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 <[email protected]>
  10  * BIGMEM support, Andrea Arcangeli <[email protected]>
  11  * SMP thread shm, Jean-Luc Boyard <[email protected]>
  12  * HIGHMEM support, Ingo Molnar <[email protected]>
  13  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <[email protected]>
  14  * Shared /dev/zero support, Kanoj Sarcar <[email protected]>
  15  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <[email protected]>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <[email protected]>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <[email protected]>
  23  *
  24  * Better ipc lock (kern_ipc_perm.lock) handling
  25  * Davidlohr Bueso <[email protected]>, 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 int 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 #ifdef CONFIG_NUMA
 431 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 432 {
 433         struct file *file = vma->vm_file;
 434         struct shm_file_data *sfd = shm_file_data(file);
 435         int err = 0;
 436
 437         if (sfd->vm_ops->set_policy)
 438                 err = sfd->vm_ops->set_policy(vma, new);
 439         return err;
 440 }
 441
 442 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 443                                         unsigned long addr)
 444 {
 445         struct file *file = vma->vm_file;
 446         struct shm_file_data *sfd = shm_file_data(file);
 447         struct mempolicy *pol = NULL;
 448
 449         if (sfd->vm_ops->get_policy)
 450                 pol = sfd->vm_ops->get_policy(vma, addr);
 451         else if (vma->vm_policy)
 452                 pol = vma->vm_policy;
 453
 454         return pol;
 455 }
 456 #endif
 457
 458 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 459 {
 460         struct shm_file_data *sfd = shm_file_data(file);
 461         int ret;
 462
 463         /*
 464          * In case of remap_file_pages() emulation, the file can represent an
 465          * IPC ID that was removed, and possibly even reused by another shm
 466          * segment already.  Propagate this case as an error to caller.
 467          */
 468         ret = __shm_open(vma);
 469         if (ret)
 470                 return ret;
 471
 472         ret = call_mmap(sfd->file, vma);
 473         if (ret) {
 474                 shm_close(vma);
 475                 return ret;
 476         }
 477         sfd->vm_ops = vma->vm_ops;
 478 #ifdef CONFIG_MMU
 479         WARN_ON(!sfd->vm_ops->fault);
 480 #endif
 481         vma->vm_ops = &shm_vm_ops;
 482         return 0;
 483 }
 484
 485 static int shm_release(struct inode *ino, struct file *file)
 486 {
 487         struct shm_file_data *sfd = shm_file_data(file);
 488
 489         put_ipc_ns(sfd->ns);
 490         fput(sfd->file);
 491         shm_file_data(file) = NULL;
 492         kfree(sfd);
 493         return 0;
 494 }
 495
 496 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 497 {
 498         struct shm_file_data *sfd = shm_file_data(file);
 499
 500         if (!sfd->file->f_op->fsync)
 501                 return -EINVAL;
 502         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 503 }
 504
 505 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 506                           loff_t len)
 507 {
 508         struct shm_file_data *sfd = shm_file_data(file);
 509
 510         if (!sfd->file->f_op->fallocate)
 511                 return -EOPNOTSUPP;
 512         return sfd->file->f_op->fallocate(file, mode, offset, len);
 513 }
 514
 515 static unsigned long shm_get_unmapped_area(struct file *file,
 516         unsigned long addr, unsigned long len, unsigned long pgoff,
 517         unsigned long flags)
 518 {
 519         struct shm_file_data *sfd = shm_file_data(file);
 520
 521         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 522                                                 pgoff, flags);
 523 }
 524
 525 static const struct file_operations shm_file_operations = {
 526         .mmap           = shm_mmap,
 527         .fsync          = shm_fsync,
 528         .release        = shm_release,
 529         .get_unmapped_area      = shm_get_unmapped_area,
 530         .llseek         = noop_llseek,
 531         .fallocate      = shm_fallocate,
 532 };
 533
 534 /*
 535  * shm_file_operations_huge is now identical to shm_file_operations,
 536  * but we keep it distinct for the sake of is_file_shm_hugepages().
 537  */
 538 static const struct file_operations shm_file_operations_huge = {
 539         .mmap           = shm_mmap,
 540         .fsync          = shm_fsync,
 541         .release        = shm_release,
 542         .get_unmapped_area      = shm_get_unmapped_area,
 543         .llseek         = noop_llseek,
 544         .fallocate      = shm_fallocate,
 545 };
 546
 547 bool is_file_shm_hugepages(struct file *file)
 548 {
 549         return file->f_op == &shm_file_operations_huge;
 550 }
 551
 552 static const struct vm_operations_struct shm_vm_ops = {
 553         .open   = shm_open,     /* callback for a new vm-area open */
 554         .close  = shm_close,    /* callback for when the vm-area is released */
 555         .fault  = shm_fault,
 556         .split  = shm_split,
 557 #if defined(CONFIG_NUMA)
 558         .set_policy = shm_set_policy,
 559         .get_policy = shm_get_policy,
 560 #endif
 561 };
 562
 563 /**
 564  * newseg - Create a new shared memory segment
 565  * @ns: namespace
 566  * @params: ptr to the structure that contains key, size and shmflg
 567  *
 568  * Called with shm_ids.rwsem held as a writer.
 569  */
 570 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 571 {
 572         key_t key = params->key;
 573         int shmflg = params->flg;
 574         size_t size = params->u.size;
 575         int error;
 576         struct shmid_kernel *shp;
 577         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 578         struct file *file;
 579         char name[13];
 580         vm_flags_t acctflag = 0;
 581
 582         if (size < SHMMIN || size > ns->shm_ctlmax)
 583                 return -EINVAL;
 584
 585         if (numpages << PAGE_SHIFT < size)
 586                 return -ENOSPC;
 587
 588         if (ns->shm_tot + numpages < ns->shm_tot ||
 589                         ns->shm_tot + numpages > ns->shm_ctlall)
 590                 return -ENOSPC;
 591
 592         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
 593         if (unlikely(!shp))
 594                 return -ENOMEM;
 595
 596         shp->shm_perm.key = key;
 597         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 598         shp->mlock_user = NULL;
 599
 600         shp->shm_perm.security = NULL;
 601         error = security_shm_alloc(&shp->shm_perm);
 602         if (error) {
 603                 kvfree(shp);
 604                 return error;
 605         }
 606
 607         sprintf(name, "SYSV%08x", key);
 608         if (shmflg & SHM_HUGETLB) {
 609                 struct hstate *hs;
 610                 size_t hugesize;
 611
 612                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 613                 if (!hs) {
 614                         error = -EINVAL;
 615                         goto no_file;
 616                 }
 617                 hugesize = ALIGN(size, huge_page_size(hs));
 618
 619                 /* hugetlb_file_setup applies strict accounting */
 620                 if (shmflg & SHM_NORESERVE)
 621                         acctflag = VM_NORESERVE;
 622                 file = hugetlb_file_setup(name, hugesize, acctflag,
 623                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 624                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 625         } else {
 626                 /*
 627                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 628                  * if it's asked for.
 629                  */
 630                 if  ((shmflg & SHM_NORESERVE) &&
 631                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 632                         acctflag = VM_NORESERVE;
 633                 file = shmem_kernel_file_setup(name, size, acctflag);
 634         }
 635         error = PTR_ERR(file);
 636         if (IS_ERR(file))
 637                 goto no_file;
 638
 639         shp->shm_cprid = get_pid(task_tgid(current));
 640         shp->shm_lprid = NULL;
 641         shp->shm_atim = shp->shm_dtim = 0;
 642         shp->shm_ctim = ktime_get_real_seconds();
 643         shp->shm_segsz = size;
 644         shp->shm_nattch = 0;
 645         shp->shm_file = file;
 646         shp->shm_creator = current;
 647
 648         /* ipc_addid() locks shp upon success. */
 649         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 650         if (error < 0)
 651                 goto no_id;
 652
 653         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 654
 655         /*
 656          * shmid gets reported as "inode#" in /proc/pid/maps.
 657          * proc-ps tools use this. Changing this will break them.
 658          */
 659         file_inode(file)->i_ino = shp->shm_perm.id;
 660
 661         ns->shm_tot += numpages;
 662         error = shp->shm_perm.id;
 663
 664         ipc_unlock_object(&shp->shm_perm);
 665         rcu_read_unlock();
 666         return error;
 667
 668 no_id:
 669         ipc_update_pid(&shp->shm_cprid, NULL);
 670         ipc_update_pid(&shp->shm_lprid, NULL);
 671         if (is_file_hugepages(file) && shp->mlock_user)
 672                 user_shm_unlock(size, shp->mlock_user);
 673         fput(file);
 674 no_file:
 675         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 676         return error;
 677 }
 678
 679 /*
 680  * Called with shm_ids.rwsem and ipcp locked.
 681  */
 682 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 683                                 struct ipc_params *params)
 684 {
 685         struct shmid_kernel *shp;
 686
 687         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 688         if (shp->shm_segsz < params->u.size)
 689                 return -EINVAL;
 690
 691         return 0;
 692 }
 693
 694 long ksys_shmget(key_t key, size_t size, int shmflg)
 695 {
 696         struct ipc_namespace *ns;
 697         static const struct ipc_ops shm_ops = {
 698                 .getnew = newseg,
 699                 .associate = security_shm_associate,
 700                 .more_checks = shm_more_checks,
 701         };
 702         struct ipc_params shm_params;
 703
 704         ns = current->nsproxy->ipc_ns;
 705
 706         shm_params.key = key;
 707         shm_params.flg = shmflg;
 708         shm_params.u.size = size;
 709
 710         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 711 }
 712
 713 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 714 {
 715         return ksys_shmget(key, size, shmflg);
 716 }
 717
 718 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 719 {
 720         switch (version) {
 721         case IPC_64:
 722                 return copy_to_user(buf, in, sizeof(*in));
 723         case IPC_OLD:
 724             {
 725                 struct shmid_ds out;
 726
 727                 memset(&out, 0, sizeof(out));
 728                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 729                 out.shm_segsz   = in->shm_segsz;
 730                 out.shm_atime   = in->shm_atime;
 731                 out.shm_dtime   = in->shm_dtime;
 732                 out.shm_ctime   = in->shm_ctime;
 733                 out.shm_cpid    = in->shm_cpid;
 734                 out.shm_lpid    = in->shm_lpid;
 735                 out.shm_nattch  = in->shm_nattch;
 736
 737                 return copy_to_user(buf, &out, sizeof(out));
 738             }
 739         default:
 740                 return -EINVAL;
 741         }
 742 }
 743
 744 static inline unsigned long
 745 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 746 {
 747         switch (version) {
 748         case IPC_64:
 749                 if (copy_from_user(out, buf, sizeof(*out)))
 750                         return -EFAULT;
 751                 return 0;
 752         case IPC_OLD:
 753             {
 754                 struct shmid_ds tbuf_old;
 755
 756                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 757                         return -EFAULT;
 758
 759                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 760                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 761                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 762
 763                 return 0;
 764             }
 765         default:
 766                 return -EINVAL;
 767         }
 768 }
 769
 770 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 771 {
 772         switch (version) {
 773         case IPC_64:
 774                 return copy_to_user(buf, in, sizeof(*in));
 775         case IPC_OLD:
 776             {
 777                 struct shminfo out;
 778
 779                 if (in->shmmax > INT_MAX)
 780                         out.shmmax = INT_MAX;
 781                 else
 782                         out.shmmax = (int)in->shmmax;
 783
 784                 out.shmmin      = in->shmmin;
 785                 out.shmmni      = in->shmmni;
 786                 out.shmseg      = in->shmseg;
 787                 out.shmall      = in->shmall;
 788
 789                 return copy_to_user(buf, &out, sizeof(out));
 790             }
 791         default:
 792                 return -EINVAL;
 793         }
 794 }
 795
 796 /*
 797  * Calculate and add used RSS and swap pages of a shm.
 798  * Called with shm_ids.rwsem held as a reader
 799  */
 800 static void shm_add_rss_swap(struct shmid_kernel *shp,
 801         unsigned long *rss_add, unsigned long *swp_add)
 802 {
 803         struct inode *inode;
 804
 805         inode = file_inode(shp->shm_file);
 806
 807         if (is_file_hugepages(shp->shm_file)) {
 808                 struct address_space *mapping = inode->i_mapping;
 809                 struct hstate *h = hstate_file(shp->shm_file);
 810                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 811         } else {
 812 #ifdef CONFIG_SHMEM
 813                 struct shmem_inode_info *info = SHMEM_I(inode);
 814
 815                 spin_lock_irq(&info->lock);
 816                 *rss_add += inode->i_mapping->nrpages;
 817                 *swp_add += info->swapped;
 818                 spin_unlock_irq(&info->lock);
 819 #else
 820                 *rss_add += inode->i_mapping->nrpages;
 821 #endif
 822         }
 823 }
 824
 825 /*
 826  * Called with shm_ids.rwsem held as a reader
 827  */
 828 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 829                 unsigned long *swp)
 830 {
 831         int next_id;
 832         int total, in_use;
 833
 834         *rss = 0;
 835         *swp = 0;
 836
 837         in_use = shm_ids(ns).in_use;
 838
 839         for (total = 0, next_id = 0; total < in_use; next_id++) {
 840                 struct kern_ipc_perm *ipc;
 841                 struct shmid_kernel *shp;
 842
 843                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 844                 if (ipc == NULL)
 845                         continue;
 846                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 847
 848                 shm_add_rss_swap(shp, rss, swp);
 849
 850                 total++;
 851         }
 852 }
 853
 854 /*
 855  * This function handles some shmctl commands which require the rwsem
 856  * to be held in write mode.
 857  * NOTE: no locks must be held, the rwsem is taken inside this function.
 858  */
 859 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 860                        struct shmid64_ds *shmid64)
 861 {
 862         struct kern_ipc_perm *ipcp;
 863         struct shmid_kernel *shp;
 864         int err;
 865
 866         down_write(&shm_ids(ns).rwsem);
 867         rcu_read_lock();
 868
 869         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 870                                       &shmid64->shm_perm, 0);
 871         if (IS_ERR(ipcp)) {
 872                 err = PTR_ERR(ipcp);
 873                 goto out_unlock1;
 874         }
 875
 876         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 877
 878         err = security_shm_shmctl(&shp->shm_perm, cmd);
 879         if (err)
 880                 goto out_unlock1;
 881
 882         switch (cmd) {
 883         case IPC_RMID:
 884                 ipc_lock_object(&shp->shm_perm);
 885                 /* do_shm_rmid unlocks the ipc object and rcu */
 886                 do_shm_rmid(ns, ipcp);
 887                 goto out_up;
 888         case IPC_SET:
 889                 ipc_lock_object(&shp->shm_perm);
 890                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
 891                 if (err)
 892                         goto out_unlock0;
 893                 shp->shm_ctim = ktime_get_real_seconds();
 894                 break;
 895         default:
 896                 err = -EINVAL;
 897                 goto out_unlock1;
 898         }
 899
 900 out_unlock0:
 901         ipc_unlock_object(&shp->shm_perm);
 902 out_unlock1:
 903         rcu_read_unlock();
 904 out_up:
 905         up_write(&shm_ids(ns).rwsem);
 906         return err;
 907 }
 908
 909 static int shmctl_ipc_info(struct ipc_namespace *ns,
 910                            struct shminfo64 *shminfo)
 911 {
 912         int err = security_shm_shmctl(NULL, IPC_INFO);
 913         if (!err) {
 914                 memset(shminfo, 0, sizeof(*shminfo));
 915                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
 916                 shminfo->shmmax = ns->shm_ctlmax;
 917                 shminfo->shmall = ns->shm_ctlall;
 918                 shminfo->shmmin = SHMMIN;
 919                 down_read(&shm_ids(ns).rwsem);
 920                 err = ipc_get_maxid(&shm_ids(ns));
 921                 up_read(&shm_ids(ns).rwsem);
 922                 if (err < 0)
 923                         err = 0;
 924         }
 925         return err;
 926 }
 927
 928 static int shmctl_shm_info(struct ipc_namespace *ns,
 929                            struct shm_info *shm_info)
 930 {
 931         int err = security_shm_shmctl(NULL, SHM_INFO);
 932         if (!err) {
 933                 memset(shm_info, 0, sizeof(*shm_info));
 934                 down_read(&shm_ids(ns).rwsem);
 935                 shm_info->used_ids = shm_ids(ns).in_use;
 936                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
 937                 shm_info->shm_tot = ns->shm_tot;
 938                 shm_info->swap_attempts = 0;
 939                 shm_info->swap_successes = 0;
 940                 err = ipc_get_maxid(&shm_ids(ns));
 941                 up_read(&shm_ids(ns).rwsem);
 942                 if (err < 0)
 943                         err = 0;
 944         }
 945         return err;
 946 }
 947
 948 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
 949                         int cmd, struct shmid64_ds *tbuf)
 950 {
 951         struct shmid_kernel *shp;
 952         int id = 0;
 953         int err;
 954
 955         memset(tbuf, 0, sizeof(*tbuf));
 956
 957         rcu_read_lock();
 958         if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
 959                 shp = shm_obtain_object(ns, shmid);
 960                 if (IS_ERR(shp)) {
 961                         err = PTR_ERR(shp);
 962                         goto out_unlock;
 963                 }
 964                 id = shp->shm_perm.id;
 965         } else { /* IPC_STAT */
 966                 shp = shm_obtain_object_check(ns, shmid);
 967                 if (IS_ERR(shp)) {
 968                         err = PTR_ERR(shp);
 969                         goto out_unlock;
 970                 }
 971         }
 972
 973         /*
 974          * Semantically SHM_STAT_ANY ought to be identical to
 975          * that functionality provided by the /proc/sysvipc/
 976          * interface. As such, only audit these calls and
 977          * do not do traditional S_IRUGO permission checks on
 978          * the ipc object.
 979          */
 980         if (cmd == SHM_STAT_ANY)
 981                 audit_ipc_obj(&shp->shm_perm);
 982         else {
 983                 err = -EACCES;
 984                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 985                         goto out_unlock;
 986         }
 987
 988         err = security_shm_shmctl(&shp->shm_perm, cmd);
 989         if (err)
 990                 goto out_unlock;
 991
 992         ipc_lock_object(&shp->shm_perm);
 993
 994         if (!ipc_valid_object(&shp->shm_perm)) {
 995                 ipc_unlock_object(&shp->shm_perm);
 996                 err = -EIDRM;
 997                 goto out_unlock;
 998         }
 999
1000         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1001         tbuf->shm_segsz = shp->shm_segsz;
1002         tbuf->shm_atime = shp->shm_atim;
1003         tbuf->shm_dtime = shp->shm_dtim;
1004         tbuf->shm_ctime = shp->shm_ctim;
1005         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
1006         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
1007         tbuf->shm_nattch = shp->shm_nattch;
1008
1009         ipc_unlock_object(&shp->shm_perm);
1010         rcu_read_unlock();
1011         return id;
1012
1013 out_unlock:
1014         rcu_read_unlock();
1015         return err;
1016 }
1017
1018 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1019 {
1020         struct shmid_kernel *shp;
1021         struct file *shm_file;
1022         int err;
1023
1024         rcu_read_lock();
1025         shp = shm_obtain_object_check(ns, shmid);
1026         if (IS_ERR(shp)) {
1027                 err = PTR_ERR(shp);
1028                 goto out_unlock1;
1029         }
1030
1031         audit_ipc_obj(&(shp->shm_perm));
1032         err = security_shm_shmctl(&shp->shm_perm, cmd);
1033         if (err)
1034                 goto out_unlock1;
1035
1036         ipc_lock_object(&shp->shm_perm);
1037
1038         /* check if shm_destroy() is tearing down shp */
1039         if (!ipc_valid_object(&shp->shm_perm)) {
1040                 err = -EIDRM;
1041                 goto out_unlock0;
1042         }
1043
1044         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1045                 kuid_t euid = current_euid();
1046
1047                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1048                     !uid_eq(euid, shp->shm_perm.cuid)) {
1049                         err = -EPERM;
1050                         goto out_unlock0;
1051                 }
1052                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1053                         err = -EPERM;
1054                         goto out_unlock0;
1055                 }
1056         }
1057
1058         shm_file = shp->shm_file;
1059         if (is_file_hugepages(shm_file))
1060                 goto out_unlock0;
1061
1062         if (cmd == SHM_LOCK) {
1063                 struct user_struct *user = current_user();
1064
1065                 err = shmem_lock(shm_file, 1, user);
1066                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1067                         shp->shm_perm.mode |= SHM_LOCKED;
1068                         shp->mlock_user = user;
1069                 }
1070                 goto out_unlock0;
1071         }
1072
1073         /* SHM_UNLOCK */
1074         if (!(shp->shm_perm.mode & SHM_LOCKED))
1075                 goto out_unlock0;
1076         shmem_lock(shm_file, 0, shp->mlock_user);
1077         shp->shm_perm.mode &= ~SHM_LOCKED;
1078         shp->mlock_user = NULL;
1079         get_file(shm_file);
1080         ipc_unlock_object(&shp->shm_perm);
1081         rcu_read_unlock();
1082         shmem_unlock_mapping(shm_file->f_mapping);
1083
1084         fput(shm_file);
1085         return err;
1086
1087 out_unlock0:
1088         ipc_unlock_object(&shp->shm_perm);
1089 out_unlock1:
1090         rcu_read_unlock();
1091         return err;
1092 }
1093
1094 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1095 {
1096         int err, version;
1097         struct ipc_namespace *ns;
1098         struct shmid64_ds sem64;
1099
1100         if (cmd < 0 || shmid < 0)
1101                 return -EINVAL;
1102
1103         version = ipc_parse_version(&cmd);
1104         ns = current->nsproxy->ipc_ns;
1105
1106         switch (cmd) {
1107         case IPC_INFO: {
1108                 struct shminfo64 shminfo;
1109                 err = shmctl_ipc_info(ns, &shminfo);
1110                 if (err < 0)
1111                         return err;
1112                 if (copy_shminfo_to_user(buf, &shminfo, version))
1113                         err = -EFAULT;
1114                 return err;
1115         }
1116         case SHM_INFO: {
1117                 struct shm_info shm_info;
1118                 err = shmctl_shm_info(ns, &shm_info);
1119                 if (err < 0)
1120                         return err;
1121                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1122                         err = -EFAULT;
1123                 return err;
1124         }
1125         case SHM_STAT:
1126         case SHM_STAT_ANY:
1127         case IPC_STAT: {
1128                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1129                 if (err < 0)
1130                         return err;
1131                 if (copy_shmid_to_user(buf, &sem64, version))
1132                         err = -EFAULT;
1133                 return err;
1134         }
1135         case IPC_SET:
1136                 if (copy_shmid_from_user(&sem64, buf, version))
1137                         return -EFAULT;
1138                 /* fallthru */
1139         case IPC_RMID:
1140                 return shmctl_down(ns, shmid, cmd, &sem64);
1141         case SHM_LOCK:
1142         case SHM_UNLOCK:
1143                 return shmctl_do_lock(ns, shmid, cmd);
1144         default:
1145                 return -EINVAL;
1146         }
1147 }
1148
1149 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1150 {
1151         return ksys_shmctl(shmid, cmd, buf);
1152 }
1153
1154 #ifdef CONFIG_COMPAT
1155
1156 struct compat_shmid_ds {
1157         struct compat_ipc_perm shm_perm;
1158         int shm_segsz;
1159         compat_time_t shm_atime;
1160         compat_time_t shm_dtime;
1161         compat_time_t shm_ctime;
1162         compat_ipc_pid_t shm_cpid;
1163         compat_ipc_pid_t shm_lpid;
1164         unsigned short shm_nattch;
1165         unsigned short shm_unused;
1166         compat_uptr_t shm_unused2;
1167         compat_uptr_t shm_unused3;
1168 };
1169
1170 struct compat_shminfo64 {
1171         compat_ulong_t shmmax;
1172         compat_ulong_t shmmin;
1173         compat_ulong_t shmmni;
1174         compat_ulong_t shmseg;
1175         compat_ulong_t shmall;
1176         compat_ulong_t __unused1;
1177         compat_ulong_t __unused2;
1178         compat_ulong_t __unused3;
1179         compat_ulong_t __unused4;
1180 };
1181
1182 struct compat_shm_info {
1183         compat_int_t used_ids;
1184         compat_ulong_t shm_tot, shm_rss, shm_swp;
1185         compat_ulong_t swap_attempts, swap_successes;
1186 };
1187
1188 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1189                                         int version)
1190 {
1191         if (in->shmmax > INT_MAX)
1192                 in->shmmax = INT_MAX;
1193         if (version == IPC_64) {
1194                 struct compat_shminfo64 info;
1195                 memset(&info, 0, sizeof(info));
1196                 info.shmmax = in->shmmax;
1197                 info.shmmin = in->shmmin;
1198                 info.shmmni = in->shmmni;
1199                 info.shmseg = in->shmseg;
1200                 info.shmall = in->shmall;
1201                 return copy_to_user(buf, &info, sizeof(info));
1202         } else {
1203                 struct shminfo info;
1204                 memset(&info, 0, sizeof(info));
1205                 info.shmmax = in->shmmax;
1206                 info.shmmin = in->shmmin;
1207                 info.shmmni = in->shmmni;
1208                 info.shmseg = in->shmseg;
1209                 info.shmall = in->shmall;
1210                 return copy_to_user(buf, &info, sizeof(info));
1211         }
1212 }
1213
1214 static int put_compat_shm_info(struct shm_info *ip,
1215                                 struct compat_shm_info __user *uip)
1216 {
1217         struct compat_shm_info info;
1218
1219         memset(&info, 0, sizeof(info));
1220         info.used_ids = ip->used_ids;
1221         info.shm_tot = ip->shm_tot;
1222         info.shm_rss = ip->shm_rss;
1223         info.shm_swp = ip->shm_swp;
1224         info.swap_attempts = ip->swap_attempts;
1225         info.swap_successes = ip->swap_successes;
1226         return copy_to_user(uip, &info, sizeof(info));
1227 }
1228
1229 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1230                                         int version)
1231 {
1232         if (version == IPC_64) {
1233                 struct compat_shmid64_ds v;
1234                 memset(&v, 0, sizeof(v));
1235                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1236                 v.shm_atime = in->shm_atime;
1237                 v.shm_dtime = in->shm_dtime;
1238                 v.shm_ctime = in->shm_ctime;
1239                 v.shm_segsz = in->shm_segsz;
1240                 v.shm_nattch = in->shm_nattch;
1241                 v.shm_cpid = in->shm_cpid;
1242                 v.shm_lpid = in->shm_lpid;
1243                 return copy_to_user(buf, &v, sizeof(v));
1244         } else {
1245                 struct compat_shmid_ds v;
1246                 memset(&v, 0, sizeof(v));
1247                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1248                 v.shm_perm.key = in->shm_perm.key;
1249                 v.shm_atime = in->shm_atime;
1250                 v.shm_dtime = in->shm_dtime;
1251                 v.shm_ctime = in->shm_ctime;
1252                 v.shm_segsz = in->shm_segsz;
1253                 v.shm_nattch = in->shm_nattch;
1254                 v.shm_cpid = in->shm_cpid;
1255                 v.shm_lpid = in->shm_lpid;
1256                 return copy_to_user(buf, &v, sizeof(v));
1257         }
1258 }
1259
1260 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1261                                         int version)
1262 {
1263         memset(out, 0, sizeof(*out));
1264         if (version == IPC_64) {
1265                 struct compat_shmid64_ds __user *p = buf;
1266                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1267         } else {
1268                 struct compat_shmid_ds __user *p = buf;
1269                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1270         }
1271 }
1272
1273 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1274 {
1275         struct ipc_namespace *ns;
1276         struct shmid64_ds sem64;
1277         int version = compat_ipc_parse_version(&cmd);
1278         int err;
1279
1280         ns = current->nsproxy->ipc_ns;
1281
1282         if (cmd < 0 || shmid < 0)
1283                 return -EINVAL;
1284
1285         switch (cmd) {
1286         case IPC_INFO: {
1287                 struct shminfo64 shminfo;
1288                 err = shmctl_ipc_info(ns, &shminfo);
1289                 if (err < 0)
1290                         return err;
1291                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1292                         err = -EFAULT;
1293                 return err;
1294         }
1295         case SHM_INFO: {
1296                 struct shm_info shm_info;
1297                 err = shmctl_shm_info(ns, &shm_info);
1298                 if (err < 0)
1299                         return err;
1300                 if (put_compat_shm_info(&shm_info, uptr))
1301                         err = -EFAULT;
1302                 return err;
1303         }
1304         case IPC_STAT:
1305         case SHM_STAT_ANY:
1306         case SHM_STAT:
1307                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1308                 if (err < 0)
1309                         return err;
1310                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1311                         err = -EFAULT;
1312                 return err;
1313
1314         case IPC_SET:
1315                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1316                         return -EFAULT;
1317                 /* fallthru */
1318         case IPC_RMID:
1319                 return shmctl_down(ns, shmid, cmd, &sem64);
1320         case SHM_LOCK:
1321         case SHM_UNLOCK:
1322                 return shmctl_do_lock(ns, shmid, cmd);
1323                 break;
1324         default:
1325                 return -EINVAL;
1326         }
1327         return err;
1328 }
1329
1330 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1331 {
1332         return compat_ksys_shmctl(shmid, cmd, uptr);
1333 }
1334 #endif
1335
1336 /*
1337  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1338  *
1339  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1340  * "raddr" thing points to kernel space, and there has to be a wrapper around
1341  * this.
1342  */
1343 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1344               ulong *raddr, unsigned long shmlba)
1345 {
1346         struct shmid_kernel *shp;
1347         unsigned long addr = (unsigned long)shmaddr;
1348         unsigned long size;
1349         struct file *file;
1350         int    err;
1351         unsigned long flags = MAP_SHARED;
1352         unsigned long prot;
1353         int acc_mode;
1354         struct ipc_namespace *ns;
1355         struct shm_file_data *sfd;
1356         struct path path;
1357         fmode_t f_mode;
1358         unsigned long populate = 0;
1359
1360         err = -EINVAL;
1361         if (shmid < 0)
1362                 goto out;
1363
1364         if (addr) {
1365                 if (addr & (shmlba - 1)) {
1366                         /*
1367                          * Round down to the nearest multiple of shmlba.
1368                          * For sane do_mmap_pgoff() parameters, avoid
1369                          * round downs that trigger nil-page and MAP_FIXED.
1370                          */
1371                         if ((shmflg & SHM_RND) && addr >= shmlba)
1372                                 addr &= ~(shmlba - 1);
1373                         else
1374 #ifndef __ARCH_FORCE_SHMLBA
1375                                 if (addr & ~PAGE_MASK)
1376 #endif
1377                                         goto out;
1378                 }
1379
1380                 flags |= MAP_FIXED;
1381         } else if ((shmflg & SHM_REMAP))
1382                 goto out;
1383
1384         if (shmflg & SHM_RDONLY) {
1385                 prot = PROT_READ;
1386                 acc_mode = S_IRUGO;
1387                 f_mode = FMODE_READ;
1388         } else {
1389                 prot = PROT_READ | PROT_WRITE;
1390                 acc_mode = S_IRUGO | S_IWUGO;
1391                 f_mode = FMODE_READ | FMODE_WRITE;
1392         }
1393         if (shmflg & SHM_EXEC) {
1394                 prot |= PROT_EXEC;
1395                 acc_mode |= S_IXUGO;
1396         }
1397
1398         /*
1399          * We cannot rely on the fs check since SYSV IPC does have an
1400          * additional creator id...
1401          */
1402         ns = current->nsproxy->ipc_ns;
1403         rcu_read_lock();
1404         shp = shm_obtain_object_check(ns, shmid);
1405         if (IS_ERR(shp)) {
1406                 err = PTR_ERR(shp);
1407                 goto out_unlock;
1408         }
1409
1410         err = -EACCES;
1411         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1412                 goto out_unlock;
1413
1414         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1415         if (err)
1416                 goto out_unlock;
1417
1418         ipc_lock_object(&shp->shm_perm);
1419
1420         /* check if shm_destroy() is tearing down shp */
1421         if (!ipc_valid_object(&shp->shm_perm)) {
1422                 ipc_unlock_object(&shp->shm_perm);
1423                 err = -EIDRM;
1424                 goto out_unlock;
1425         }
1426
1427         path = shp->shm_file->f_path;
1428         path_get(&path);
1429         shp->shm_nattch++;
1430         size = i_size_read(d_inode(path.dentry));
1431         ipc_unlock_object(&shp->shm_perm);
1432         rcu_read_unlock();
1433
1434         err = -ENOMEM;
1435         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1436         if (!sfd) {
1437                 path_put(&path);
1438                 goto out_nattch;
1439         }
1440
1441         file = alloc_file(&path, f_mode,
1442                           is_file_hugepages(shp->shm_file) ?
1443                                 &shm_file_operations_huge :
1444                                 &shm_file_operations);
1445         err = PTR_ERR(file);
1446         if (IS_ERR(file)) {
1447                 kfree(sfd);
1448                 path_put(&path);
1449                 goto out_nattch;
1450         }
1451
1452         file->private_data = sfd;
1453         file->f_mapping = shp->shm_file->f_mapping;
1454         sfd->id = shp->shm_perm.id;
1455         sfd->ns = get_ipc_ns(ns);
1456         /*
1457          * We need to take a reference to the real shm file to prevent the
1458          * pointer from becoming stale in cases where the lifetime of the outer
1459          * file extends beyond that of the shm segment.  It's not usually
1460          * possible, but it can happen during remap_file_pages() emulation as
1461          * that unmaps the memory, then does ->mmap() via file reference only.
1462          * We'll deny the ->mmap() if the shm segment was since removed, but to
1463          * detect shm ID reuse we need to compare the file pointers.
1464          */
1465         sfd->file = get_file(shp->shm_file);
1466         sfd->vm_ops = NULL;
1467
1468         err = security_mmap_file(file, prot, flags);
1469         if (err)
1470                 goto out_fput;
1471
1472         if (down_write_killable(&current->mm->mmap_sem)) {
1473                 err = -EINTR;
1474                 goto out_fput;
1475         }
1476
1477         if (addr && !(shmflg & SHM_REMAP)) {
1478                 err = -EINVAL;
1479                 if (addr + size < addr)
1480                         goto invalid;
1481
1482                 if (find_vma_intersection(current->mm, addr, addr + size))
1483                         goto invalid;
1484         }
1485
1486         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1487         *raddr = addr;
1488         err = 0;
1489         if (IS_ERR_VALUE(addr))
1490                 err = (long)addr;
1491 invalid:
1492         up_write(&current->mm->mmap_sem);
1493         if (populate)
1494                 mm_populate(addr, populate);
1495
1496 out_fput:
1497         fput(file);
1498
1499 out_nattch:
1500         down_write(&shm_ids(ns).rwsem);
1501         shp = shm_lock(ns, shmid);
1502         shp->shm_nattch--;
1503         if (shm_may_destroy(ns, shp))
1504                 shm_destroy(ns, shp);
1505         else
1506                 shm_unlock(shp);
1507         up_write(&shm_ids(ns).rwsem);
1508         return err;
1509
1510 out_unlock:
1511         rcu_read_unlock();
1512 out:
1513         return err;
1514 }
1515
1516 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1517 {
1518         unsigned long ret;
1519         long err;
1520
1521         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1522         if (err)
1523                 return err;
1524         force_successful_syscall_return();
1525         return (long)ret;
1526 }
1527
1528 #ifdef CONFIG_COMPAT
1529
1530 #ifndef COMPAT_SHMLBA
1531 #define COMPAT_SHMLBA   SHMLBA
1532 #endif
1533
1534 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1535 {
1536         unsigned long ret;
1537         long err;
1538
1539         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1540         if (err)
1541                 return err;
1542         force_successful_syscall_return();
1543         return (long)ret;
1544 }
1545 #endif
1546
1547 /*
1548  * detach and kill segment if marked destroyed.
1549  * The work is done in shm_close.
1550  */
1551 long ksys_shmdt(char __user *shmaddr)
1552 {
1553         struct mm_struct *mm = current->mm;
1554         struct vm_area_struct *vma;
1555         unsigned long addr = (unsigned long)shmaddr;
1556         int retval = -EINVAL;
1557 #ifdef CONFIG_MMU
1558         loff_t size = 0;
1559         struct file *file;
1560         struct vm_area_struct *next;
1561 #endif
1562
1563         if (addr & ~PAGE_MASK)
1564                 return retval;
1565
1566         if (down_write_killable(&mm->mmap_sem))
1567                 return -EINTR;
1568
1569         /*
1570          * This function tries to be smart and unmap shm segments that
1571          * were modified by partial mlock or munmap calls:
1572          * - It first determines the size of the shm segment that should be
1573          *   unmapped: It searches for a vma that is backed by shm and that
1574          *   started at address shmaddr. It records it's size and then unmaps
1575          *   it.
1576          * - Then it unmaps all shm vmas that started at shmaddr and that
1577          *   are within the initially determined size and that are from the
1578          *   same shm segment from which we determined the size.
1579          * Errors from do_munmap are ignored: the function only fails if
1580          * it's called with invalid parameters or if it's called to unmap
1581          * a part of a vma. Both calls in this function are for full vmas,
1582          * the parameters are directly copied from the vma itself and always
1583          * valid - therefore do_munmap cannot fail. (famous last words?)
1584          */
1585         /*
1586          * If it had been mremap()'d, the starting address would not
1587          * match the usual checks anyway. So assume all vma's are
1588          * above the starting address given.
1589          */
1590         vma = find_vma(mm, addr);
1591
1592 #ifdef CONFIG_MMU
1593         while (vma) {
1594                 next = vma->vm_next;
1595
1596                 /*
1597                  * Check if the starting address would match, i.e. it's
1598                  * a fragment created by mprotect() and/or munmap(), or it
1599                  * otherwise it starts at this address with no hassles.
1600                  */
1601                 if ((vma->vm_ops == &shm_vm_ops) &&
1602                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1603
1604                         /*
1605                          * Record the file of the shm segment being
1606                          * unmapped.  With mremap(), someone could place
1607                          * page from another segment but with equal offsets
1608                          * in the range we are unmapping.
1609                          */
1610                         file = vma->vm_file;
1611                         size = i_size_read(file_inode(vma->vm_file));
1612                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1613                         /*
1614                          * We discovered the size of the shm segment, so
1615                          * break out of here and fall through to the next
1616                          * loop that uses the size information to stop
1617                          * searching for matching vma's.
1618                          */
1619                         retval = 0;
1620                         vma = next;
1621                         break;
1622                 }
1623                 vma = next;
1624         }
1625
1626         /*
1627          * We need look no further than the maximum address a fragment
1628          * could possibly have landed at. Also cast things to loff_t to
1629          * prevent overflows and make comparisons vs. equal-width types.
1630          */
1631         size = PAGE_ALIGN(size);
1632         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1633                 next = vma->vm_next;
1634
1635                 /* finding a matching vma now does not alter retval */
1636                 if ((vma->vm_ops == &shm_vm_ops) &&
1637                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1638                     (vma->vm_file == file))
1639                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1640                 vma = next;
1641         }
1642
1643 #else   /* CONFIG_MMU */
1644         /* under NOMMU conditions, the exact address to be destroyed must be
1645          * given
1646          */
1647         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1648                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1649                 retval = 0;
1650         }
1651
1652 #endif
1653
1654         up_write(&mm->mmap_sem);
1655         return retval;
1656 }
1657
1658 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1659 {
1660         return ksys_shmdt(shmaddr);
1661 }
1662
1663 #ifdef CONFIG_PROC_FS
1664 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1665 {
1666         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1667         struct user_namespace *user_ns = seq_user_ns(s);
1668         struct kern_ipc_perm *ipcp = it;
1669         struct shmid_kernel *shp;
1670         unsigned long rss = 0, swp = 0;
1671
1672         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1673         shm_add_rss_swap(shp, &rss, &swp);
1674
1675 #if BITS_PER_LONG <= 32
1676 #define SIZE_SPEC "%10lu"
1677 #else
1678 #define SIZE_SPEC "%21lu"
1679 #endif
1680
1681         seq_printf(s,
1682                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1683                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1684                    SIZE_SPEC " " SIZE_SPEC "\n",
1685                    shp->shm_perm.key,
1686                    shp->shm_perm.id,
1687                    shp->shm_perm.mode,
1688                    shp->shm_segsz,
1689                    pid_nr_ns(shp->shm_cprid, pid_ns),
1690                    pid_nr_ns(shp->shm_lprid, pid_ns),
1691                    shp->shm_nattch,
1692                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1693                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1694                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1695                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1696                    shp->shm_atim,
1697                    shp->shm_dtim,
1698                    shp->shm_ctim,
1699                    rss * PAGE_SIZE,
1700                    swp * PAGE_SIZE);
1701
1702         return 0;
1703 }
1704 #endif