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