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 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
 390 {
 391         struct file *file = vma->vm_file;
 392         struct shm_file_data *sfd = shm_file_data(file);
 393
 394         if (sfd->vm_ops && sfd->vm_ops->split)
 395                 return sfd->vm_ops->split(vma, addr);
 396
 397         return 0;
 398 }
 399
 400 #ifdef CONFIG_NUMA
 401 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 402 {
 403         struct file *file = vma->vm_file;
 404         struct shm_file_data *sfd = shm_file_data(file);
 405         int err = 0;
 406
 407         if (sfd->vm_ops->set_policy)
 408                 err = sfd->vm_ops->set_policy(vma, new);
 409         return err;
 410 }
 411
 412 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 413                                         unsigned long addr)
 414 {
 415         struct file *file = vma->vm_file;
 416         struct shm_file_data *sfd = shm_file_data(file);
 417         struct mempolicy *pol = NULL;
 418
 419         if (sfd->vm_ops->get_policy)
 420                 pol = sfd->vm_ops->get_policy(vma, addr);
 421         else if (vma->vm_policy)
 422                 pol = vma->vm_policy;
 423
 424         return pol;
 425 }
 426 #endif
 427
 428 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 429 {
 430         struct shm_file_data *sfd = shm_file_data(file);
 431         int ret;
 432
 433         /*
 434          * In case of remap_file_pages() emulation, the file can represent
 435          * removed IPC ID: propogate shm_lock() error to caller.
 436          */
 437         ret = __shm_open(vma);
 438         if (ret)
 439                 return ret;
 440
 441         ret = call_mmap(sfd->file, vma);
 442         if (ret) {
 443                 shm_close(vma);
 444                 return ret;
 445         }
 446         sfd->vm_ops = vma->vm_ops;
 447 #ifdef CONFIG_MMU
 448         WARN_ON(!sfd->vm_ops->fault);
 449 #endif
 450         vma->vm_ops = &shm_vm_ops;
 451         return 0;
 452 }
 453
 454 static int shm_release(struct inode *ino, struct file *file)
 455 {
 456         struct shm_file_data *sfd = shm_file_data(file);
 457
 458         put_ipc_ns(sfd->ns);
 459         shm_file_data(file) = NULL;
 460         kfree(sfd);
 461         return 0;
 462 }
 463
 464 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 465 {
 466         struct shm_file_data *sfd = shm_file_data(file);
 467
 468         if (!sfd->file->f_op->fsync)
 469                 return -EINVAL;
 470         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 471 }
 472
 473 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 474                           loff_t len)
 475 {
 476         struct shm_file_data *sfd = shm_file_data(file);
 477
 478         if (!sfd->file->f_op->fallocate)
 479                 return -EOPNOTSUPP;
 480         return sfd->file->f_op->fallocate(file, mode, offset, len);
 481 }
 482
 483 static unsigned long shm_get_unmapped_area(struct file *file,
 484         unsigned long addr, unsigned long len, unsigned long pgoff,
 485         unsigned long flags)
 486 {
 487         struct shm_file_data *sfd = shm_file_data(file);
 488
 489         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 490                                                 pgoff, flags);
 491 }
 492
 493 static const struct file_operations shm_file_operations = {
 494         .mmap           = shm_mmap,
 495         .fsync          = shm_fsync,
 496         .release        = shm_release,
 497         .get_unmapped_area      = shm_get_unmapped_area,
 498         .llseek         = noop_llseek,
 499         .fallocate      = shm_fallocate,
 500 };
 501
 502 /*
 503  * shm_file_operations_huge is now identical to shm_file_operations,
 504  * but we keep it distinct for the sake of is_file_shm_hugepages().
 505  */
 506 static const struct file_operations shm_file_operations_huge = {
 507         .mmap           = shm_mmap,
 508         .fsync          = shm_fsync,
 509         .release        = shm_release,
 510         .get_unmapped_area      = shm_get_unmapped_area,
 511         .llseek         = noop_llseek,
 512         .fallocate      = shm_fallocate,
 513 };
 514
 515 bool is_file_shm_hugepages(struct file *file)
 516 {
 517         return file->f_op == &shm_file_operations_huge;
 518 }
 519
 520 static const struct vm_operations_struct shm_vm_ops = {
 521         .open   = shm_open,     /* callback for a new vm-area open */
 522         .close  = shm_close,    /* callback for when the vm-area is released */
 523         .fault  = shm_fault,
 524         .split  = shm_split,
 525 #if defined(CONFIG_NUMA)
 526         .set_policy = shm_set_policy,
 527         .get_policy = shm_get_policy,
 528 #endif
 529 };
 530
 531 /**
 532  * newseg - Create a new shared memory segment
 533  * @ns: namespace
 534  * @params: ptr to the structure that contains key, size and shmflg
 535  *
 536  * Called with shm_ids.rwsem held as a writer.
 537  */
 538 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 539 {
 540         key_t key = params->key;
 541         int shmflg = params->flg;
 542         size_t size = params->u.size;
 543         int error;
 544         struct shmid_kernel *shp;
 545         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 546         struct file *file;
 547         char name[13];
 548         vm_flags_t acctflag = 0;
 549
 550         if (size < SHMMIN || size > ns->shm_ctlmax)
 551                 return -EINVAL;
 552
 553         if (numpages << PAGE_SHIFT < size)
 554                 return -ENOSPC;
 555
 556         if (ns->shm_tot + numpages < ns->shm_tot ||
 557                         ns->shm_tot + numpages > ns->shm_ctlall)
 558                 return -ENOSPC;
 559
 560         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
 561         if (unlikely(!shp))
 562                 return -ENOMEM;
 563
 564         shp->shm_perm.key = key;
 565         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 566         shp->mlock_user = NULL;
 567
 568         shp->shm_perm.security = NULL;
 569         error = security_shm_alloc(shp);
 570         if (error) {
 571                 kvfree(shp);
 572                 return error;
 573         }
 574
 575         sprintf(name, "SYSV%08x", key);
 576         if (shmflg & SHM_HUGETLB) {
 577                 struct hstate *hs;
 578                 size_t hugesize;
 579
 580                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 581                 if (!hs) {
 582                         error = -EINVAL;
 583                         goto no_file;
 584                 }
 585                 hugesize = ALIGN(size, huge_page_size(hs));
 586
 587                 /* hugetlb_file_setup applies strict accounting */
 588                 if (shmflg & SHM_NORESERVE)
 589                         acctflag = VM_NORESERVE;
 590                 file = hugetlb_file_setup(name, hugesize, acctflag,
 591                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 592                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 593         } else {
 594                 /*
 595                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 596                  * if it's asked for.
 597                  */
 598                 if  ((shmflg & SHM_NORESERVE) &&
 599                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 600                         acctflag = VM_NORESERVE;
 601                 file = shmem_kernel_file_setup(name, size, acctflag);
 602         }
 603         error = PTR_ERR(file);
 604         if (IS_ERR(file))
 605                 goto no_file;
 606
 607         shp->shm_cprid = task_tgid_vnr(current);
 608         shp->shm_lprid = 0;
 609         shp->shm_atim = shp->shm_dtim = 0;
 610         shp->shm_ctim = ktime_get_real_seconds();
 611         shp->shm_segsz = size;
 612         shp->shm_nattch = 0;
 613         shp->shm_file = file;
 614         shp->shm_creator = current;
 615
 616         /* ipc_addid() locks shp upon success. */
 617         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 618         if (error < 0)
 619                 goto no_id;
 620
 621         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 622
 623         /*
 624          * shmid gets reported as "inode#" in /proc/pid/maps.
 625          * proc-ps tools use this. Changing this will break them.
 626          */
 627         file_inode(file)->i_ino = shp->shm_perm.id;
 628
 629         ns->shm_tot += numpages;
 630         error = shp->shm_perm.id;
 631
 632         ipc_unlock_object(&shp->shm_perm);
 633         rcu_read_unlock();
 634         return error;
 635
 636 no_id:
 637         if (is_file_hugepages(file) && shp->mlock_user)
 638                 user_shm_unlock(size, shp->mlock_user);
 639         fput(file);
 640 no_file:
 641         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 642         return error;
 643 }
 644
 645 /*
 646  * Called with shm_ids.rwsem and ipcp locked.
 647  */
 648 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 649 {
 650         struct shmid_kernel *shp;
 651
 652         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 653         return security_shm_associate(shp, shmflg);
 654 }
 655
 656 /*
 657  * Called with shm_ids.rwsem and ipcp locked.
 658  */
 659 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 660                                 struct ipc_params *params)
 661 {
 662         struct shmid_kernel *shp;
 663
 664         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 665         if (shp->shm_segsz < params->u.size)
 666                 return -EINVAL;
 667
 668         return 0;
 669 }
 670
 671 long ksys_shmget(key_t key, size_t size, int shmflg)
 672 {
 673         struct ipc_namespace *ns;
 674         static const struct ipc_ops shm_ops = {
 675                 .getnew = newseg,
 676                 .associate = shm_security,
 677                 .more_checks = shm_more_checks,
 678         };
 679         struct ipc_params shm_params;
 680
 681         ns = current->nsproxy->ipc_ns;
 682
 683         shm_params.key = key;
 684         shm_params.flg = shmflg;
 685         shm_params.u.size = size;
 686
 687         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 688 }
 689
 690 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 691 {
 692         return ksys_shmget(key, size, shmflg);
 693 }
 694
 695 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 696 {
 697         switch (version) {
 698         case IPC_64:
 699                 return copy_to_user(buf, in, sizeof(*in));
 700         case IPC_OLD:
 701             {
 702                 struct shmid_ds out;
 703
 704                 memset(&out, 0, sizeof(out));
 705                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 706                 out.shm_segsz   = in->shm_segsz;
 707                 out.shm_atime   = in->shm_atime;
 708                 out.shm_dtime   = in->shm_dtime;
 709                 out.shm_ctime   = in->shm_ctime;
 710                 out.shm_cpid    = in->shm_cpid;
 711                 out.shm_lpid    = in->shm_lpid;
 712                 out.shm_nattch  = in->shm_nattch;
 713
 714                 return copy_to_user(buf, &out, sizeof(out));
 715             }
 716         default:
 717                 return -EINVAL;
 718         }
 719 }
 720
 721 static inline unsigned long
 722 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 723 {
 724         switch (version) {
 725         case IPC_64:
 726                 if (copy_from_user(out, buf, sizeof(*out)))
 727                         return -EFAULT;
 728                 return 0;
 729         case IPC_OLD:
 730             {
 731                 struct shmid_ds tbuf_old;
 732
 733                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 734                         return -EFAULT;
 735
 736                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 737                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 738                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 739
 740                 return 0;
 741             }
 742         default:
 743                 return -EINVAL;
 744         }
 745 }
 746
 747 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 748 {
 749         switch (version) {
 750         case IPC_64:
 751                 return copy_to_user(buf, in, sizeof(*in));
 752         case IPC_OLD:
 753             {
 754                 struct shminfo out;
 755
 756                 if (in->shmmax > INT_MAX)
 757                         out.shmmax = INT_MAX;
 758                 else
 759                         out.shmmax = (int)in->shmmax;
 760
 761                 out.shmmin      = in->shmmin;
 762                 out.shmmni      = in->shmmni;
 763                 out.shmseg      = in->shmseg;
 764                 out.shmall      = in->shmall;
 765
 766                 return copy_to_user(buf, &out, sizeof(out));
 767             }
 768         default:
 769                 return -EINVAL;
 770         }
 771 }
 772
 773 /*
 774  * Calculate and add used RSS and swap pages of a shm.
 775  * Called with shm_ids.rwsem held as a reader
 776  */
 777 static void shm_add_rss_swap(struct shmid_kernel *shp,
 778         unsigned long *rss_add, unsigned long *swp_add)
 779 {
 780         struct inode *inode;
 781
 782         inode = file_inode(shp->shm_file);
 783
 784         if (is_file_hugepages(shp->shm_file)) {
 785                 struct address_space *mapping = inode->i_mapping;
 786                 struct hstate *h = hstate_file(shp->shm_file);
 787                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 788         } else {
 789 #ifdef CONFIG_SHMEM
 790                 struct shmem_inode_info *info = SHMEM_I(inode);
 791
 792                 spin_lock_irq(&info->lock);
 793                 *rss_add += inode->i_mapping->nrpages;
 794                 *swp_add += info->swapped;
 795                 spin_unlock_irq(&info->lock);
 796 #else
 797                 *rss_add += inode->i_mapping->nrpages;
 798 #endif
 799         }
 800 }
 801
 802 /*
 803  * Called with shm_ids.rwsem held as a reader
 804  */
 805 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 806                 unsigned long *swp)
 807 {
 808         int next_id;
 809         int total, in_use;
 810
 811         *rss = 0;
 812         *swp = 0;
 813
 814         in_use = shm_ids(ns).in_use;
 815
 816         for (total = 0, next_id = 0; total < in_use; next_id++) {
 817                 struct kern_ipc_perm *ipc;
 818                 struct shmid_kernel *shp;
 819
 820                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 821                 if (ipc == NULL)
 822                         continue;
 823                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 824
 825                 shm_add_rss_swap(shp, rss, swp);
 826
 827                 total++;
 828         }
 829 }
 830
 831 /*
 832  * This function handles some shmctl commands which require the rwsem
 833  * to be held in write mode.
 834  * NOTE: no locks must be held, the rwsem is taken inside this function.
 835  */
 836 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 837                        struct shmid64_ds *shmid64)
 838 {
 839         struct kern_ipc_perm *ipcp;
 840         struct shmid_kernel *shp;
 841         int err;
 842
 843         down_write(&shm_ids(ns).rwsem);
 844         rcu_read_lock();
 845
 846         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 847                                       &shmid64->shm_perm, 0);
 848         if (IS_ERR(ipcp)) {
 849                 err = PTR_ERR(ipcp);
 850                 goto out_unlock1;
 851         }
 852
 853         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 854
 855         err = security_shm_shmctl(shp, cmd);
 856         if (err)
 857                 goto out_unlock1;
 858
 859         switch (cmd) {
 860         case IPC_RMID:
 861                 ipc_lock_object(&shp->shm_perm);
 862                 /* do_shm_rmid unlocks the ipc object and rcu */
 863                 do_shm_rmid(ns, ipcp);
 864                 goto out_up;
 865         case IPC_SET:
 866                 ipc_lock_object(&shp->shm_perm);
 867                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
 868                 if (err)
 869                         goto out_unlock0;
 870                 shp->shm_ctim = ktime_get_real_seconds();
 871                 break;
 872         default:
 873                 err = -EINVAL;
 874                 goto out_unlock1;
 875         }
 876
 877 out_unlock0:
 878         ipc_unlock_object(&shp->shm_perm);
 879 out_unlock1:
 880         rcu_read_unlock();
 881 out_up:
 882         up_write(&shm_ids(ns).rwsem);
 883         return err;
 884 }
 885
 886 static int shmctl_ipc_info(struct ipc_namespace *ns,
 887                            struct shminfo64 *shminfo)
 888 {
 889         int err = security_shm_shmctl(NULL, IPC_INFO);
 890         if (!err) {
 891                 memset(shminfo, 0, sizeof(*shminfo));
 892                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
 893                 shminfo->shmmax = ns->shm_ctlmax;
 894                 shminfo->shmall = ns->shm_ctlall;
 895                 shminfo->shmmin = SHMMIN;
 896                 down_read(&shm_ids(ns).rwsem);
 897                 err = ipc_get_maxid(&shm_ids(ns));
 898                 up_read(&shm_ids(ns).rwsem);
 899                 if (err < 0)
 900                         err = 0;
 901         }
 902         return err;
 903 }
 904
 905 static int shmctl_shm_info(struct ipc_namespace *ns,
 906                            struct shm_info *shm_info)
 907 {
 908         int err = security_shm_shmctl(NULL, SHM_INFO);
 909         if (!err) {
 910                 memset(shm_info, 0, sizeof(*shm_info));
 911                 down_read(&shm_ids(ns).rwsem);
 912                 shm_info->used_ids = shm_ids(ns).in_use;
 913                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
 914                 shm_info->shm_tot = ns->shm_tot;
 915                 shm_info->swap_attempts = 0;
 916                 shm_info->swap_successes = 0;
 917                 err = ipc_get_maxid(&shm_ids(ns));
 918                 up_read(&shm_ids(ns).rwsem);
 919                 if (err < 0)
 920                         err = 0;
 921         }
 922         return err;
 923 }
 924
 925 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
 926                         int cmd, struct shmid64_ds *tbuf)
 927 {
 928         struct shmid_kernel *shp;
 929         int id = 0;
 930         int err;
 931
 932         memset(tbuf, 0, sizeof(*tbuf));
 933
 934         rcu_read_lock();
 935         if (cmd == SHM_STAT) {
 936                 shp = shm_obtain_object(ns, shmid);
 937                 if (IS_ERR(shp)) {
 938                         err = PTR_ERR(shp);
 939                         goto out_unlock;
 940                 }
 941                 id = shp->shm_perm.id;
 942         } else {
 943                 shp = shm_obtain_object_check(ns, shmid);
 944                 if (IS_ERR(shp)) {
 945                         err = PTR_ERR(shp);
 946                         goto out_unlock;
 947                 }
 948         }
 949
 950         err = -EACCES;
 951         if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 952                 goto out_unlock;
 953
 954         err = security_shm_shmctl(shp, cmd);
 955         if (err)
 956                 goto out_unlock;
 957
 958         ipc_lock_object(&shp->shm_perm);
 959
 960         if (!ipc_valid_object(&shp->shm_perm)) {
 961                 ipc_unlock_object(&shp->shm_perm);
 962                 err = -EIDRM;
 963                 goto out_unlock;
 964         }
 965
 966         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
 967         tbuf->shm_segsz = shp->shm_segsz;
 968         tbuf->shm_atime = shp->shm_atim;
 969         tbuf->shm_dtime = shp->shm_dtim;
 970         tbuf->shm_ctime = shp->shm_ctim;
 971         tbuf->shm_cpid  = shp->shm_cprid;
 972         tbuf->shm_lpid  = shp->shm_lprid;
 973         tbuf->shm_nattch = shp->shm_nattch;
 974
 975         ipc_unlock_object(&shp->shm_perm);
 976         rcu_read_unlock();
 977         return id;
 978
 979 out_unlock:
 980         rcu_read_unlock();
 981         return err;
 982 }
 983
 984 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
 985 {
 986         struct shmid_kernel *shp;
 987         struct file *shm_file;
 988         int err;
 989
 990         rcu_read_lock();
 991         shp = shm_obtain_object_check(ns, shmid);
 992         if (IS_ERR(shp)) {
 993                 err = PTR_ERR(shp);
 994                 goto out_unlock1;
 995         }
 996
 997         audit_ipc_obj(&(shp->shm_perm));
 998         err = security_shm_shmctl(shp, cmd);
 999         if (err)
1000                 goto out_unlock1;
1001
1002         ipc_lock_object(&shp->shm_perm);
1003
1004         /* check if shm_destroy() is tearing down shp */
1005         if (!ipc_valid_object(&shp->shm_perm)) {
1006                 err = -EIDRM;
1007                 goto out_unlock0;
1008         }
1009
1010         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1011                 kuid_t euid = current_euid();
1012
1013                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1014                     !uid_eq(euid, shp->shm_perm.cuid)) {
1015                         err = -EPERM;
1016                         goto out_unlock0;
1017                 }
1018                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1019                         err = -EPERM;
1020                         goto out_unlock0;
1021                 }
1022         }
1023
1024         shm_file = shp->shm_file;
1025         if (is_file_hugepages(shm_file))
1026                 goto out_unlock0;
1027
1028         if (cmd == SHM_LOCK) {
1029                 struct user_struct *user = current_user();
1030
1031                 err = shmem_lock(shm_file, 1, user);
1032                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1033                         shp->shm_perm.mode |= SHM_LOCKED;
1034                         shp->mlock_user = user;
1035                 }
1036                 goto out_unlock0;
1037         }
1038
1039         /* SHM_UNLOCK */
1040         if (!(shp->shm_perm.mode & SHM_LOCKED))
1041                 goto out_unlock0;
1042         shmem_lock(shm_file, 0, shp->mlock_user);
1043         shp->shm_perm.mode &= ~SHM_LOCKED;
1044         shp->mlock_user = NULL;
1045         get_file(shm_file);
1046         ipc_unlock_object(&shp->shm_perm);
1047         rcu_read_unlock();
1048         shmem_unlock_mapping(shm_file->f_mapping);
1049
1050         fput(shm_file);
1051         return err;
1052
1053 out_unlock0:
1054         ipc_unlock_object(&shp->shm_perm);
1055 out_unlock1:
1056         rcu_read_unlock();
1057         return err;
1058 }
1059
1060 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1061 {
1062         int err, version;
1063         struct ipc_namespace *ns;
1064         struct shmid64_ds sem64;
1065
1066         if (cmd < 0 || shmid < 0)
1067                 return -EINVAL;
1068
1069         version = ipc_parse_version(&cmd);
1070         ns = current->nsproxy->ipc_ns;
1071
1072         switch (cmd) {
1073         case IPC_INFO: {
1074                 struct shminfo64 shminfo;
1075                 err = shmctl_ipc_info(ns, &shminfo);
1076                 if (err < 0)
1077                         return err;
1078                 if (copy_shminfo_to_user(buf, &shminfo, version))
1079                         err = -EFAULT;
1080                 return err;
1081         }
1082         case SHM_INFO: {
1083                 struct shm_info shm_info;
1084                 err = shmctl_shm_info(ns, &shm_info);
1085                 if (err < 0)
1086                         return err;
1087                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1088                         err = -EFAULT;
1089                 return err;
1090         }
1091         case SHM_STAT:
1092         case IPC_STAT: {
1093                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1094                 if (err < 0)
1095                         return err;
1096                 if (copy_shmid_to_user(buf, &sem64, version))
1097                         err = -EFAULT;
1098                 return err;
1099         }
1100         case IPC_SET:
1101                 if (copy_shmid_from_user(&sem64, buf, version))
1102                         return -EFAULT;
1103                 /* fallthru */
1104         case IPC_RMID:
1105                 return shmctl_down(ns, shmid, cmd, &sem64);
1106         case SHM_LOCK:
1107         case SHM_UNLOCK:
1108                 return shmctl_do_lock(ns, shmid, cmd);
1109         default:
1110                 return -EINVAL;
1111         }
1112 }
1113
1114 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1115 {
1116         return ksys_shmctl(shmid, cmd, buf);
1117 }
1118
1119 #ifdef CONFIG_COMPAT
1120
1121 struct compat_shmid_ds {
1122         struct compat_ipc_perm shm_perm;
1123         int shm_segsz;
1124         compat_time_t shm_atime;
1125         compat_time_t shm_dtime;
1126         compat_time_t shm_ctime;
1127         compat_ipc_pid_t shm_cpid;
1128         compat_ipc_pid_t shm_lpid;
1129         unsigned short shm_nattch;
1130         unsigned short shm_unused;
1131         compat_uptr_t shm_unused2;
1132         compat_uptr_t shm_unused3;
1133 };
1134
1135 struct compat_shminfo64 {
1136         compat_ulong_t shmmax;
1137         compat_ulong_t shmmin;
1138         compat_ulong_t shmmni;
1139         compat_ulong_t shmseg;
1140         compat_ulong_t shmall;
1141         compat_ulong_t __unused1;
1142         compat_ulong_t __unused2;
1143         compat_ulong_t __unused3;
1144         compat_ulong_t __unused4;
1145 };
1146
1147 struct compat_shm_info {
1148         compat_int_t used_ids;
1149         compat_ulong_t shm_tot, shm_rss, shm_swp;
1150         compat_ulong_t swap_attempts, swap_successes;
1151 };
1152
1153 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1154                                         int version)
1155 {
1156         if (in->shmmax > INT_MAX)
1157                 in->shmmax = INT_MAX;
1158         if (version == IPC_64) {
1159                 struct compat_shminfo64 info;
1160                 memset(&info, 0, sizeof(info));
1161                 info.shmmax = in->shmmax;
1162                 info.shmmin = in->shmmin;
1163                 info.shmmni = in->shmmni;
1164                 info.shmseg = in->shmseg;
1165                 info.shmall = in->shmall;
1166                 return copy_to_user(buf, &info, sizeof(info));
1167         } else {
1168                 struct shminfo info;
1169                 memset(&info, 0, sizeof(info));
1170                 info.shmmax = in->shmmax;
1171                 info.shmmin = in->shmmin;
1172                 info.shmmni = in->shmmni;
1173                 info.shmseg = in->shmseg;
1174                 info.shmall = in->shmall;
1175                 return copy_to_user(buf, &info, sizeof(info));
1176         }
1177 }
1178
1179 static int put_compat_shm_info(struct shm_info *ip,
1180                                 struct compat_shm_info __user *uip)
1181 {
1182         struct compat_shm_info info;
1183
1184         memset(&info, 0, sizeof(info));
1185         info.used_ids = ip->used_ids;
1186         info.shm_tot = ip->shm_tot;
1187         info.shm_rss = ip->shm_rss;
1188         info.shm_swp = ip->shm_swp;
1189         info.swap_attempts = ip->swap_attempts;
1190         info.swap_successes = ip->swap_successes;
1191         return copy_to_user(uip, &info, sizeof(info));
1192 }
1193
1194 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1195                                         int version)
1196 {
1197         if (version == IPC_64) {
1198                 struct compat_shmid64_ds v;
1199                 memset(&v, 0, sizeof(v));
1200                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1201                 v.shm_atime = in->shm_atime;
1202                 v.shm_dtime = in->shm_dtime;
1203                 v.shm_ctime = in->shm_ctime;
1204                 v.shm_segsz = in->shm_segsz;
1205                 v.shm_nattch = in->shm_nattch;
1206                 v.shm_cpid = in->shm_cpid;
1207                 v.shm_lpid = in->shm_lpid;
1208                 return copy_to_user(buf, &v, sizeof(v));
1209         } else {
1210                 struct compat_shmid_ds v;
1211                 memset(&v, 0, sizeof(v));
1212                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1213                 v.shm_perm.key = in->shm_perm.key;
1214                 v.shm_atime = in->shm_atime;
1215                 v.shm_dtime = in->shm_dtime;
1216                 v.shm_ctime = in->shm_ctime;
1217                 v.shm_segsz = in->shm_segsz;
1218                 v.shm_nattch = in->shm_nattch;
1219                 v.shm_cpid = in->shm_cpid;
1220                 v.shm_lpid = in->shm_lpid;
1221                 return copy_to_user(buf, &v, sizeof(v));
1222         }
1223 }
1224
1225 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1226                                         int version)
1227 {
1228         memset(out, 0, sizeof(*out));
1229         if (version == IPC_64) {
1230                 struct compat_shmid64_ds __user *p = buf;
1231                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1232         } else {
1233                 struct compat_shmid_ds __user *p = buf;
1234                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1235         }
1236 }
1237
1238 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1239 {
1240         struct ipc_namespace *ns;
1241         struct shmid64_ds sem64;
1242         int version = compat_ipc_parse_version(&cmd);
1243         int err;
1244
1245         ns = current->nsproxy->ipc_ns;
1246
1247         if (cmd < 0 || shmid < 0)
1248                 return -EINVAL;
1249
1250         switch (cmd) {
1251         case IPC_INFO: {
1252                 struct shminfo64 shminfo;
1253                 err = shmctl_ipc_info(ns, &shminfo);
1254                 if (err < 0)
1255                         return err;
1256                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1257                         err = -EFAULT;
1258                 return err;
1259         }
1260         case SHM_INFO: {
1261                 struct shm_info shm_info;
1262                 err = shmctl_shm_info(ns, &shm_info);
1263                 if (err < 0)
1264                         return err;
1265                 if (put_compat_shm_info(&shm_info, uptr))
1266                         err = -EFAULT;
1267                 return err;
1268         }
1269         case IPC_STAT:
1270         case SHM_STAT:
1271                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1272                 if (err < 0)
1273                         return err;
1274                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1275                         err = -EFAULT;
1276                 return err;
1277
1278         case IPC_SET:
1279                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1280                         return -EFAULT;
1281                 /* fallthru */
1282         case IPC_RMID:
1283                 return shmctl_down(ns, shmid, cmd, &sem64);
1284         case SHM_LOCK:
1285         case SHM_UNLOCK:
1286                 return shmctl_do_lock(ns, shmid, cmd);
1287                 break;
1288         default:
1289                 return -EINVAL;
1290         }
1291         return err;
1292 }
1293
1294 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1295 {
1296         return compat_ksys_shmctl(shmid, cmd, uptr);
1297 }
1298 #endif
1299
1300 /*
1301  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1302  *
1303  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1304  * "raddr" thing points to kernel space, and there has to be a wrapper around
1305  * this.
1306  */
1307 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1308               ulong *raddr, unsigned long shmlba)
1309 {
1310         struct shmid_kernel *shp;
1311         unsigned long addr = (unsigned long)shmaddr;
1312         unsigned long size;
1313         struct file *file;
1314         int    err;
1315         unsigned long flags = MAP_SHARED;
1316         unsigned long prot;
1317         int acc_mode;
1318         struct ipc_namespace *ns;
1319         struct shm_file_data *sfd;
1320         struct path path;
1321         fmode_t f_mode;
1322         unsigned long populate = 0;
1323
1324         err = -EINVAL;
1325         if (shmid < 0)
1326                 goto out;
1327
1328         if (addr) {
1329                 if (addr & (shmlba - 1)) {
1330                         /*
1331                          * Round down to the nearest multiple of shmlba.
1332                          * For sane do_mmap_pgoff() parameters, avoid
1333                          * round downs that trigger nil-page and MAP_FIXED.
1334                          */
1335                         if ((shmflg & SHM_RND) && addr >= shmlba)
1336                                 addr &= ~(shmlba - 1);
1337                         else
1338 #ifndef __ARCH_FORCE_SHMLBA
1339                                 if (addr & ~PAGE_MASK)
1340 #endif
1341                                         goto out;
1342                 }
1343
1344                 flags |= MAP_FIXED;
1345         } else if ((shmflg & SHM_REMAP))
1346                 goto out;
1347
1348         if (shmflg & SHM_RDONLY) {
1349                 prot = PROT_READ;
1350                 acc_mode = S_IRUGO;
1351                 f_mode = FMODE_READ;
1352         } else {
1353                 prot = PROT_READ | PROT_WRITE;
1354                 acc_mode = S_IRUGO | S_IWUGO;
1355                 f_mode = FMODE_READ | FMODE_WRITE;
1356         }
1357         if (shmflg & SHM_EXEC) {
1358                 prot |= PROT_EXEC;
1359                 acc_mode |= S_IXUGO;
1360         }
1361
1362         /*
1363          * We cannot rely on the fs check since SYSV IPC does have an
1364          * additional creator id...
1365          */
1366         ns = current->nsproxy->ipc_ns;
1367         rcu_read_lock();
1368         shp = shm_obtain_object_check(ns, shmid);
1369         if (IS_ERR(shp)) {
1370                 err = PTR_ERR(shp);
1371                 goto out_unlock;
1372         }
1373
1374         err = -EACCES;
1375         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1376                 goto out_unlock;
1377
1378         err = security_shm_shmat(shp, shmaddr, shmflg);
1379         if (err)
1380                 goto out_unlock;
1381
1382         ipc_lock_object(&shp->shm_perm);
1383
1384         /* check if shm_destroy() is tearing down shp */
1385         if (!ipc_valid_object(&shp->shm_perm)) {
1386                 ipc_unlock_object(&shp->shm_perm);
1387                 err = -EIDRM;
1388                 goto out_unlock;
1389         }
1390
1391         path = shp->shm_file->f_path;
1392         path_get(&path);
1393         shp->shm_nattch++;
1394         size = i_size_read(d_inode(path.dentry));
1395         ipc_unlock_object(&shp->shm_perm);
1396         rcu_read_unlock();
1397
1398         err = -ENOMEM;
1399         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1400         if (!sfd) {
1401                 path_put(&path);
1402                 goto out_nattch;
1403         }
1404
1405         file = alloc_file(&path, f_mode,
1406                           is_file_hugepages(shp->shm_file) ?
1407                                 &shm_file_operations_huge :
1408                                 &shm_file_operations);
1409         err = PTR_ERR(file);
1410         if (IS_ERR(file)) {
1411                 kfree(sfd);
1412                 path_put(&path);
1413                 goto out_nattch;
1414         }
1415
1416         file->private_data = sfd;
1417         file->f_mapping = shp->shm_file->f_mapping;
1418         sfd->id = shp->shm_perm.id;
1419         sfd->ns = get_ipc_ns(ns);
1420         sfd->file = shp->shm_file;
1421         sfd->vm_ops = NULL;
1422
1423         err = security_mmap_file(file, prot, flags);
1424         if (err)
1425                 goto out_fput;
1426
1427         if (down_write_killable(&current->mm->mmap_sem)) {
1428                 err = -EINTR;
1429                 goto out_fput;
1430         }
1431
1432         if (addr && !(shmflg & SHM_REMAP)) {
1433                 err = -EINVAL;
1434                 if (addr + size < addr)
1435                         goto invalid;
1436
1437                 if (find_vma_intersection(current->mm, addr, addr + size))
1438                         goto invalid;
1439         }
1440
1441         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1442         *raddr = addr;
1443         err = 0;
1444         if (IS_ERR_VALUE(addr))
1445                 err = (long)addr;
1446 invalid:
1447         up_write(&current->mm->mmap_sem);
1448         if (populate)
1449                 mm_populate(addr, populate);
1450
1451 out_fput:
1452         fput(file);
1453
1454 out_nattch:
1455         down_write(&shm_ids(ns).rwsem);
1456         shp = shm_lock(ns, shmid);
1457         shp->shm_nattch--;
1458         if (shm_may_destroy(ns, shp))
1459                 shm_destroy(ns, shp);
1460         else
1461                 shm_unlock(shp);
1462         up_write(&shm_ids(ns).rwsem);
1463         return err;
1464
1465 out_unlock:
1466         rcu_read_unlock();
1467 out:
1468         return err;
1469 }
1470
1471 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1472 {
1473         unsigned long ret;
1474         long err;
1475
1476         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1477         if (err)
1478                 return err;
1479         force_successful_syscall_return();
1480         return (long)ret;
1481 }
1482
1483 #ifdef CONFIG_COMPAT
1484
1485 #ifndef COMPAT_SHMLBA
1486 #define COMPAT_SHMLBA   SHMLBA
1487 #endif
1488
1489 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1490 {
1491         unsigned long ret;
1492         long err;
1493
1494         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1495         if (err)
1496                 return err;
1497         force_successful_syscall_return();
1498         return (long)ret;
1499 }
1500 #endif
1501
1502 /*
1503  * detach and kill segment if marked destroyed.
1504  * The work is done in shm_close.
1505  */
1506 long ksys_shmdt(char __user *shmaddr)
1507 {
1508         struct mm_struct *mm = current->mm;
1509         struct vm_area_struct *vma;
1510         unsigned long addr = (unsigned long)shmaddr;
1511         int retval = -EINVAL;
1512 #ifdef CONFIG_MMU
1513         loff_t size = 0;
1514         struct file *file;
1515         struct vm_area_struct *next;
1516 #endif
1517
1518         if (addr & ~PAGE_MASK)
1519                 return retval;
1520
1521         if (down_write_killable(&mm->mmap_sem))
1522                 return -EINTR;
1523
1524         /*
1525          * This function tries to be smart and unmap shm segments that
1526          * were modified by partial mlock or munmap calls:
1527          * - It first determines the size of the shm segment that should be
1528          *   unmapped: It searches for a vma that is backed by shm and that
1529          *   started at address shmaddr. It records it's size and then unmaps
1530          *   it.
1531          * - Then it unmaps all shm vmas that started at shmaddr and that
1532          *   are within the initially determined size and that are from the
1533          *   same shm segment from which we determined the size.
1534          * Errors from do_munmap are ignored: the function only fails if
1535          * it's called with invalid parameters or if it's called to unmap
1536          * a part of a vma. Both calls in this function are for full vmas,
1537          * the parameters are directly copied from the vma itself and always
1538          * valid - therefore do_munmap cannot fail. (famous last words?)
1539          */
1540         /*
1541          * If it had been mremap()'d, the starting address would not
1542          * match the usual checks anyway. So assume all vma's are
1543          * above the starting address given.
1544          */
1545         vma = find_vma(mm, addr);
1546
1547 #ifdef CONFIG_MMU
1548         while (vma) {
1549                 next = vma->vm_next;
1550
1551                 /*
1552                  * Check if the starting address would match, i.e. it's
1553                  * a fragment created by mprotect() and/or munmap(), or it
1554                  * otherwise it starts at this address with no hassles.
1555                  */
1556                 if ((vma->vm_ops == &shm_vm_ops) &&
1557                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1558
1559                         /*
1560                          * Record the file of the shm segment being
1561                          * unmapped.  With mremap(), someone could place
1562                          * page from another segment but with equal offsets
1563                          * in the range we are unmapping.
1564                          */
1565                         file = vma->vm_file;
1566                         size = i_size_read(file_inode(vma->vm_file));
1567                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1568                         /*
1569                          * We discovered the size of the shm segment, so
1570                          * break out of here and fall through to the next
1571                          * loop that uses the size information to stop
1572                          * searching for matching vma's.
1573                          */
1574                         retval = 0;
1575                         vma = next;
1576                         break;
1577                 }
1578                 vma = next;
1579         }
1580
1581         /*
1582          * We need look no further than the maximum address a fragment
1583          * could possibly have landed at. Also cast things to loff_t to
1584          * prevent overflows and make comparisons vs. equal-width types.
1585          */
1586         size = PAGE_ALIGN(size);
1587         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1588                 next = vma->vm_next;
1589
1590                 /* finding a matching vma now does not alter retval */
1591                 if ((vma->vm_ops == &shm_vm_ops) &&
1592                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1593                     (vma->vm_file == file))
1594                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1595                 vma = next;
1596         }
1597
1598 #else   /* CONFIG_MMU */
1599         /* under NOMMU conditions, the exact address to be destroyed must be
1600          * given
1601          */
1602         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1603                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1604                 retval = 0;
1605         }
1606
1607 #endif
1608
1609         up_write(&mm->mmap_sem);
1610         return retval;
1611 }
1612
1613 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1614 {
1615         return ksys_shmdt(shmaddr);
1616 }
1617
1618 #ifdef CONFIG_PROC_FS
1619 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1620 {
1621         struct user_namespace *user_ns = seq_user_ns(s);
1622         struct kern_ipc_perm *ipcp = it;
1623         struct shmid_kernel *shp;
1624         unsigned long rss = 0, swp = 0;
1625
1626         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1627         shm_add_rss_swap(shp, &rss, &swp);
1628
1629 #if BITS_PER_LONG <= 32
1630 #define SIZE_SPEC "%10lu"
1631 #else
1632 #define SIZE_SPEC "%21lu"
1633 #endif
1634
1635         seq_printf(s,
1636                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1637                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1638                    SIZE_SPEC " " SIZE_SPEC "\n",
1639                    shp->shm_perm.key,
1640                    shp->shm_perm.id,
1641                    shp->shm_perm.mode,
1642                    shp->shm_segsz,
1643                    shp->shm_cprid,
1644                    shp->shm_lprid,
1645                    shp->shm_nattch,
1646                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1647                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1648                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1649                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1650                    shp->shm_atim,
1651                    shp->shm_dtim,
1652                    shp->shm_ctim,
1653                    rss * PAGE_SIZE,
1654                    swp * PAGE_SIZE);
1655
1656         return 0;
1657 }
1658 #endif