminfo;
1444                 err = shmctl_ipc_info(ns, &shminfo);
1445                 if (err < 0)
1446                         return err;
1447                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1448                         err = -EFAULT;
1449                 return err;
1450         }
1451         case SHM_INFO: {
1452                 struct shm_info shm_info;
1453                 err = shmctl_shm_info(ns, &shm_info);
1454                 if (err < 0)
1455                         return err;
1456                 if (put_compat_shm_info(&shm_info, uptr))
1457                         err = -EFAULT;
1458                 return err;
1459         }
1460         case IPC_STAT:
1461         case SHM_STAT_ANY:
1462         case SHM_STAT:
1463                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1464                 if (err < 0)
1465                         return err;
1466                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1467                         err = -EFAULT;
1468                 return err;
1469
1470         case IPC_SET:
1471                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1472                         return -EFAULT;
1473                 fallthrough;
1474         case IPC_RMID:
1475                 return shmctl_down(ns, shmid, cmd, &sem64);
1476         case SHM_LOCK:
1477         case SHM_UNLOCK:
1478                 return shmctl_do_lock(ns, shmid, cmd);
1479         default:
1480                 return -EINVAL;
1481         }
1482         return err;
1483 }
1484
1485 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1486 {
1487         return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1488 }
1489
1490 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1491 long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1492 {
1493         int version = compat_ipc_parse_version(&cmd);
1494
1495         return compat_ksys_shmctl(shmid, cmd, uptr, version);
1496 }
1497
1498 COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1499 {
1500         return compat_ksys_old_shmctl(shmid, cmd, uptr);
1501 }
1502 #endif
1503 #endif
1504
1505 /*
1506  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1507  *
1508  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1509  * "raddr" thing points to kernel space, and there has to be a wrapper around
1510  * this.
1511  */
1512 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1513               ulong *raddr, unsigned long shmlba)
1514 {
1515         struct shmid_kernel *shp;
1516         unsigned long addr = (unsigned long)shmaddr;
1517         unsigned long size;
1518         struct file *file, *base;
1519         int    err;
1520         unsigned long flags = MAP_SHARED;
1521         unsigned long prot;
1522         int acc_mode;
1523         struct ipc_namespace *ns;
1524         struct shm_file_data *sfd;
1525         int f_flags;
1526         unsigned long populate = 0;
1527
1528         err = -EINVAL;
1529         if (shmid < 0)
1530                 goto out;
1531
1532         if (addr) {
1533                 if (addr & (shmlba - 1)) {
1534                         if (shmflg & SHM_RND) {
1535                                 addr &= ~(shmlba - 1);  /* round down */
1536
1537                                 /*
1538                                  * Ensure that the round-down is non-nil
1539                                  * when remapping. This can happen for
1540                                  * cases when addr < shmlba.
1541                                  */
1542                                 if (!addr && (shmflg & SHM_REMAP))
1543                                         goto out;
1544                         } else
1545 #ifndef __ARCH_FORCE_SHMLBA
1546                                 if (addr & ~PAGE_MASK)
1547 #endif
1548                                         goto out;
1549                 }
1550
1551                 flags |= MAP_FIXED;
1552         } else if ((shmflg & SHM_REMAP))
1553                 goto out;
1554
1555         if (shmflg & SHM_RDONLY) {
1556                 prot = PROT_READ;
1557                 acc_mode = S_IRUGO;
1558                 f_flags = O_RDONLY;
1559         } else {
1560                 prot = PROT_READ | PROT_WRITE;
1561                 acc_mode = S_IRUGO | S_IWUGO;
1562                 f_flags = O_RDWR;
1563         }
1564         if (shmflg & SHM_EXEC) {
1565                 prot |= PROT_EXEC;
1566                 acc_mode |= S_IXUGO;
1567         }
1568
1569         /*
1570          * We cannot rely on the fs check since SYSV IPC does have an
1571          * additional creator id...
1572          */
1573         ns = current->nsproxy->ipc_ns;
1574         rcu_read_lock();
1575         shp = shm_obtain_object_check(ns, shmid);
1576         if (IS_ERR(shp)) {
1577                 err = PTR_ERR(shp);
1578                 goto out_unlock;
1579         }
1580
1581         err = -EACCES;
1582         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1583                 goto out_unlock;
1584
1585         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1586         if (err)
1587                 goto out_unlock;
1588
1589         ipc_lock_object(&shp->shm_perm);
1590
1591         /* check if shm_destroy() is tearing down shp */
1592         if (!ipc_valid_object(&shp->shm_perm)) {
1593                 ipc_unlock_object(&shp->shm_perm);
1594                 err = -EIDRM;
1595                 goto out_unlock;
1596         }
1597
1598         /*
1599          * We need to take a reference to the real shm file to prevent the
1600          * pointer from becoming stale in cases where the lifetime of the outer
1601          * file extends beyond that of the shm segment.  It's not usually
1602          * possible, but it can happen during remap_file_pages() emulation as
1603          * that unmaps the memory, then does ->mmap() via file reference only.
1604          * We'll deny the ->mmap() if the shm segment was since removed, but to
1605          * detect shm ID reuse we need to compare the file pointers.
1606          */
1607         base = get_file(shp->shm_file);
1608         shp->shm_nattch++;
1609         size = i_size_read(file_inode(base));
1610         ipc_unlock_object(&shp->shm_perm);
1611         rcu_read_unlock();
1612
1613         err = -ENOMEM;
1614         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1615         if (!sfd) {
1616                 fput(base);
1617                 goto out_nattch;
1618         }
1619
1620         file = alloc_file_clone(base, f_flags,
1621                           is_file_hugepages(base) ?
1622                                 &shm_file_operations_huge :
1623                                 &shm_file_operations);
1624         err = PTR_ERR(file);
1625         if (IS_ERR(file)) {
1626                 kfree(sfd);
1627                 fput(base);
1628                 goto out_nattch;
1629         }
1630
1631         sfd->id = shp->shm_perm.id;
1632         sfd->ns = get_ipc_ns(ns);
1633         sfd->file = base;
1634         sfd->vm_ops = NULL;
1635         file->private_data = sfd;
1636
1637         err = security_mmap_file(file, prot, flags);
1638         if (err)
1639                 goto out_fput;
1640
1641         if (mmap_write_lock_killable(current->mm)) {
1642                 err = -EINTR;
1643                 goto out_fput;
1644         }
1645
1646         if (addr && !(shmflg & SHM_REMAP)) {
1647                 err = -EINVAL;
1648                 if (addr + size < addr)
1649                         goto invalid;
1650
1651                 if (find_vma_intersection(current->mm, addr, addr + size))
1652                         goto invalid;
1653         }
1654
1655         addr = do_mmap(file, addr, size, prot, flags, 0, &populate, NULL);
1656         *raddr = addr;
1657         err = 0;
1658         if (IS_ERR_VALUE(addr))
1659                 err = (long)addr;
1660 invalid:
1661         mmap_write_unlock(current->mm);
1662         if (populate)
1663                 mm_populate(addr, populate);
1664
1665 out_fput:
1666         fput(file);
1667
1668 out_nattch:
1669         down_write(&shm_ids(ns).rwsem);
1670         shp = shm_lock(ns, shmid);
1671         shp->shm_nattch--;
1672
1673         if (shm_may_destroy(shp))
1674                 shm_destroy(ns, shp);
1675         else
1676                 shm_unlock(shp);
1677         up_write(&shm_ids(ns).rwsem);
1678         return err;
1679
1680 out_unlock:
1681         rcu_read_unlock();
1682 out:
1683         return err;
1684 }
1685
1686 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1687 {
1688         unsigned long ret;
1689         long err;
1690
1691         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1692         if (err)
1693                 return err;
1694         force_successful_syscall_return();
1695         return (long)ret;
1696 }
1697
1698 #ifdef CONFIG_COMPAT
1699
1700 #ifndef COMPAT_SHMLBA
1701 #define COMPAT_SHMLBA   SHMLBA
1702 #endif
1703
1704 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1705 {
1706         unsigned long ret;
1707         long err;
1708
1709         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1710         if (err)
1711                 return err;
1712         force_successful_syscall_return();
1713         return (long)ret;
1714 }
1715 #endif
1716
1717 /*
1718  * detach and kill segment if marked destroyed.
1719  * The work is done in shm_close.
1720  */
1721 long ksys_shmdt(char __user *shmaddr)
1722 {
1723         struct mm_struct *mm = current->mm;
1724         struct vm_area_struct *vma;
1725         unsigned long addr = (unsigned long)shmaddr;
1726         int retval = -EINVAL;
1727 #ifdef CONFIG_MMU
1728         loff_t size = 0;
1729         struct file *file;
1730         struct vm_area_struct *next;
1731 #endif
1732
1733         if (addr & ~PAGE_MASK)
1734                 return retval;
1735
1736         if (mmap_write_lock_killable(mm))
1737                 return -EINTR;
1738
1739         /*
1740          * This function tries to be smart and unmap shm segments that
1741          * were modified by partial mlock or munmap calls:
1742          * - It first determines the size of the shm segment that should be
1743          *   unmapped: It searches for a vma that is backed by shm and that
1744          *   started at address shmaddr. It records it's size and then unmaps
1745          *   it.
1746          * - Then it unmaps all shm vmas that started at shmaddr and that
1747          *   are within the initially determined size and that are from the
1748          *   same shm segment from which we determined the size.
1749          * Errors from do_munmap are ignored: the function only fails if
1750          * it's called with invalid parameters or if it's called to unmap
1751          * a part of a vma. Both calls in this function are for full vmas,
1752          * the parameters are directly copied from the vma itself and always
1753          * valid - therefore do_munmap cannot fail. (famous last words?)
1754          */
1755         /*
1756          * If it had been mremap()'d, the starting address would not
1757          * match the usual checks anyway. So assume all vma's are
1758          * above the starting address given.
1759          */
1760         vma = find_vma(mm, addr);
1761
1762 #ifdef CONFIG_MMU
1763         while (vma) {
1764                 next = vma->vm_next;
1765
1766                 /*
1767                  * Check if the starting address would match, i.e. it's
1768                  * a fragment created by mprotect() and/or munmap(), or it
1769                  * otherwise it starts at this address with no hassles.
1770                  */
1771                 if ((vma->vm_ops == &shm_vm_ops) &&
1772                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1773
1774                         /*
1775                          * Record the file of the shm segment being
1776                          * unmapped.  With mremap(), someone could place
1777                          * page from another segment but with equal offsets
1778                          * in the range we are unmapping.
1779                          */
1780                         file = vma->vm_file;
1781                         size = i_size_read(file_inode(vma->vm_file));
1782                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1783                         /*
1784                          * We discovered the size of the shm segment, so
1785                          * break out of here and fall through to the next
1786                          * loop that uses the size information to stop
1787                          * searching for matching vma's.
1788                          */
1789                         retval = 0;
1790                         vma = next;
1791                         break;
1792                 }
1793                 vma = next;
1794         }
1795
1796         /*
1797          * We need look no further than the maximum address a fragment
1798          * could possibly have landed at. Also cast things to loff_t to
1799          * prevent overflows and make comparisons vs. equal-width types.
1800          */
1801         size = PAGE_ALIGN(size);
1802         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1803                 next = vma->vm_next;
1804
1805                 /* finding a matching vma now does not alter retval */
1806                 if ((vma->vm_ops == &shm_vm_ops) &&
1807                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1808                     (vma->vm_file == file))
1809                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1810                 vma = next;
1811         }
1812
1813 #else   /* CONFIG_MMU */
1814         /* under NOMMU conditions, the exact address to be destroyed must be
1815          * given
1816          */
1817         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1818                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1819                 retval = 0;
1820         }
1821
1822 #endif
1823
1824         mmap_write_unlock(mm);
1825         return retval;
1826 }
1827
1828 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1829 {
1830         return ksys_shmdt(shmaddr);
1831 }
1832
1833 #ifdef CONFIG_PROC_FS
1834 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1835 {
1836         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1837         struct user_namespace *user_ns = seq_user_ns(s);
1838         struct kern_ipc_perm *ipcp = it;
1839         struct shmid_kernel *shp;
1840         unsigned long rss = 0, swp = 0;
1841
1842         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1843         shm_add_rss_swap(shp, &rss, &swp);
1844
1845 #if BITS_PER_LONG <= 32
1846 #define SIZE_SPEC "%10lu"
1847 #else
1848 #define SIZE_SPEC "%21lu"
1849 #endif
1850
1851         seq_printf(s,
1852                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1853                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1854                    SIZE_SPEC " " SIZE_SPEC "\n",
1855                    shp->shm_perm.key,
1856                    shp->shm_perm.id,
1857                    shp->shm_perm.mode,
1858                    shp->shm_segsz,
1859                    pid_nr_ns(shp->shm_cprid, pid_ns),
1860                    pid_nr_ns(shp->shm_lprid, pid_ns),
1861                    shp->shm_nattch,
1862                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1863                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1864                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1865                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1866                    shp->shm_atim,
1867                    shp->shm_dtim,
1868                    shp->shm_ctim,
1869                    rss * PAGE_SIZE,
1870                    swp * PAGE_SIZE);
1871
1872         return 0;
1873 }
1874 #endif