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