ipc/shm.c

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *       Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15  *
  16  * support for audit of ipc object properties and permission changes
  17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  18  *
  19  * namespaces support
  20  * OpenVZ, SWsoft Inc.
  21  * Pavel Emelianov <xemul@openvz.org>
  22  */
  23
  24 #include <linux/slab.h>
  25 #include <linux/mm.h>
  26 #include <linux/hugetlb.h>
  27 #include <linux/shm.h>
  28 #include <linux/init.h>
  29 #include <linux/file.h>
  30 #include <linux/mman.h>
  31 #include <linux/shmem_fs.h>
  32 #include <linux/security.h>
  33 #include <linux/syscalls.h>
  34 #include <linux/audit.h>
  35 #include <linux/capability.h>
  36 #include <linux/ptrace.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/rwsem.h>
  39 #include <linux/nsproxy.h>
  40 #include <linux/mount.h>
  41 #include <linux/ipc_namespace.h>
  42
  43 #include <asm/uaccess.h>
  44
  45 #include "util.h"
  46
  47 struct shm_file_data {
  48         int id;
  49         struct ipc_namespace *ns;
  50         struct file *file;
  51         const struct vm_operations_struct *vm_ops;
  52 };
  53
  54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  55
  56 static const struct file_operations shm_file_operations;
  57 static const struct vm_operations_struct shm_vm_ops;
  58
  59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  60
  61 #define shm_unlock(shp)                 \
  62         ipc_unlock(&(shp)->shm_perm)
  63
  64 static int newseg(struct ipc_namespace *, struct ipc_params *);
  65 static void shm_open(struct vm_area_struct *vma);
  66 static void shm_close(struct vm_area_struct *vma);
  67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  68 #ifdef CONFIG_PROC_FS
  69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  70 #endif
  71
  72 void shm_init_ns(struct ipc_namespace *ns)
  73 {
  74         ns->shm_ctlmax = SHMMAX;
  75         ns->shm_ctlall = SHMALL;
  76         ns->shm_ctlmni = SHMMNI;
  77         ns->shm_rmid_forced = 0;
  78         ns->shm_tot = 0;
  79         ipc_init_ids(&shm_ids(ns));
  80 }
  81
  82 /*
  83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  84  * Only shm_ids.rw_mutex remains locked on exit.
  85  */
  86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  87 {
  88         struct shmid_kernel *shp;
  89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  90
  91         if (shp->shm_nattch){
  92                 shp->shm_perm.mode |= SHM_DEST;
  93                 /* Do not find it any more */
  94                 shp->shm_perm.key = IPC_PRIVATE;
  95                 shm_unlock(shp);
  96         } else
  97                 shm_destroy(ns, shp);
  98 }
  99
 100 #ifdef CONFIG_IPC_NS
 101 void shm_exit_ns(struct ipc_namespace *ns)
 102 {
 103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 104         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 105 }
 106 #endif
 107
 108 static int __init ipc_ns_init(void)
 109 {
 110         shm_init_ns(&init_ipc_ns);
 111         return 0;
 112 }
 113
 114 pure_initcall(ipc_ns_init);
 115
 116 void __init shm_init (void)
 117 {
 118         ipc_init_proc_interface("sysvipc/shm",
 119 #if BITS_PER_LONG <= 32
 120                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 121 #else
 122                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 123 #endif
 124                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 125 }
 126
 127 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 128 {
 129         struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
 130
 131         if (IS_ERR(ipcp))
 132                 return ERR_CAST(ipcp);
 133
 134         return container_of(ipcp, struct shmid_kernel, shm_perm);
 135 }
 136
 137 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 138 {
 139         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 140
 141         if (IS_ERR(ipcp))
 142                 return ERR_CAST(ipcp);
 143
 144         return container_of(ipcp, struct shmid_kernel, shm_perm);
 145 }
 146
 147 /*
 148  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 149  * is not necessarily held.
 150  */
 151 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 152 {
 153         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 154
 155         if (IS_ERR(ipcp))
 156                 return (struct shmid_kernel *)ipcp;
 157
 158         return container_of(ipcp, struct shmid_kernel, shm_perm);
 159 }
 160
 161 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 162 {
 163         rcu_read_lock();
 164         ipc_lock_object(&ipcp->shm_perm);
 165 }
 166
 167 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 168                                                 int id)
 169 {
 170         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 171
 172         if (IS_ERR(ipcp))
 173                 return (struct shmid_kernel *)ipcp;
 174
 175         return container_of(ipcp, struct shmid_kernel, shm_perm);
 176 }
 177
 178 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 179 {
 180         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 181 }
 182
 183
 184 /* This is called by fork, once for every shm attach. */
 185 static void shm_open(struct vm_area_struct *vma)
 186 {
 187         struct file *file = vma->vm_file;
 188         struct shm_file_data *sfd = shm_file_data(file);
 189         struct shmid_kernel *shp;
 190
 191         shp = shm_lock(sfd->ns, sfd->id);
 192         BUG_ON(IS_ERR(shp));
 193         shp->shm_atim = get_seconds();
 194         shp->shm_lprid = task_tgid_vnr(current);
 195         shp->shm_nattch++;
 196         shm_unlock(shp);
 197 }
 198
 199 /*
 200  * shm_destroy - free the struct shmid_kernel
 201  *
 202  * @ns: namespace
 203  * @shp: struct to free
 204  *
 205  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 206  * but returns with shp unlocked and freed.
 207  */
 208 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 209 {
 210         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 211         shm_rmid(ns, shp);
 212         shm_unlock(shp);
 213         if (!is_file_hugepages(shp->shm_file))
 214                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 215         else if (shp->mlock_user)
 216                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
 217                                                 shp->mlock_user);
 218         fput (shp->shm_file);
 219         security_shm_free(shp);
 220         ipc_rcu_putref(shp);
 221 }
 222
 223 /*
 224  * shm_may_destroy - identifies whether shm segment should be destroyed now
 225  *
 226  * Returns true if and only if there are no active users of the segment and
 227  * one of the following is true:
 228  *
 229  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 230  *
 231  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 232  */
 233 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 234 {
 235         return (shp->shm_nattch == 0) &&
 236                (ns->shm_rmid_forced ||
 237                 (shp->shm_perm.mode & SHM_DEST));
 238 }
 239
 240 /*
 241  * remove the attach descriptor vma.
 242  * free memory for segment if it is marked destroyed.
 243  * The descriptor has already been removed from the current->mm->mmap list
 244  * and will later be kfree()d.
 245  */
 246 static void shm_close(struct vm_area_struct *vma)
 247 {
 248         struct file * file = vma->vm_file;
 249         struct shm_file_data *sfd = shm_file_data(file);
 250         struct shmid_kernel *shp;
 251         struct ipc_namespace *ns = sfd->ns;
 252
 253         down_write(&shm_ids(ns).rw_mutex);
 254         /* remove from the list of attaches of the shm segment */
 255         shp = shm_lock(ns, sfd->id);
 256         BUG_ON(IS_ERR(shp));
 257         shp->shm_lprid = task_tgid_vnr(current);
 258         shp->shm_dtim = get_seconds();
 259         shp->shm_nattch--;
 260         if (shm_may_destroy(ns, shp))
 261                 shm_destroy(ns, shp);
 262         else
 263                 shm_unlock(shp);
 264         up_write(&shm_ids(ns).rw_mutex);
 265 }
 266
 267 /* Called with ns->shm_ids(ns).rw_mutex locked */
 268 static int shm_try_destroy_current(int id, void *p, void *data)
 269 {
 270         struct ipc_namespace *ns = data;
 271         struct kern_ipc_perm *ipcp = p;
 272         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 273
 274         if (shp->shm_creator != current)
 275                 return 0;
 276
 277         /*
 278          * Mark it as orphaned to destroy the segment when
 279          * kernel.shm_rmid_forced is changed.
 280          * It is noop if the following shm_may_destroy() returns true.
 281          */
 282         shp->shm_creator = NULL;
 283
 284         /*
 285          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
 286          * is not set, it shouldn't be deleted here.
 287          */
 288         if (!ns->shm_rmid_forced)
 289                 return 0;
 290
 291         if (shm_may_destroy(ns, shp)) {
 292                 shm_lock_by_ptr(shp);
 293                 shm_destroy(ns, shp);
 294         }
 295         return 0;
 296 }
 297
 298 /* Called with ns->shm_ids(ns).rw_mutex locked */
 299 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 300 {
 301         struct ipc_namespace *ns = data;
 302         struct kern_ipc_perm *ipcp = p;
 303         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 304
 305         /*
 306          * We want to destroy segments without users and with already
 307          * exit'ed originating process.
 308          *
 309          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
 310          */
 311         if (shp->shm_creator != NULL)
 312                 return 0;
 313
 314         if (shm_may_destroy(ns, shp)) {
 315                 shm_lock_by_ptr(shp);
 316                 shm_destroy(ns, shp);
 317         }
 318         return 0;
 319 }
 320
 321 void shm_destroy_orphaned(struct ipc_namespace *ns)
 322 {
 323         down_write(&shm_ids(ns).rw_mutex);
 324         if (shm_ids(ns).in_use)
 325                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 326         up_write(&shm_ids(ns).rw_mutex);
 327 }
 328
 329
 330 void exit_shm(struct task_struct *task)
 331 {
 332         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 333
 334         if (shm_ids(ns).in_use == 0)
 335                 return;
 336
 337         /* Destroy all already created segments, but not mapped yet */
 338         down_write(&shm_ids(ns).rw_mutex);
 339         if (shm_ids(ns).in_use)
 340                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
 341         up_write(&shm_ids(ns).rw_mutex);
 342 }
 343
 344 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 345 {
 346         struct file *file = vma->vm_file;
 347         struct shm_file_data *sfd = shm_file_data(file);
 348
 349         return sfd->vm_ops->fault(vma, vmf);
 350 }
 351
 352 #ifdef CONFIG_NUMA
 353 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 354 {
 355         struct file *file = vma->vm_file;
 356         struct shm_file_data *sfd = shm_file_data(file);
 357         int err = 0;
 358         if (sfd->vm_ops->set_policy)
 359                 err = sfd->vm_ops->set_policy(vma, new);
 360         return err;
 361 }
 362
 363 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 364                                         unsigned long addr)
 365 {
 366         struct file *file = vma->vm_file;
 367         struct shm_file_data *sfd = shm_file_data(file);
 368         struct mempolicy *pol = NULL;
 369
 370         if (sfd->vm_ops->get_policy)
 371                 pol = sfd->vm_ops->get_policy(vma, addr);
 372         else if (vma->vm_policy)
 373                 pol = vma->vm_policy;
 374
 375         return pol;
 376 }
 377 #endif
 378
 379 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 380 {
 381         struct shm_file_data *sfd = shm_file_data(file);
 382         int ret;
 383
 384         ret = sfd->file->f_op->mmap(sfd->file, vma);
 385         if (ret != 0)
 386                 return ret;
 387         sfd->vm_ops = vma->vm_ops;
 388 #ifdef CONFIG_MMU
 389         BUG_ON(!sfd->vm_ops->fault);
 390 #endif
 391         vma->vm_ops = &shm_vm_ops;
 392         shm_open(vma);
 393
 394         return ret;
 395 }
 396
 397 static int shm_release(struct inode *ino, struct file *file)
 398 {
 399         struct shm_file_data *sfd = shm_file_data(file);
 400
 401         put_ipc_ns(sfd->ns);
 402         shm_file_data(file) = NULL;
 403         kfree(sfd);
 404         return 0;
 405 }
 406
 407 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 408 {
 409         struct shm_file_data *sfd = shm_file_data(file);
 410
 411         if (!sfd->file->f_op->fsync)
 412                 return -EINVAL;
 413         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 414 }
 415
 416 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 417                           loff_t len)
 418 {
 419         struct shm_file_data *sfd = shm_file_data(file);
 420
 421         if (!sfd->file->f_op->fallocate)
 422                 return -EOPNOTSUPP;
 423         return sfd->file->f_op->fallocate(file, mode, offset, len);
 424 }
 425
 426 static unsigned long shm_get_unmapped_area(struct file *file,
 427         unsigned long addr, unsigned long len, unsigned long pgoff,
 428         unsigned long flags)
 429 {
 430         struct shm_file_data *sfd = shm_file_data(file);
 431         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 432                                                 pgoff, flags);
 433 }
 434
 435 static const struct file_operations shm_file_operations = {
 436         .mmap           = shm_mmap,
 437         .fsync          = shm_fsync,
 438         .release        = shm_release,
 439 #ifndef CONFIG_MMU
 440         .get_unmapped_area      = shm_get_unmapped_area,
 441 #endif
 442         .llseek         = noop_llseek,
 443         .fallocate      = shm_fallocate,
 444 };
 445
 446 static const struct file_operations shm_file_operations_huge = {
 447         .mmap           = shm_mmap,
 448         .fsync          = shm_fsync,
 449         .release        = shm_release,
 450         .get_unmapped_area      = shm_get_unmapped_area,
 451         .llseek         = noop_llseek,
 452         .fallocate      = shm_fallocate,
 453 };
 454
 455 int is_file_shm_hugepages(struct file *file)
 456 {
 457         return file->f_op == &shm_file_operations_huge;
 458 }
 459
 460 static const struct vm_operations_struct shm_vm_ops = {
 461         .open   = shm_open,     /* callback for a new vm-area open */
 462         .close  = shm_close,    /* callback for when the vm-area is released */
 463         .fault  = shm_fault,
 464 #if defined(CONFIG_NUMA)
 465         .set_policy = shm_set_policy,
 466         .get_policy = shm_get_policy,
 467 #endif
 468 };
 469
 470 /**
 471  * newseg - Create a new shared memory segment
 472  * @ns: namespace
 473  * @params: ptr to the structure that contains key, size and shmflg
 474  *
 475  * Called with shm_ids.rw_mutex held as a writer.
 476  */
 477
 478 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 479 {
 480         key_t key = params->key;
 481         int shmflg = params->flg;
 482         size_t size = params->u.size;
 483         int error;
 484         struct shmid_kernel *shp;
 485         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 486         struct file * file;
 487         char name[13];
 488         int id;
 489         vm_flags_t acctflag = 0;
 490
 491         if (size < SHMMIN || size > ns->shm_ctlmax)
 492                 return -EINVAL;
 493
 494         if (ns->shm_tot + numpages > ns->shm_ctlall)
 495                 return -ENOSPC;
 496
 497         shp = ipc_rcu_alloc(sizeof(*shp));
 498         if (!shp)
 499                 return -ENOMEM;
 500
 501         shp->shm_perm.key = key;
 502         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 503         shp->mlock_user = NULL;
 504
 505         shp->shm_perm.security = NULL;
 506         error = security_shm_alloc(shp);
 507         if (error) {
 508                 ipc_rcu_putref(shp);
 509                 return error;
 510         }
 511
 512         sprintf (name, "SYSV%08x", key);
 513         if (shmflg & SHM_HUGETLB) {
 514                 struct hstate *hs;
 515                 size_t hugesize;
 516
 517                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 518                 if (!hs) {
 519                         error = -EINVAL;
 520                         goto no_file;
 521                 }
 522                 hugesize = ALIGN(size, huge_page_size(hs));
 523
 524                 /* hugetlb_file_setup applies strict accounting */
 525                 if (shmflg & SHM_NORESERVE)
 526                         acctflag = VM_NORESERVE;
 527                 file = hugetlb_file_setup(name, hugesize, acctflag,
 528                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 529                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 530         } else {
 531                 /*
 532                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 533                  * if it's asked for.
 534                  */
 535                 if  ((shmflg & SHM_NORESERVE) &&
 536                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 537                         acctflag = VM_NORESERVE;
 538                 file = shmem_file_setup(name, size, acctflag);
 539         }
 540         error = PTR_ERR(file);
 541         if (IS_ERR(file))
 542                 goto no_file;
 543
 544         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 545         if (id < 0) {
 546                 error = id;
 547                 goto no_id;
 548         }
 549
 550         shp->shm_cprid = task_tgid_vnr(current);
 551         shp->shm_lprid = 0;
 552         shp->shm_atim = shp->shm_dtim = 0;
 553         shp->shm_ctim = get_seconds();
 554         shp->shm_segsz = size;
 555         shp->shm_nattch = 0;
 556         shp->shm_file = file;
 557         shp->shm_creator = current;
 558
 559         /*
 560          * shmid gets reported as "inode#" in /proc/pid/maps.
 561          * proc-ps tools use this. Changing this will break them.
 562          */
 563         file_inode(file)->i_ino = shp->shm_perm.id;
 564
 565         ns->shm_tot += numpages;
 566         error = shp->shm_perm.id;
 567
 568         ipc_unlock_object(&shp->shm_perm);
 569         rcu_read_unlock();
 570         return error;
 571
 572 no_id:
 573         if (is_file_hugepages(file) && shp->mlock_user)
 574                 user_shm_unlock(size, shp->mlock_user);
 575         fput(file);
 576 no_file:
 577         security_shm_free(shp);
 578         ipc_rcu_putref(shp);
 579         return error;
 580 }
 581
 582 /*
 583  * Called with shm_ids.rw_mutex and ipcp locked.
 584  */
 585 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 586 {
 587         struct shmid_kernel *shp;
 588
 589         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 590         return security_shm_associate(shp, shmflg);
 591 }
 592
 593 /*
 594  * Called with shm_ids.rw_mutex and ipcp locked.
 595  */
 596 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 597                                 struct ipc_params *params)
 598 {
 599         struct shmid_kernel *shp;
 600
 601         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 602         if (shp->shm_segsz < params->u.size)
 603                 return -EINVAL;
 604
 605         return 0;
 606 }
 607
 608 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 609 {
 610         struct ipc_namespace *ns;
 611         struct ipc_ops shm_ops;
 612         struct ipc_params shm_params;
 613
 614         ns = current->nsproxy->ipc_ns;
 615
 616         shm_ops.getnew = newseg;
 617         shm_ops.associate = shm_security;
 618         shm_ops.more_checks = shm_more_checks;
 619
 620         shm_params.key = key;
 621         shm_params.flg = shmflg;
 622         shm_params.u.size = size;
 623
 624         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 625 }
 626
 627 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 628 {
 629         switch(version) {
 630         case IPC_64:
 631                 return copy_to_user(buf, in, sizeof(*in));
 632         case IPC_OLD:
 633             {
 634                 struct shmid_ds out;
 635
 636                 memset(&out, 0, sizeof(out));
 637                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 638                 out.shm_segsz   = in->shm_segsz;
 639                 out.shm_atime   = in->shm_atime;
 640                 out.shm_dtime   = in->shm_dtime;
 641                 out.shm_ctime   = in->shm_ctime;
 642                 out.shm_cpid    = in->shm_cpid;
 643                 out.shm_lpid    = in->shm_lpid;
 644                 out.shm_nattch  = in->shm_nattch;
 645
 646                 return copy_to_user(buf, &out, sizeof(out));
 647             }
 648         default:
 649                 return -EINVAL;
 650         }
 651 }
 652
 653 static inline unsigned long
 654 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 655 {
 656         switch(version) {
 657         case IPC_64:
 658                 if (copy_from_user(out, buf, sizeof(*out)))
 659                         return -EFAULT;
 660                 return 0;
 661         case IPC_OLD:
 662             {
 663                 struct shmid_ds tbuf_old;
 664
 665                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 666                         return -EFAULT;
 667
 668                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 669                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 670                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 671
 672                 return 0;
 673             }
 674         default:
 675                 return -EINVAL;
 676         }
 677 }
 678
 679 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 680 {
 681         switch(version) {
 682         case IPC_64:
 683                 return copy_to_user(buf, in, sizeof(*in));
 684         case IPC_OLD:
 685             {
 686                 struct shminfo out;
 687
 688                 if(in->shmmax > INT_MAX)
 689                         out.shmmax = INT_MAX;
 690                 else
 691                         out.shmmax = (int)in->shmmax;
 692
 693                 out.shmmin      = in->shmmin;
 694                 out.shmmni      = in->shmmni;
 695                 out.shmseg      = in->shmseg;
 696                 out.shmall      = in->shmall;
 697
 698                 return copy_to_user(buf, &out, sizeof(out));
 699             }
 700         default:
 701                 return -EINVAL;
 702         }
 703 }
 704
 705 /*
 706  * Calculate and add used RSS and swap pages of a shm.
 707  * Called with shm_ids.rw_mutex held as a reader
 708  */
 709 static void shm_add_rss_swap(struct shmid_kernel *shp,
 710         unsigned long *rss_add, unsigned long *swp_add)
 711 {
 712         struct inode *inode;
 713
 714         inode = file_inode(shp->shm_file);
 715
 716         if (is_file_hugepages(shp->shm_file)) {
 717                 struct address_space *mapping = inode->i_mapping;
 718                 struct hstate *h = hstate_file(shp->shm_file);
 719                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 720         } else {
 721 #ifdef CONFIG_SHMEM
 722                 struct shmem_inode_info *info = SHMEM_I(inode);
 723                 spin_lock(&info->lock);
 724                 *rss_add += inode->i_mapping->nrpages;
 725                 *swp_add += info->swapped;
 726                 spin_unlock(&info->lock);
 727 #else
 728                 *rss_add += inode->i_mapping->nrpages;
 729 #endif
 730         }
 731 }
 732
 733 /*
 734  * Called with shm_ids.rw_mutex held as a reader
 735  */
 736 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 737                 unsigned long *swp)
 738 {
 739         int next_id;
 740         int total, in_use;
 741
 742         *rss = 0;
 743         *swp = 0;
 744
 745         in_use = shm_ids(ns).in_use;
 746
 747         for (total = 0, next_id = 0; total < in_use; next_id++) {
 748                 struct kern_ipc_perm *ipc;
 749                 struct shmid_kernel *shp;
 750
 751                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 752                 if (ipc == NULL)
 753                         continue;
 754                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 755
 756                 shm_add_rss_swap(shp, rss, swp);
 757
 758                 total++;
 759         }
 760 }
 761
 762 /*
 763  * This function handles some shmctl commands which require the rw_mutex
 764  * to be held in write mode.
 765  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 766  */
 767 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 768                        struct shmid_ds __user *buf, int version)
 769 {
 770         struct kern_ipc_perm *ipcp;
 771         struct shmid64_ds shmid64;
 772         struct shmid_kernel *shp;
 773         int err;
 774
 775         if (cmd == IPC_SET) {
 776                 if (copy_shmid_from_user(&shmid64, buf, version))
 777                         return -EFAULT;
 778         }
 779
 780         down_write(&shm_ids(ns).rw_mutex);
 781         rcu_read_lock();
 782
 783         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 784                                       &shmid64.shm_perm, 0);
 785         if (IS_ERR(ipcp)) {
 786                 err = PTR_ERR(ipcp);
 787                 goto out_unlock1;
 788         }
 789
 790         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 791
 792         err = security_shm_shmctl(shp, cmd);
 793         if (err)
 794                 goto out_unlock1;
 795
 796         switch (cmd) {
 797         case IPC_RMID:
 798                 ipc_lock_object(&shp->shm_perm);
 799                 /* do_shm_rmid unlocks the ipc object and rcu */
 800                 do_shm_rmid(ns, ipcp);
 801                 goto out_up;
 802         case IPC_SET:
 803                 ipc_lock_object(&shp->shm_perm);
 804                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
 805                 if (err)
 806                         goto out_unlock0;
 807                 shp->shm_ctim = get_seconds();
 808                 break;
 809         default:
 810                 err = -EINVAL;
 811                 goto out_unlock1;
 812         }
 813
 814 out_unlock0:
 815         ipc_unlock_object(&shp->shm_perm);
 816 out_unlock1:
 817         rcu_read_unlock();
 818 out_up:
 819         up_write(&shm_ids(ns).rw_mutex);
 820         return err;
 821 }
 822
 823 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 824 {
 825         struct shmid_kernel *shp;
 826         int err, version;
 827         struct ipc_namespace *ns;
 828
 829         if (cmd < 0 || shmid < 0) {
 830                 err = -EINVAL;
 831                 goto out;
 832         }
 833
 834         version = ipc_parse_version(&cmd);
 835         ns = current->nsproxy->ipc_ns;
 836
 837         switch (cmd) { /* replace with proc interface ? */
 838         case IPC_INFO:
 839         {
 840                 struct shminfo64 shminfo;
 841
 842                 err = security_shm_shmctl(NULL, cmd);
 843                 if (err)
 844                         return err;
 845
 846                 memset(&shminfo, 0, sizeof(shminfo));
 847                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 848                 shminfo.shmmax = ns->shm_ctlmax;
 849                 shminfo.shmall = ns->shm_ctlall;
 850
 851                 shminfo.shmmin = SHMMIN;
 852                 if(copy_shminfo_to_user (buf, &shminfo, version))
 853                         return -EFAULT;
 854
 855                 down_read(&shm_ids(ns).rw_mutex);
 856                 err = ipc_get_maxid(&shm_ids(ns));
 857                 up_read(&shm_ids(ns).rw_mutex);
 858
 859                 if(err<0)
 860                         err = 0;
 861                 goto out;
 862         }
 863         case SHM_INFO:
 864         {
 865                 struct shm_info shm_info;
 866
 867                 err = security_shm_shmctl(NULL, cmd);
 868                 if (err)
 869                         return err;
 870
 871                 memset(&shm_info, 0, sizeof(shm_info));
 872                 down_read(&shm_ids(ns).rw_mutex);
 873                 shm_info.used_ids = shm_ids(ns).in_use;
 874                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 875                 shm_info.shm_tot = ns->shm_tot;
 876                 shm_info.swap_attempts = 0;
 877                 shm_info.swap_successes = 0;
 878                 err = ipc_get_maxid(&shm_ids(ns));
 879                 up_read(&shm_ids(ns).rw_mutex);
 880                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 881                         err = -EFAULT;
 882                         goto out;
 883                 }
 884
 885                 err = err < 0 ? 0 : err;
 886                 goto out;
 887         }
 888         case SHM_STAT:
 889         case IPC_STAT:
 890         {
 891                 struct shmid64_ds tbuf;
 892                 int result;
 893
 894                 if (cmd == SHM_STAT) {
 895                         shp = shm_lock(ns, shmid);
 896                         if (IS_ERR(shp)) {
 897                                 err = PTR_ERR(shp);
 898                                 goto out;
 899                         }
 900                         result = shp->shm_perm.id;
 901                 } else {
 902                         shp = shm_lock_check(ns, shmid);
 903                         if (IS_ERR(shp)) {
 904                                 err = PTR_ERR(shp);
 905                                 goto out;
 906                         }
 907                         result = 0;
 908                 }
 909                 err = -EACCES;
 910                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 911                         goto out_unlock;
 912                 err = security_shm_shmctl(shp, cmd);
 913                 if (err)
 914                         goto out_unlock;
 915                 memset(&tbuf, 0, sizeof(tbuf));
 916                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 917                 tbuf.shm_segsz  = shp->shm_segsz;
 918                 tbuf.shm_atime  = shp->shm_atim;
 919                 tbuf.shm_dtime  = shp->shm_dtim;
 920                 tbuf.shm_ctime  = shp->shm_ctim;
 921                 tbuf.shm_cpid   = shp->shm_cprid;
 922                 tbuf.shm_lpid   = shp->shm_lprid;
 923                 tbuf.shm_nattch = shp->shm_nattch;
 924                 shm_unlock(shp);
 925                 if(copy_shmid_to_user (buf, &tbuf, version))
 926                         err = -EFAULT;
 927                 else
 928                         err = result;
 929                 goto out;
 930         }
 931         case SHM_LOCK:
 932         case SHM_UNLOCK:
 933         {
 934                 struct file *shm_file;
 935
 936                 shp = shm_lock_check(ns, shmid);
 937                 if (IS_ERR(shp)) {
 938                         err = PTR_ERR(shp);
 939                         goto out;
 940                 }
 941
 942                 audit_ipc_obj(&(shp->shm_perm));
 943
 944                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
 945                         kuid_t euid = current_euid();
 946                         err = -EPERM;
 947                         if (!uid_eq(euid, shp->shm_perm.uid) &&
 948                             !uid_eq(euid, shp->shm_perm.cuid))
 949                                 goto out_unlock;
 950                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
 951                                 goto out_unlock;
 952                 }
 953
 954                 err = security_shm_shmctl(shp, cmd);
 955                 if (err)
 956                         goto out_unlock;
 957
 958                 shm_file = shp->shm_file;
 959                 if (is_file_hugepages(shm_file))
 960                         goto out_unlock;
 961
 962                 if (cmd == SHM_LOCK) {
 963                         struct user_struct *user = current_user();
 964                         err = shmem_lock(shm_file, 1, user);
 965                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
 966                                 shp->shm_perm.mode |= SHM_LOCKED;
 967                                 shp->mlock_user = user;
 968                         }
 969                         goto out_unlock;
 970                 }
 971
 972                 /* SHM_UNLOCK */
 973                 if (!(shp->shm_perm.mode & SHM_LOCKED))
 974                         goto out_unlock;
 975                 shmem_lock(shm_file, 0, shp->mlock_user);
 976                 shp->shm_perm.mode &= ~SHM_LOCKED;
 977                 shp->mlock_user = NULL;
 978                 get_file(shm_file);
 979                 shm_unlock(shp);
 980                 shmem_unlock_mapping(shm_file->f_mapping);
 981                 fput(shm_file);
 982                 goto out;
 983         }
 984         case IPC_RMID:
 985         case IPC_SET:
 986                 err = shmctl_down(ns, shmid, cmd, buf, version);
 987                 return err;
 988         default:
 989                 return -EINVAL;
 990         }
 991
 992 out_unlock:
 993         shm_unlock(shp);
 994 out:
 995         return err;
 996 }
 997
 998 /*
 999  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1000  *
1001  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1002  * "raddr" thing points to kernel space, and there has to be a wrapper around
1003  * this.
1004  */
1005 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1006               unsigned long shmlba)
1007 {
1008         struct shmid_kernel *shp;
1009         unsigned long addr;
1010         unsigned long size;
1011         struct file * file;
1012         int    err;
1013         unsigned long flags;
1014         unsigned long prot;
1015         int acc_mode;
1016         struct ipc_namespace *ns;
1017         struct shm_file_data *sfd;
1018         struct path path;
1019         fmode_t f_mode;
1020         unsigned long populate = 0;
1021
1022         err = -EINVAL;
1023         if (shmid < 0)
1024                 goto out;
1025         else if ((addr = (ulong)shmaddr)) {
1026                 if (addr & (shmlba - 1)) {
1027                         if (shmflg & SHM_RND)
1028                                 addr &= ~(shmlba - 1);     /* round down */
1029                         else
1030 #ifndef __ARCH_FORCE_SHMLBA
1031                                 if (addr & ~PAGE_MASK)
1032 #endif
1033                                         goto out;
1034                 }
1035                 flags = MAP_SHARED | MAP_FIXED;
1036         } else {
1037                 if ((shmflg & SHM_REMAP))
1038                         goto out;
1039
1040                 flags = MAP_SHARED;
1041         }
1042
1043         if (shmflg & SHM_RDONLY) {
1044                 prot = PROT_READ;
1045                 acc_mode = S_IRUGO;
1046                 f_mode = FMODE_READ;
1047         } else {
1048                 prot = PROT_READ | PROT_WRITE;
1049                 acc_mode = S_IRUGO | S_IWUGO;
1050                 f_mode = FMODE_READ | FMODE_WRITE;
1051         }
1052         if (shmflg & SHM_EXEC) {
1053                 prot |= PROT_EXEC;
1054                 acc_mode |= S_IXUGO;
1055         }
1056
1057         /*
1058          * We cannot rely on the fs check since SYSV IPC does have an
1059          * additional creator id...
1060          */
1061         ns = current->nsproxy->ipc_ns;
1062         shp = shm_lock_check(ns, shmid);
1063         if (IS_ERR(shp)) {
1064                 err = PTR_ERR(shp);
1065                 goto out;
1066         }
1067
1068         err = -EACCES;
1069         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1070                 goto out_unlock;
1071
1072         err = security_shm_shmat(shp, shmaddr, shmflg);
1073         if (err)
1074                 goto out_unlock;
1075
1076         path = shp->shm_file->f_path;
1077         path_get(&path);
1078         shp->shm_nattch++;
1079         size = i_size_read(path.dentry->d_inode);
1080         shm_unlock(shp);
1081
1082         err = -ENOMEM;
1083         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1084         if (!sfd)
1085                 goto out_put_dentry;
1086
1087         file = alloc_file(&path, f_mode,
1088                           is_file_hugepages(shp->shm_file) ?
1089                                 &shm_file_operations_huge :
1090                                 &shm_file_operations);
1091         err = PTR_ERR(file);
1092         if (IS_ERR(file))
1093                 goto out_free;
1094
1095         file->private_data = sfd;
1096         file->f_mapping = shp->shm_file->f_mapping;
1097         sfd->id = shp->shm_perm.id;
1098         sfd->ns = get_ipc_ns(ns);
1099         sfd->file = shp->shm_file;
1100         sfd->vm_ops = NULL;
1101
1102         err = security_mmap_file(file, prot, flags);
1103         if (err)
1104                 goto out_fput;
1105
1106         down_write(&current->mm->mmap_sem);
1107         if (addr && !(shmflg & SHM_REMAP)) {
1108                 err = -EINVAL;
1109                 if (find_vma_intersection(current->mm, addr, addr + size))
1110                         goto invalid;
1111                 /*
1112                  * If shm segment goes below stack, make sure there is some
1113                  * space left for the stack to grow (at least 4 pages).
1114                  */
1115                 if (addr < current->mm->start_stack &&
1116                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1117                         goto invalid;
1118         }
1119
1120         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1121         *raddr = addr;
1122         err = 0;
1123         if (IS_ERR_VALUE(addr))
1124                 err = (long)addr;
1125 invalid:
1126         up_write(&current->mm->mmap_sem);
1127         if (populate)
1128                 mm_populate(addr, populate);
1129
1130 out_fput:
1131         fput(file);
1132
1133 out_nattch:
1134         down_write(&shm_ids(ns).rw_mutex);
1135         shp = shm_lock(ns, shmid);
1136         BUG_ON(IS_ERR(shp));
1137         shp->shm_nattch--;
1138         if (shm_may_destroy(ns, shp))
1139                 shm_destroy(ns, shp);
1140         else
1141                 shm_unlock(shp);
1142         up_write(&shm_ids(ns).rw_mutex);
1143
1144 out:
1145         return err;
1146
1147 out_unlock:
1148         shm_unlock(shp);
1149         goto out;
1150
1151 out_free:
1152         kfree(sfd);
1153 out_put_dentry:
1154         path_put(&path);
1155         goto out_nattch;
1156 }
1157
1158 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1159 {
1160         unsigned long ret;
1161         long err;
1162
1163         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1164         if (err)
1165                 return err;
1166         force_successful_syscall_return();
1167         return (long)ret;
1168 }
1169
1170 /*
1171  * detach and kill segment if marked destroyed.
1172  * The work is done in shm_close.
1173  */
1174 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1175 {
1176         struct mm_struct *mm = current->mm;
1177         struct vm_area_struct *vma;
1178         unsigned long addr = (unsigned long)shmaddr;
1179         int retval = -EINVAL;
1180 #ifdef CONFIG_MMU
1181         loff_t size = 0;
1182         struct vm_area_struct *next;
1183 #endif
1184
1185         if (addr & ~PAGE_MASK)
1186                 return retval;
1187
1188         down_write(&mm->mmap_sem);
1189
1190         /*
1191          * This function tries to be smart and unmap shm segments that
1192          * were modified by partial mlock or munmap calls:
1193          * - It first determines the size of the shm segment that should be
1194          *   unmapped: It searches for a vma that is backed by shm and that
1195          *   started at address shmaddr. It records it's size and then unmaps
1196          *   it.
1197          * - Then it unmaps all shm vmas that started at shmaddr and that
1198          *   are within the initially determined size.
1199          * Errors from do_munmap are ignored: the function only fails if
1200          * it's called with invalid parameters or if it's called to unmap
1201          * a part of a vma. Both calls in this function are for full vmas,
1202          * the parameters are directly copied from the vma itself and always
1203          * valid - therefore do_munmap cannot fail. (famous last words?)
1204          */
1205         /*
1206          * If it had been mremap()'d, the starting address would not
1207          * match the usual checks anyway. So assume all vma's are
1208          * above the starting address given.
1209          */
1210         vma = find_vma(mm, addr);
1211
1212 #ifdef CONFIG_MMU
1213         while (vma) {
1214                 next = vma->vm_next;
1215
1216                 /*
1217                  * Check if the starting address would match, i.e. it's
1218                  * a fragment created by mprotect() and/or munmap(), or it
1219                  * otherwise it starts at this address with no hassles.
1220                  */
1221                 if ((vma->vm_ops == &shm_vm_ops) &&
1222                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1223
1224
1225                         size = file_inode(vma->vm_file)->i_size;
1226                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1227                         /*
1228                          * We discovered the size of the shm segment, so
1229                          * break out of here and fall through to the next
1230                          * loop that uses the size information to stop
1231                          * searching for matching vma's.
1232                          */
1233                         retval = 0;
1234                         vma = next;
1235                         break;
1236                 }
1237                 vma = next;
1238         }
1239
1240         /*
1241          * We need look no further than the maximum address a fragment
1242          * could possibly have landed at. Also cast things to loff_t to
1243          * prevent overflows and make comparisons vs. equal-width types.
1244          */
1245         size = PAGE_ALIGN(size);
1246         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1247                 next = vma->vm_next;
1248
1249                 /* finding a matching vma now does not alter retval */
1250                 if ((vma->vm_ops == &shm_vm_ops) &&
1251                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1252
1253                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1254                 vma = next;
1255         }
1256
1257 #else /* CONFIG_MMU */
1258         /* under NOMMU conditions, the exact address to be destroyed must be
1259          * given */
1260         retval = -EINVAL;
1261         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1262                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1263                 retval = 0;
1264         }
1265
1266 #endif
1267
1268         up_write(&mm->mmap_sem);
1269         return retval;
1270 }
1271
1272 #ifdef CONFIG_PROC_FS
1273 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1274 {
1275         struct user_namespace *user_ns = seq_user_ns(s);
1276         struct shmid_kernel *shp = it;
1277         unsigned long rss = 0, swp = 0;
1278
1279         shm_add_rss_swap(shp, &rss, &swp);
1280
1281 #if BITS_PER_LONG <= 32
1282 #define SIZE_SPEC "%10lu"
1283 #else
1284 #define SIZE_SPEC "%21lu"
1285 #endif
1286
1287         return seq_printf(s,
1288                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1289                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1290                           SIZE_SPEC " " SIZE_SPEC "\n",
1291                           shp->shm_perm.key,
1292                           shp->shm_perm.id,
1293                           shp->shm_perm.mode,
1294                           shp->shm_segsz,
1295                           shp->shm_cprid,
1296                           shp->shm_lprid,
1297                           shp->shm_nattch,
1298                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1299                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1300                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1301                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1302                           shp->shm_atim,
1303                           shp->shm_dtim,
1304                           shp->shm_ctim,
1305                           rss * PAGE_SIZE,
1306                           swp * PAGE_SIZE);
1307 }
1308 #endif