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 SYSCALL_DEFINE3(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 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 679 {
 680         switch (version) {
 681         case IPC_64:
 682                 return copy_to_user(buf, in, sizeof(*in));
 683         case IPC_OLD:
 684             {
 685                 struct shmid_ds out;
 686
 687                 memset(&out, 0, sizeof(out));
 688                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 689                 out.shm_segsz   = in->shm_segsz;
 690                 out.shm_atime   = in->shm_atime;
 691                 out.shm_dtime   = in->shm_dtime;
 692                 out.shm_ctime   = in->shm_ctime;
 693                 out.shm_cpid    = in->shm_cpid;
 694                 out.shm_lpid    = in->shm_lpid;
 695                 out.shm_nattch  = in->shm_nattch;
 696
 697                 return copy_to_user(buf, &out, sizeof(out));
 698             }
 699         default:
 700                 return -EINVAL;
 701         }
 702 }
 703
 704 static inline unsigned long
 705 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 706 {
 707         switch (version) {
 708         case IPC_64:
 709                 if (copy_from_user(out, buf, sizeof(*out)))
 710                         return -EFAULT;
 711                 return 0;
 712         case IPC_OLD:
 713             {
 714                 struct shmid_ds tbuf_old;
 715
 716                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 717                         return -EFAULT;
 718
 719                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 720                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 721                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 722
 723                 return 0;
 724             }
 725         default:
 726                 return -EINVAL;
 727         }
 728 }
 729
 730 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 731 {
 732         switch (version) {
 733         case IPC_64:
 734                 return copy_to_user(buf, in, sizeof(*in));
 735         case IPC_OLD:
 736             {
 737                 struct shminfo out;
 738
 739                 if (in->shmmax > INT_MAX)
 740                         out.shmmax = INT_MAX;
 741                 else
 742                         out.shmmax = (int)in->shmmax;
 743
 744                 out.shmmin      = in->shmmin;
 745                 out.shmmni      = in->shmmni;
 746                 out.shmseg      = in->shmseg;
 747                 out.shmall      = in->shmall;
 748
 749                 return copy_to_user(buf, &out, sizeof(out));
 750             }
 751         default:
 752                 return -EINVAL;
 753         }
 754 }
 755
 756 /*
 757  * Calculate and add used RSS and swap pages of a shm.
 758  * Called with shm_ids.rwsem held as a reader
 759  */
 760 static void shm_add_rss_swap(struct shmid_kernel *shp,
 761         unsigned long *rss_add, unsigned long *swp_add)
 762 {
 763         struct inode *inode;
 764
 765         inode = file_inode(shp->shm_file);
 766
 767         if (is_file_hugepages(shp->shm_file)) {
 768                 struct address_space *mapping = inode->i_mapping;
 769                 struct hstate *h = hstate_file(shp->shm_file);
 770                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 771         } else {
 772 #ifdef CONFIG_SHMEM
 773                 struct shmem_inode_info *info = SHMEM_I(inode);
 774
 775                 spin_lock_irq(&info->lock);
 776                 *rss_add += inode->i_mapping->nrpages;
 777                 *swp_add += info->swapped;
 778                 spin_unlock_irq(&info->lock);
 779 #else
 780                 *rss_add += inode->i_mapping->nrpages;
 781 #endif
 782         }
 783 }
 784
 785 /*
 786  * Called with shm_ids.rwsem held as a reader
 787  */
 788 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 789                 unsigned long *swp)
 790 {
 791         int next_id;
 792         int total, in_use;
 793
 794         *rss = 0;
 795         *swp = 0;
 796
 797         in_use = shm_ids(ns).in_use;
 798
 799         for (total = 0, next_id = 0; total < in_use; next_id++) {
 800                 struct kern_ipc_perm *ipc;
 801                 struct shmid_kernel *shp;
 802
 803                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 804                 if (ipc == NULL)
 805                         continue;
 806                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 807
 808                 shm_add_rss_swap(shp, rss, swp);
 809
 810                 total++;
 811         }
 812 }
 813
 814 /*
 815  * This function handles some shmctl commands which require the rwsem
 816  * to be held in write mode.
 817  * NOTE: no locks must be held, the rwsem is taken inside this function.
 818  */
 819 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 820                        struct shmid64_ds *shmid64)
 821 {
 822         struct kern_ipc_perm *ipcp;
 823         struct shmid_kernel *shp;
 824         int err;
 825
 826         down_write(&shm_ids(ns).rwsem);
 827         rcu_read_lock();
 828
 829         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 830                                       &shmid64->shm_perm, 0);
 831         if (IS_ERR(ipcp)) {
 832                 err = PTR_ERR(ipcp);
 833                 goto out_unlock1;
 834         }
 835
 836         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 837
 838         err = security_shm_shmctl(shp, cmd);
 839         if (err)
 840                 goto out_unlock1;
 841
 842         switch (cmd) {
 843         case IPC_RMID:
 844                 ipc_lock_object(&shp->shm_perm);
 845                 /* do_shm_rmid unlocks the ipc object and rcu */
 846                 do_shm_rmid(ns, ipcp);
 847                 goto out_up;
 848         case IPC_SET:
 849                 ipc_lock_object(&shp->shm_perm);
 850                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
 851                 if (err)
 852                         goto out_unlock0;
 853                 shp->shm_ctim = ktime_get_real_seconds();
 854                 break;
 855         default:
 856                 err = -EINVAL;
 857                 goto out_unlock1;
 858         }
 859
 860 out_unlock0:
 861         ipc_unlock_object(&shp->shm_perm);
 862 out_unlock1:
 863         rcu_read_unlock();
 864 out_up:
 865         up_write(&shm_ids(ns).rwsem);
 866         return err;
 867 }
 868
 869 static int shmctl_ipc_info(struct ipc_namespace *ns,
 870                            struct shminfo64 *shminfo)
 871 {
 872         int err = security_shm_shmctl(NULL, IPC_INFO);
 873         if (!err) {
 874                 memset(shminfo, 0, sizeof(*shminfo));
 875                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
 876                 shminfo->shmmax = ns->shm_ctlmax;
 877                 shminfo->shmall = ns->shm_ctlall;
 878                 shminfo->shmmin = SHMMIN;
 879                 down_read(&shm_ids(ns).rwsem);
 880                 err = ipc_get_maxid(&shm_ids(ns));
 881                 up_read(&shm_ids(ns).rwsem);
 882                 if (err < 0)
 883                         err = 0;
 884         }
 885         return err;
 886 }
 887
 888 static int shmctl_shm_info(struct ipc_namespace *ns,
 889                            struct shm_info *shm_info)
 890 {
 891         int err = security_shm_shmctl(NULL, SHM_INFO);
 892         if (!err) {
 893                 memset(shm_info, 0, sizeof(*shm_info));
 894                 down_read(&shm_ids(ns).rwsem);
 895                 shm_info->used_ids = shm_ids(ns).in_use;
 896                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
 897                 shm_info->shm_tot = ns->shm_tot;
 898                 shm_info->swap_attempts = 0;
 899                 shm_info->swap_successes = 0;
 900                 err = ipc_get_maxid(&shm_ids(ns));
 901                 up_read(&shm_ids(ns).rwsem);
 902                 if (err < 0)
 903                         err = 0;
 904         }
 905         return err;
 906 }
 907
 908 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
 909                         int cmd, struct shmid64_ds *tbuf)
 910 {
 911         struct shmid_kernel *shp;
 912         int id = 0;
 913         int err;
 914
 915         memset(tbuf, 0, sizeof(*tbuf));
 916
 917         rcu_read_lock();
 918         if (cmd == SHM_STAT) {
 919                 shp = shm_obtain_object(ns, shmid);
 920                 if (IS_ERR(shp)) {
 921                         err = PTR_ERR(shp);
 922                         goto out_unlock;
 923                 }
 924                 id = shp->shm_perm.id;
 925         } else {
 926                 shp = shm_obtain_object_check(ns, shmid);
 927                 if (IS_ERR(shp)) {
 928                         err = PTR_ERR(shp);
 929                         goto out_unlock;
 930                 }
 931         }
 932
 933         err = -EACCES;
 934         if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 935                 goto out_unlock;
 936
 937         err = security_shm_shmctl(shp, cmd);
 938         if (err)
 939                 goto out_unlock;
 940
 941         ipc_lock_object(&shp->shm_perm);
 942
 943         if (!ipc_valid_object(&shp->shm_perm)) {
 944                 ipc_unlock_object(&shp->shm_perm);
 945                 err = -EIDRM;
 946                 goto out_unlock;
 947         }
 948
 949         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
 950         tbuf->shm_segsz = shp->shm_segsz;
 951         tbuf->shm_atime = shp->shm_atim;
 952         tbuf->shm_dtime = shp->shm_dtim;
 953         tbuf->shm_ctime = shp->shm_ctim;
 954         tbuf->shm_cpid  = shp->shm_cprid;
 955         tbuf->shm_lpid  = shp->shm_lprid;
 956         tbuf->shm_nattch = shp->shm_nattch;
 957
 958         ipc_unlock_object(&shp->shm_perm);
 959         rcu_read_unlock();
 960         return id;
 961
 962 out_unlock:
 963         rcu_read_unlock();
 964         return err;
 965 }
 966
 967 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
 968 {
 969         struct shmid_kernel *shp;
 970         struct file *shm_file;
 971         int err;
 972
 973         rcu_read_lock();
 974         shp = shm_obtain_object_check(ns, shmid);
 975         if (IS_ERR(shp)) {
 976                 err = PTR_ERR(shp);
 977                 goto out_unlock1;
 978         }
 979
 980         audit_ipc_obj(&(shp->shm_perm));
 981         err = security_shm_shmctl(shp, cmd);
 982         if (err)
 983                 goto out_unlock1;
 984
 985         ipc_lock_object(&shp->shm_perm);
 986
 987         /* check if shm_destroy() is tearing down shp */
 988         if (!ipc_valid_object(&shp->shm_perm)) {
 989                 err = -EIDRM;
 990                 goto out_unlock0;
 991         }
 992
 993         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
 994                 kuid_t euid = current_euid();
 995
 996                 if (!uid_eq(euid, shp->shm_perm.uid) &&
 997                     !uid_eq(euid, shp->shm_perm.cuid)) {
 998                         err = -EPERM;
 999                         goto out_unlock0;
1000                 }
1001                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1002                         err = -EPERM;
1003                         goto out_unlock0;
1004                 }
1005         }
1006
1007         shm_file = shp->shm_file;
1008         if (is_file_hugepages(shm_file))
1009                 goto out_unlock0;
1010
1011         if (cmd == SHM_LOCK) {
1012                 struct user_struct *user = current_user();
1013
1014                 err = shmem_lock(shm_file, 1, user);
1015                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1016                         shp->shm_perm.mode |= SHM_LOCKED;
1017                         shp->mlock_user = user;
1018                 }
1019                 goto out_unlock0;
1020         }
1021
1022         /* SHM_UNLOCK */
1023         if (!(shp->shm_perm.mode & SHM_LOCKED))
1024                 goto out_unlock0;
1025         shmem_lock(shm_file, 0, shp->mlock_user);
1026         shp->shm_perm.mode &= ~SHM_LOCKED;
1027         shp->mlock_user = NULL;
1028         get_file(shm_file);
1029         ipc_unlock_object(&shp->shm_perm);
1030         rcu_read_unlock();
1031         shmem_unlock_mapping(shm_file->f_mapping);
1032
1033         fput(shm_file);
1034         return err;
1035
1036 out_unlock0:
1037         ipc_unlock_object(&shp->shm_perm);
1038 out_unlock1:
1039         rcu_read_unlock();
1040         return err;
1041 }
1042
1043 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1044 {
1045         int err, version;
1046         struct ipc_namespace *ns;
1047         struct shmid64_ds sem64;
1048
1049         if (cmd < 0 || shmid < 0)
1050                 return -EINVAL;
1051
1052         version = ipc_parse_version(&cmd);
1053         ns = current->nsproxy->ipc_ns;
1054
1055         switch (cmd) {
1056         case IPC_INFO: {
1057                 struct shminfo64 shminfo;
1058                 err = shmctl_ipc_info(ns, &shminfo);
1059                 if (err < 0)
1060                         return err;
1061                 if (copy_shminfo_to_user(buf, &shminfo, version))
1062                         err = -EFAULT;
1063                 return err;
1064         }
1065         case SHM_INFO: {
1066                 struct shm_info shm_info;
1067                 err = shmctl_shm_info(ns, &shm_info);
1068                 if (err < 0)
1069                         return err;
1070                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1071                         err = -EFAULT;
1072                 return err;
1073         }
1074         case SHM_STAT:
1075         case IPC_STAT: {
1076                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1077                 if (err < 0)
1078                         return err;
1079                 if (copy_shmid_to_user(buf, &sem64, version))
1080                         err = -EFAULT;
1081                 return err;
1082         }
1083         case IPC_SET:
1084                 if (copy_shmid_from_user(&sem64, buf, version))
1085                         return -EFAULT;
1086                 /* fallthru */
1087         case IPC_RMID:
1088                 return shmctl_down(ns, shmid, cmd, &sem64);
1089         case SHM_LOCK:
1090         case SHM_UNLOCK:
1091                 return shmctl_do_lock(ns, shmid, cmd);
1092         default:
1093                 return -EINVAL;
1094         }
1095 }
1096
1097 #ifdef CONFIG_COMPAT
1098
1099 struct compat_shmid_ds {
1100         struct compat_ipc_perm shm_perm;
1101         int shm_segsz;
1102         compat_time_t shm_atime;
1103         compat_time_t shm_dtime;
1104         compat_time_t shm_ctime;
1105         compat_ipc_pid_t shm_cpid;
1106         compat_ipc_pid_t shm_lpid;
1107         unsigned short shm_nattch;
1108         unsigned short shm_unused;
1109         compat_uptr_t shm_unused2;
1110         compat_uptr_t shm_unused3;
1111 };
1112
1113 struct compat_shminfo64 {
1114         compat_ulong_t shmmax;
1115         compat_ulong_t shmmin;
1116         compat_ulong_t shmmni;
1117         compat_ulong_t shmseg;
1118         compat_ulong_t shmall;
1119         compat_ulong_t __unused1;
1120         compat_ulong_t __unused2;
1121         compat_ulong_t __unused3;
1122         compat_ulong_t __unused4;
1123 };
1124
1125 struct compat_shm_info {
1126         compat_int_t used_ids;
1127         compat_ulong_t shm_tot, shm_rss, shm_swp;
1128         compat_ulong_t swap_attempts, swap_successes;
1129 };
1130
1131 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1132                                         int version)
1133 {
1134         if (in->shmmax > INT_MAX)
1135                 in->shmmax = INT_MAX;
1136         if (version == IPC_64) {
1137                 struct compat_shminfo64 info;
1138                 memset(&info, 0, sizeof(info));
1139                 info.shmmax = in->shmmax;
1140                 info.shmmin = in->shmmin;
1141                 info.shmmni = in->shmmni;
1142                 info.shmseg = in->shmseg;
1143                 info.shmall = in->shmall;
1144                 return copy_to_user(buf, &info, sizeof(info));
1145         } else {
1146                 struct shminfo info;
1147                 memset(&info, 0, sizeof(info));
1148                 info.shmmax = in->shmmax;
1149                 info.shmmin = in->shmmin;
1150                 info.shmmni = in->shmmni;
1151                 info.shmseg = in->shmseg;
1152                 info.shmall = in->shmall;
1153                 return copy_to_user(buf, &info, sizeof(info));
1154         }
1155 }
1156
1157 static int put_compat_shm_info(struct shm_info *ip,
1158                                 struct compat_shm_info __user *uip)
1159 {
1160         struct compat_shm_info info;
1161
1162         memset(&info, 0, sizeof(info));
1163         info.used_ids = ip->used_ids;
1164         info.shm_tot = ip->shm_tot;
1165         info.shm_rss = ip->shm_rss;
1166         info.shm_swp = ip->shm_swp;
1167         info.swap_attempts = ip->swap_attempts;
1168         info.swap_successes = ip->swap_successes;
1169         return copy_to_user(uip, &info, sizeof(info));
1170 }
1171
1172 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1173                                         int version)
1174 {
1175         if (version == IPC_64) {
1176                 struct compat_shmid64_ds v;
1177                 memset(&v, 0, sizeof(v));
1178                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1179                 v.shm_atime = in->shm_atime;
1180                 v.shm_dtime = in->shm_dtime;
1181                 v.shm_ctime = in->shm_ctime;
1182                 v.shm_segsz = in->shm_segsz;
1183                 v.shm_nattch = in->shm_nattch;
1184                 v.shm_cpid = in->shm_cpid;
1185                 v.shm_lpid = in->shm_lpid;
1186                 return copy_to_user(buf, &v, sizeof(v));
1187         } else {
1188                 struct compat_shmid_ds v;
1189                 memset(&v, 0, sizeof(v));
1190                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1191                 v.shm_perm.key = in->shm_perm.key;
1192                 v.shm_atime = in->shm_atime;
1193                 v.shm_dtime = in->shm_dtime;
1194                 v.shm_ctime = in->shm_ctime;
1195                 v.shm_segsz = in->shm_segsz;
1196                 v.shm_nattch = in->shm_nattch;
1197                 v.shm_cpid = in->shm_cpid;
1198                 v.shm_lpid = in->shm_lpid;
1199                 return copy_to_user(buf, &v, sizeof(v));
1200         }
1201 }
1202
1203 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1204                                         int version)
1205 {
1206         memset(out, 0, sizeof(*out));
1207         if (version == IPC_64) {
1208                 struct compat_shmid64_ds __user *p = buf;
1209                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1210         } else {
1211                 struct compat_shmid_ds __user *p = buf;
1212                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1213         }
1214 }
1215
1216 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1217 {
1218         struct ipc_namespace *ns;
1219         struct shmid64_ds sem64;
1220         int version = compat_ipc_parse_version(&cmd);
1221         int err;
1222
1223         ns = current->nsproxy->ipc_ns;
1224
1225         if (cmd < 0 || shmid < 0)
1226                 return -EINVAL;
1227
1228         switch (cmd) {
1229         case IPC_INFO: {
1230                 struct shminfo64 shminfo;
1231                 err = shmctl_ipc_info(ns, &shminfo);
1232                 if (err < 0)
1233                         return err;
1234                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1235                         err = -EFAULT;
1236                 return err;
1237         }
1238         case SHM_INFO: {
1239                 struct shm_info shm_info;
1240                 err = shmctl_shm_info(ns, &shm_info);
1241                 if (err < 0)
1242                         return err;
1243                 if (put_compat_shm_info(&shm_info, uptr))
1244                         err = -EFAULT;
1245                 return err;
1246         }
1247         case IPC_STAT:
1248         case SHM_STAT:
1249                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1250                 if (err < 0)
1251                         return err;
1252                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1253                         err = -EFAULT;
1254                 return err;
1255
1256         case IPC_SET:
1257                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1258                         return -EFAULT;
1259                 /* fallthru */
1260         case IPC_RMID:
1261                 return shmctl_down(ns, shmid, cmd, &sem64);
1262         case SHM_LOCK:
1263         case SHM_UNLOCK:
1264                 return shmctl_do_lock(ns, shmid, cmd);
1265                 break;
1266         default:
1267                 return -EINVAL;
1268         }
1269         return err;
1270 }
1271 #endif
1272
1273 /*
1274  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1275  *
1276  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1277  * "raddr" thing points to kernel space, and there has to be a wrapper around
1278  * this.
1279  */
1280 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1281               ulong *raddr, unsigned long shmlba)
1282 {
1283         struct shmid_kernel *shp;
1284         unsigned long addr = (unsigned long)shmaddr;
1285         unsigned long size;
1286         struct file *file;
1287         int    err;
1288         unsigned long flags = MAP_SHARED;
1289         unsigned long prot;
1290         int acc_mode;
1291         struct ipc_namespace *ns;
1292         struct shm_file_data *sfd;
1293         struct path path;
1294         fmode_t f_mode;
1295         unsigned long populate = 0;
1296
1297         err = -EINVAL;
1298         if (shmid < 0)
1299                 goto out;
1300
1301         if (addr) {
1302                 if (addr & (shmlba - 1)) {
1303                         /*
1304                          * Round down to the nearest multiple of shmlba.
1305                          * For sane do_mmap_pgoff() parameters, avoid
1306                          * round downs that trigger nil-page and MAP_FIXED.
1307                          */
1308                         if ((shmflg & SHM_RND) && addr >= shmlba)
1309                                 addr &= ~(shmlba - 1);
1310                         else
1311 #ifndef __ARCH_FORCE_SHMLBA
1312                                 if (addr & ~PAGE_MASK)
1313 #endif
1314                                         goto out;
1315                 }
1316
1317                 flags |= MAP_FIXED;
1318         } else if ((shmflg & SHM_REMAP))
1319                 goto out;
1320
1321         if (shmflg & SHM_RDONLY) {
1322                 prot = PROT_READ;
1323                 acc_mode = S_IRUGO;
1324                 f_mode = FMODE_READ;
1325         } else {
1326                 prot = PROT_READ | PROT_WRITE;
1327                 acc_mode = S_IRUGO | S_IWUGO;
1328                 f_mode = FMODE_READ | FMODE_WRITE;
1329         }
1330         if (shmflg & SHM_EXEC) {
1331                 prot |= PROT_EXEC;
1332                 acc_mode |= S_IXUGO;
1333         }
1334
1335         /*
1336          * We cannot rely on the fs check since SYSV IPC does have an
1337          * additional creator id...
1338          */
1339         ns = current->nsproxy->ipc_ns;
1340         rcu_read_lock();
1341         shp = shm_obtain_object_check(ns, shmid);
1342         if (IS_ERR(shp)) {
1343                 err = PTR_ERR(shp);
1344                 goto out_unlock;
1345         }
1346
1347         err = -EACCES;
1348         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1349                 goto out_unlock;
1350
1351         err = security_shm_shmat(shp, shmaddr, shmflg);
1352         if (err)
1353                 goto out_unlock;
1354
1355         ipc_lock_object(&shp->shm_perm);
1356
1357         /* check if shm_destroy() is tearing down shp */
1358         if (!ipc_valid_object(&shp->shm_perm)) {
1359                 ipc_unlock_object(&shp->shm_perm);
1360                 err = -EIDRM;
1361                 goto out_unlock;
1362         }
1363
1364         path = shp->shm_file->f_path;
1365         path_get(&path);
1366         shp->shm_nattch++;
1367         size = i_size_read(d_inode(path.dentry));
1368         ipc_unlock_object(&shp->shm_perm);
1369         rcu_read_unlock();
1370
1371         err = -ENOMEM;
1372         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1373         if (!sfd) {
1374                 path_put(&path);
1375                 goto out_nattch;
1376         }
1377
1378         file = alloc_file(&path, f_mode,
1379                           is_file_hugepages(shp->shm_file) ?
1380                                 &shm_file_operations_huge :
1381                                 &shm_file_operations);
1382         err = PTR_ERR(file);
1383         if (IS_ERR(file)) {
1384                 kfree(sfd);
1385                 path_put(&path);
1386                 goto out_nattch;
1387         }
1388
1389         file->private_data = sfd;
1390         file->f_mapping = shp->shm_file->f_mapping;
1391         sfd->id = shp->shm_perm.id;
1392         sfd->ns = get_ipc_ns(ns);
1393         sfd->file = shp->shm_file;
1394         sfd->vm_ops = NULL;
1395
1396         err = security_mmap_file(file, prot, flags);
1397         if (err)
1398                 goto out_fput;
1399
1400         if (down_write_killable(&current->mm->mmap_sem)) {
1401                 err = -EINTR;
1402                 goto out_fput;
1403         }
1404
1405         if (addr && !(shmflg & SHM_REMAP)) {
1406                 err = -EINVAL;
1407                 if (addr + size < addr)
1408                         goto invalid;
1409
1410                 if (find_vma_intersection(current->mm, addr, addr + size))
1411                         goto invalid;
1412         }
1413
1414         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1415         *raddr = addr;
1416         err = 0;
1417         if (IS_ERR_VALUE(addr))
1418                 err = (long)addr;
1419 invalid:
1420         up_write(&current->mm->mmap_sem);
1421         if (populate)
1422                 mm_populate(addr, populate);
1423
1424 out_fput:
1425         fput(file);
1426
1427 out_nattch:
1428         down_write(&shm_ids(ns).rwsem);
1429         shp = shm_lock(ns, shmid);
1430         shp->shm_nattch--;
1431         if (shm_may_destroy(ns, shp))
1432                 shm_destroy(ns, shp);
1433         else
1434                 shm_unlock(shp);
1435         up_write(&shm_ids(ns).rwsem);
1436         return err;
1437
1438 out_unlock:
1439         rcu_read_unlock();
1440 out:
1441         return err;
1442 }
1443
1444 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1445 {
1446         unsigned long ret;
1447         long err;
1448
1449         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1450         if (err)
1451                 return err;
1452         force_successful_syscall_return();
1453         return (long)ret;
1454 }
1455
1456 #ifdef CONFIG_COMPAT
1457
1458 #ifndef COMPAT_SHMLBA
1459 #define COMPAT_SHMLBA   SHMLBA
1460 #endif
1461
1462 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1463 {
1464         unsigned long ret;
1465         long err;
1466
1467         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1468         if (err)
1469                 return err;
1470         force_successful_syscall_return();
1471         return (long)ret;
1472 }
1473 #endif
1474
1475 /*
1476  * detach and kill segment if marked destroyed.
1477  * The work is done in shm_close.
1478  */
1479 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1480 {
1481         struct mm_struct *mm = current->mm;
1482         struct vm_area_struct *vma;
1483         unsigned long addr = (unsigned long)shmaddr;
1484         int retval = -EINVAL;
1485 #ifdef CONFIG_MMU
1486         loff_t size = 0;
1487         struct file *file;
1488         struct vm_area_struct *next;
1489 #endif
1490
1491         if (addr & ~PAGE_MASK)
1492                 return retval;
1493
1494         if (down_write_killable(&mm->mmap_sem))
1495                 return -EINTR;
1496
1497         /*
1498          * This function tries to be smart and unmap shm segments that
1499          * were modified by partial mlock or munmap calls:
1500          * - It first determines the size of the shm segment that should be
1501          *   unmapped: It searches for a vma that is backed by shm and that
1502          *   started at address shmaddr. It records it's size and then unmaps
1503          *   it.
1504          * - Then it unmaps all shm vmas that started at shmaddr and that
1505          *   are within the initially determined size and that are from the
1506          *   same shm segment from which we determined the size.
1507          * Errors from do_munmap are ignored: the function only fails if
1508          * it's called with invalid parameters or if it's called to unmap
1509          * a part of a vma. Both calls in this function are for full vmas,
1510          * the parameters are directly copied from the vma itself and always
1511          * valid - therefore do_munmap cannot fail. (famous last words?)
1512          */
1513         /*
1514          * If it had been mremap()'d, the starting address would not
1515          * match the usual checks anyway. So assume all vma's are
1516          * above the starting address given.
1517          */
1518         vma = find_vma(mm, addr);
1519
1520 #ifdef CONFIG_MMU
1521         while (vma) {
1522                 next = vma->vm_next;
1523
1524                 /*
1525                  * Check if the starting address would match, i.e. it's
1526                  * a fragment created by mprotect() and/or munmap(), or it
1527                  * otherwise it starts at this address with no hassles.
1528                  */
1529                 if ((vma->vm_ops == &shm_vm_ops) &&
1530                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1531
1532                         /*
1533                          * Record the file of the shm segment being
1534                          * unmapped.  With mremap(), someone could place
1535                          * page from another segment but with equal offsets
1536                          * in the range we are unmapping.
1537                          */
1538                         file = vma->vm_file;
1539                         size = i_size_read(file_inode(vma->vm_file));
1540                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1541                         /*
1542                          * We discovered the size of the shm segment, so
1543                          * break out of here and fall through to the next
1544                          * loop that uses the size information to stop
1545                          * searching for matching vma's.
1546                          */
1547                         retval = 0;
1548                         vma = next;
1549                         break;
1550                 }
1551                 vma = next;
1552         }
1553
1554         /*
1555          * We need look no further than the maximum address a fragment
1556          * could possibly have landed at. Also cast things to loff_t to
1557          * prevent overflows and make comparisons vs. equal-width types.
1558          */
1559         size = PAGE_ALIGN(size);
1560         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1561                 next = vma->vm_next;
1562
1563                 /* finding a matching vma now does not alter retval */
1564                 if ((vma->vm_ops == &shm_vm_ops) &&
1565                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1566                     (vma->vm_file == file))
1567                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1568                 vma = next;
1569         }
1570
1571 #else   /* CONFIG_MMU */
1572         /* under NOMMU conditions, the exact address to be destroyed must be
1573          * given
1574          */
1575         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1576                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1577                 retval = 0;
1578         }
1579
1580 #endif
1581
1582         up_write(&mm->mmap_sem);
1583         return retval;
1584 }
1585
1586 #ifdef CONFIG_PROC_FS
1587 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1588 {
1589         struct user_namespace *user_ns = seq_user_ns(s);
1590         struct kern_ipc_perm *ipcp = it;
1591         struct shmid_kernel *shp;
1592         unsigned long rss = 0, swp = 0;
1593
1594         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1595         shm_add_rss_swap(shp, &rss, &swp);
1596
1597 #if BITS_PER_LONG <= 32
1598 #define SIZE_SPEC "%10lu"
1599 #else
1600 #define SIZE_SPEC "%21lu"
1601 #endif
1602
1603         seq_printf(s,
1604                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1605                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1606                    SIZE_SPEC " " SIZE_SPEC "\n",
1607                    shp->shm_perm.key,
1608                    shp->shm_perm.id,
1609                    shp->shm_perm.mode,
1610                    shp->shm_segsz,
1611                    shp->shm_cprid,
1612                    shp->shm_lprid,
1613                    shp->shm_nattch,
1614                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1615                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1616                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1617                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1618                    shp->shm_atim,
1619                    shp->shm_dtim,
1620                    shp->shm_ctim,
1621                    rss * PAGE_SIZE,
1622                    swp * PAGE_SIZE);
1623
1624         return 0;
1625 }
1626 #endif