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