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 #include <linux/ima.h>
  43
  44 #include <asm/uaccess.h>
  45
  46 #include "util.h"
  47
  48 struct shm_file_data {
  49         int id;
  50         struct ipc_namespace *ns;
  51         struct file *file;
  52         const struct vm_operations_struct *vm_ops;
  53 };
  54
  55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  56
  57 static const struct file_operations shm_file_operations;
  58 static const struct vm_operations_struct shm_vm_ops;
  59
  60 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  61
  62 #define shm_unlock(shp)                 \
  63         ipc_unlock(&(shp)->shm_perm)
  64
  65 static int newseg(struct ipc_namespace *, struct ipc_params *);
  66 static void shm_open(struct vm_area_struct *vma);
  67 static void shm_close(struct vm_area_struct *vma);
  68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  69 #ifdef CONFIG_PROC_FS
  70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  71 #endif
  72
  73 void shm_init_ns(struct ipc_namespace *ns)
  74 {
  75         ns->shm_ctlmax = SHMMAX;
  76         ns->shm_ctlall = SHMALL;
  77         ns->shm_ctlmni = SHMMNI;
  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 void __init shm_init (void)
 109 {
 110         shm_init_ns(&init_ipc_ns);
 111         ipc_init_proc_interface("sysvipc/shm",
 112                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 113                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 114 }
 115
 116 /*
 117  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 118  * is not necessarily held.
 119  */
 120 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 121 {
 122         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 123
 124         if (IS_ERR(ipcp))
 125                 return (struct shmid_kernel *)ipcp;
 126
 127         return container_of(ipcp, struct shmid_kernel, shm_perm);
 128 }
 129
 130 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 131                                                 int id)
 132 {
 133         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 134
 135         if (IS_ERR(ipcp))
 136                 return (struct shmid_kernel *)ipcp;
 137
 138         return container_of(ipcp, struct shmid_kernel, shm_perm);
 139 }
 140
 141 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 142 {
 143         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 144 }
 145
 146
 147 /* This is called by fork, once for every shm attach. */
 148 static void shm_open(struct vm_area_struct *vma)
 149 {
 150         struct file *file = vma->vm_file;
 151         struct shm_file_data *sfd = shm_file_data(file);
 152         struct shmid_kernel *shp;
 153
 154         shp = shm_lock(sfd->ns, sfd->id);
 155         BUG_ON(IS_ERR(shp));
 156         shp->shm_atim = get_seconds();
 157         shp->shm_lprid = task_tgid_vnr(current);
 158         shp->shm_nattch++;
 159         shm_unlock(shp);
 160 }
 161
 162 /*
 163  * shm_destroy - free the struct shmid_kernel
 164  *
 165  * @ns: namespace
 166  * @shp: struct to free
 167  *
 168  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 169  * but returns with shp unlocked and freed.
 170  */
 171 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 172 {
 173         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 174         shm_rmid(ns, shp);
 175         shm_unlock(shp);
 176         if (!is_file_hugepages(shp->shm_file))
 177                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 178         else if (shp->mlock_user)
 179                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 180                                                 shp->mlock_user);
 181         fput (shp->shm_file);
 182         security_shm_free(shp);
 183         ipc_rcu_putref(shp);
 184 }
 185
 186 /*
 187  * remove the attach descriptor vma.
 188  * free memory for segment if it is marked destroyed.
 189  * The descriptor has already been removed from the current->mm->mmap list
 190  * and will later be kfree()d.
 191  */
 192 static void shm_close(struct vm_area_struct *vma)
 193 {
 194         struct file * file = vma->vm_file;
 195         struct shm_file_data *sfd = shm_file_data(file);
 196         struct shmid_kernel *shp;
 197         struct ipc_namespace *ns = sfd->ns;
 198
 199         down_write(&shm_ids(ns).rw_mutex);
 200         /* remove from the list of attaches of the shm segment */
 201         shp = shm_lock(ns, sfd->id);
 202         BUG_ON(IS_ERR(shp));
 203         shp->shm_lprid = task_tgid_vnr(current);
 204         shp->shm_dtim = get_seconds();
 205         shp->shm_nattch--;
 206         if(shp->shm_nattch == 0 &&
 207            shp->shm_perm.mode & SHM_DEST)
 208                 shm_destroy(ns, shp);
 209         else
 210                 shm_unlock(shp);
 211         up_write(&shm_ids(ns).rw_mutex);
 212 }
 213
 214 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 215 {
 216         struct file *file = vma->vm_file;
 217         struct shm_file_data *sfd = shm_file_data(file);
 218
 219         return sfd->vm_ops->fault(vma, vmf);
 220 }
 221
 222 #ifdef CONFIG_NUMA
 223 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 224 {
 225         struct file *file = vma->vm_file;
 226         struct shm_file_data *sfd = shm_file_data(file);
 227         int err = 0;
 228         if (sfd->vm_ops->set_policy)
 229                 err = sfd->vm_ops->set_policy(vma, new);
 230         return err;
 231 }
 232
 233 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 234                                         unsigned long addr)
 235 {
 236         struct file *file = vma->vm_file;
 237         struct shm_file_data *sfd = shm_file_data(file);
 238         struct mempolicy *pol = NULL;
 239
 240         if (sfd->vm_ops->get_policy)
 241                 pol = sfd->vm_ops->get_policy(vma, addr);
 242         else if (vma->vm_policy)
 243                 pol = vma->vm_policy;
 244
 245         return pol;
 246 }
 247 #endif
 248
 249 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 250 {
 251         struct shm_file_data *sfd = shm_file_data(file);
 252         int ret;
 253
 254         ret = sfd->file->f_op->mmap(sfd->file, vma);
 255         if (ret != 0)
 256                 return ret;
 257         sfd->vm_ops = vma->vm_ops;
 258 #ifdef CONFIG_MMU
 259         BUG_ON(!sfd->vm_ops->fault);
 260 #endif
 261         vma->vm_ops = &shm_vm_ops;
 262         shm_open(vma);
 263
 264         return ret;
 265 }
 266
 267 static int shm_release(struct inode *ino, struct file *file)
 268 {
 269         struct shm_file_data *sfd = shm_file_data(file);
 270
 271         put_ipc_ns(sfd->ns);
 272         shm_file_data(file) = NULL;
 273         kfree(sfd);
 274         return 0;
 275 }
 276
 277 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 278 {
 279         int (*fsync) (struct file *, struct dentry *, int datasync);
 280         struct shm_file_data *sfd = shm_file_data(file);
 281         int ret = -EINVAL;
 282
 283         fsync = sfd->file->f_op->fsync;
 284         if (fsync)
 285                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 286         return ret;
 287 }
 288
 289 static unsigned long shm_get_unmapped_area(struct file *file,
 290         unsigned long addr, unsigned long len, unsigned long pgoff,
 291         unsigned long flags)
 292 {
 293         struct shm_file_data *sfd = shm_file_data(file);
 294         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 295                                                 pgoff, flags);
 296 }
 297
 298 static const struct file_operations shm_file_operations = {
 299         .mmap           = shm_mmap,
 300         .fsync          = shm_fsync,
 301         .release        = shm_release,
 302 };
 303
 304 static const struct file_operations shm_file_operations_huge = {
 305         .mmap           = shm_mmap,
 306         .fsync          = shm_fsync,
 307         .release        = shm_release,
 308         .get_unmapped_area      = shm_get_unmapped_area,
 309 };
 310
 311 int is_file_shm_hugepages(struct file *file)
 312 {
 313         return file->f_op == &shm_file_operations_huge;
 314 }
 315
 316 static const struct vm_operations_struct shm_vm_ops = {
 317         .open   = shm_open,     /* callback for a new vm-area open */
 318         .close  = shm_close,    /* callback for when the vm-area is released */
 319         .fault  = shm_fault,
 320 #if defined(CONFIG_NUMA)
 321         .set_policy = shm_set_policy,
 322         .get_policy = shm_get_policy,
 323 #endif
 324 };
 325
 326 /**
 327  * newseg - Create a new shared memory segment
 328  * @ns: namespace
 329  * @params: ptr to the structure that contains key, size and shmflg
 330  *
 331  * Called with shm_ids.rw_mutex held as a writer.
 332  */
 333
 334 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 335 {
 336         key_t key = params->key;
 337         int shmflg = params->flg;
 338         size_t size = params->u.size;
 339         int error;
 340         struct shmid_kernel *shp;
 341         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 342         struct file * file;
 343         char name[13];
 344         int id;
 345         int acctflag = 0;
 346
 347         if (size < SHMMIN || size > ns->shm_ctlmax)
 348                 return -EINVAL;
 349
 350         if (ns->shm_tot + numpages > ns->shm_ctlall)
 351                 return -ENOSPC;
 352
 353         shp = ipc_rcu_alloc(sizeof(*shp));
 354         if (!shp)
 355                 return -ENOMEM;
 356
 357         shp->shm_perm.key = key;
 358         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 359         shp->mlock_user = NULL;
 360
 361         shp->shm_perm.security = NULL;
 362         error = security_shm_alloc(shp);
 363         if (error) {
 364                 ipc_rcu_putref(shp);
 365                 return error;
 366         }
 367
 368         sprintf (name, "SYSV%08x", key);
 369         if (shmflg & SHM_HUGETLB) {
 370                 /* hugetlb_file_setup applies strict accounting */
 371                 if (shmflg & SHM_NORESERVE)
 372                         acctflag = VM_NORESERVE;
 373                 file = hugetlb_file_setup(name, size, acctflag,
 374                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
 375         } else {
 376                 /*
 377                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 378                  * if it's asked for.
 379                  */
 380                 if  ((shmflg & SHM_NORESERVE) &&
 381                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 382                         acctflag = VM_NORESERVE;
 383                 file = shmem_file_setup(name, size, acctflag);
 384         }
 385         error = PTR_ERR(file);
 386         if (IS_ERR(file))
 387                 goto no_file;
 388
 389         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 390         if (id < 0) {
 391                 error = id;
 392                 goto no_id;
 393         }
 394
 395         shp->shm_cprid = task_tgid_vnr(current);
 396         shp->shm_lprid = 0;
 397         shp->shm_atim = shp->shm_dtim = 0;
 398         shp->shm_ctim = get_seconds();
 399         shp->shm_segsz = size;
 400         shp->shm_nattch = 0;
 401         shp->shm_file = file;
 402         /*
 403          * shmid gets reported as "inode#" in /proc/pid/maps.
 404          * proc-ps tools use this. Changing this will break them.
 405          */
 406         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 407
 408         ns->shm_tot += numpages;
 409         error = shp->shm_perm.id;
 410         shm_unlock(shp);
 411         return error;
 412
 413 no_id:
 414         if (is_file_hugepages(file) && shp->mlock_user)
 415                 user_shm_unlock(size, shp->mlock_user);
 416         fput(file);
 417 no_file:
 418         security_shm_free(shp);
 419         ipc_rcu_putref(shp);
 420         return error;
 421 }
 422
 423 /*
 424  * Called with shm_ids.rw_mutex and ipcp locked.
 425  */
 426 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 427 {
 428         struct shmid_kernel *shp;
 429
 430         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 431         return security_shm_associate(shp, shmflg);
 432 }
 433
 434 /*
 435  * Called with shm_ids.rw_mutex and ipcp locked.
 436  */
 437 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 438                                 struct ipc_params *params)
 439 {
 440         struct shmid_kernel *shp;
 441
 442         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 443         if (shp->shm_segsz < params->u.size)
 444                 return -EINVAL;
 445
 446         return 0;
 447 }
 448
 449 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 450 {
 451         struct ipc_namespace *ns;
 452         struct ipc_ops shm_ops;
 453         struct ipc_params shm_params;
 454
 455         ns = current->nsproxy->ipc_ns;
 456
 457         shm_ops.getnew = newseg;
 458         shm_ops.associate = shm_security;
 459         shm_ops.more_checks = shm_more_checks;
 460
 461         shm_params.key = key;
 462         shm_params.flg = shmflg;
 463         shm_params.u.size = size;
 464
 465         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 466 }
 467
 468 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 469 {
 470         switch(version) {
 471         case IPC_64:
 472                 return copy_to_user(buf, in, sizeof(*in));
 473         case IPC_OLD:
 474             {
 475                 struct shmid_ds out;
 476
 477                 memset(&out, 0, sizeof(out));
 478                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 479                 out.shm_segsz   = in->shm_segsz;
 480                 out.shm_atime   = in->shm_atime;
 481                 out.shm_dtime   = in->shm_dtime;
 482                 out.shm_ctime   = in->shm_ctime;
 483                 out.shm_cpid    = in->shm_cpid;
 484                 out.shm_lpid    = in->shm_lpid;
 485                 out.shm_nattch  = in->shm_nattch;
 486
 487                 return copy_to_user(buf, &out, sizeof(out));
 488             }
 489         default:
 490                 return -EINVAL;
 491         }
 492 }
 493
 494 static inline unsigned long
 495 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 496 {
 497         switch(version) {
 498         case IPC_64:
 499                 if (copy_from_user(out, buf, sizeof(*out)))
 500                         return -EFAULT;
 501                 return 0;
 502         case IPC_OLD:
 503             {
 504                 struct shmid_ds tbuf_old;
 505
 506                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 507                         return -EFAULT;
 508
 509                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 510                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 511                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 512
 513                 return 0;
 514             }
 515         default:
 516                 return -EINVAL;
 517         }
 518 }
 519
 520 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 521 {
 522         switch(version) {
 523         case IPC_64:
 524                 return copy_to_user(buf, in, sizeof(*in));
 525         case IPC_OLD:
 526             {
 527                 struct shminfo out;
 528
 529                 if(in->shmmax > INT_MAX)
 530                         out.shmmax = INT_MAX;
 531                 else
 532                         out.shmmax = (int)in->shmmax;
 533
 534                 out.shmmin      = in->shmmin;
 535                 out.shmmni      = in->shmmni;
 536                 out.shmseg      = in->shmseg;
 537                 out.shmall      = in->shmall;
 538
 539                 return copy_to_user(buf, &out, sizeof(out));
 540             }
 541         default:
 542                 return -EINVAL;
 543         }
 544 }
 545
 546 /*
 547  * Called with shm_ids.rw_mutex held as a reader
 548  */
 549 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 550                 unsigned long *swp)
 551 {
 552         int next_id;
 553         int total, in_use;
 554
 555         *rss = 0;
 556         *swp = 0;
 557
 558         in_use = shm_ids(ns).in_use;
 559
 560         for (total = 0, next_id = 0; total < in_use; next_id++) {
 561                 struct kern_ipc_perm *ipc;
 562                 struct shmid_kernel *shp;
 563                 struct inode *inode;
 564
 565                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 566                 if (ipc == NULL)
 567                         continue;
 568                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 569
 570                 inode = shp->shm_file->f_path.dentry->d_inode;
 571
 572                 if (is_file_hugepages(shp->shm_file)) {
 573                         struct address_space *mapping = inode->i_mapping;
 574                         struct hstate *h = hstate_file(shp->shm_file);
 575                         *rss += pages_per_huge_page(h) * mapping->nrpages;
 576                 } else {
 577 #ifdef CONFIG_SHMEM
 578                         struct shmem_inode_info *info = SHMEM_I(inode);
 579                         spin_lock(&info->lock);
 580                         *rss += inode->i_mapping->nrpages;
 581                         *swp += info->swapped;
 582                         spin_unlock(&info->lock);
 583 #else
 584                         *rss += inode->i_mapping->nrpages;
 585 #endif
 586                 }
 587
 588                 total++;
 589         }
 590 }
 591
 592 /*
 593  * This function handles some shmctl commands which require the rw_mutex
 594  * to be held in write mode.
 595  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 596  */
 597 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 598                        struct shmid_ds __user *buf, int version)
 599 {
 600         struct kern_ipc_perm *ipcp;
 601         struct shmid64_ds shmid64;
 602         struct shmid_kernel *shp;
 603         int err;
 604
 605         if (cmd == IPC_SET) {
 606                 if (copy_shmid_from_user(&shmid64, buf, version))
 607                         return -EFAULT;
 608         }
 609
 610         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 611         if (IS_ERR(ipcp))
 612                 return PTR_ERR(ipcp);
 613
 614         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 615
 616         err = security_shm_shmctl(shp, cmd);
 617         if (err)
 618                 goto out_unlock;
 619         switch (cmd) {
 620         case IPC_RMID:
 621                 do_shm_rmid(ns, ipcp);
 622                 goto out_up;
 623         case IPC_SET:
 624                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 625                 shp->shm_ctim = get_seconds();
 626                 break;
 627         default:
 628                 err = -EINVAL;
 629         }
 630 out_unlock:
 631         shm_unlock(shp);
 632 out_up:
 633         up_write(&shm_ids(ns).rw_mutex);
 634         return err;
 635 }
 636
 637 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 638 {
 639         struct shmid_kernel *shp;
 640         int err, version;
 641         struct ipc_namespace *ns;
 642
 643         if (cmd < 0 || shmid < 0) {
 644                 err = -EINVAL;
 645                 goto out;
 646         }
 647
 648         version = ipc_parse_version(&cmd);
 649         ns = current->nsproxy->ipc_ns;
 650
 651         switch (cmd) { /* replace with proc interface ? */
 652         case IPC_INFO:
 653         {
 654                 struct shminfo64 shminfo;
 655
 656                 err = security_shm_shmctl(NULL, cmd);
 657                 if (err)
 658                         return err;
 659
 660                 memset(&shminfo, 0, sizeof(shminfo));
 661                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 662                 shminfo.shmmax = ns->shm_ctlmax;
 663                 shminfo.shmall = ns->shm_ctlall;
 664
 665                 shminfo.shmmin = SHMMIN;
 666                 if(copy_shminfo_to_user (buf, &shminfo, version))
 667                         return -EFAULT;
 668
 669                 down_read(&shm_ids(ns).rw_mutex);
 670                 err = ipc_get_maxid(&shm_ids(ns));
 671                 up_read(&shm_ids(ns).rw_mutex);
 672
 673                 if(err<0)
 674                         err = 0;
 675                 goto out;
 676         }
 677         case SHM_INFO:
 678         {
 679                 struct shm_info shm_info;
 680
 681                 err = security_shm_shmctl(NULL, cmd);
 682                 if (err)
 683                         return err;
 684
 685                 memset(&shm_info, 0, sizeof(shm_info));
 686                 down_read(&shm_ids(ns).rw_mutex);
 687                 shm_info.used_ids = shm_ids(ns).in_use;
 688                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 689                 shm_info.shm_tot = ns->shm_tot;
 690                 shm_info.swap_attempts = 0;
 691                 shm_info.swap_successes = 0;
 692                 err = ipc_get_maxid(&shm_ids(ns));
 693                 up_read(&shm_ids(ns).rw_mutex);
 694                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 695                         err = -EFAULT;
 696                         goto out;
 697                 }
 698
 699                 err = err < 0 ? 0 : err;
 700                 goto out;
 701         }
 702         case SHM_STAT:
 703         case IPC_STAT:
 704         {
 705                 struct shmid64_ds tbuf;
 706                 int result;
 707
 708                 if (cmd == SHM_STAT) {
 709                         shp = shm_lock(ns, shmid);
 710                         if (IS_ERR(shp)) {
 711                                 err = PTR_ERR(shp);
 712                                 goto out;
 713                         }
 714                         result = shp->shm_perm.id;
 715                 } else {
 716                         shp = shm_lock_check(ns, shmid);
 717                         if (IS_ERR(shp)) {
 718                                 err = PTR_ERR(shp);
 719                                 goto out;
 720                         }
 721                         result = 0;
 722                 }
 723                 err = -EACCES;
 724                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 725                         goto out_unlock;
 726                 err = security_shm_shmctl(shp, cmd);
 727                 if (err)
 728                         goto out_unlock;
 729                 memset(&tbuf, 0, sizeof(tbuf));
 730                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 731                 tbuf.shm_segsz  = shp->shm_segsz;
 732                 tbuf.shm_atime  = shp->shm_atim;
 733                 tbuf.shm_dtime  = shp->shm_dtim;
 734                 tbuf.shm_ctime  = shp->shm_ctim;
 735                 tbuf.shm_cpid   = shp->shm_cprid;
 736                 tbuf.shm_lpid   = shp->shm_lprid;
 737                 tbuf.shm_nattch = shp->shm_nattch;
 738                 shm_unlock(shp);
 739                 if(copy_shmid_to_user (buf, &tbuf, version))
 740                         err = -EFAULT;
 741                 else
 742                         err = result;
 743                 goto out;
 744         }
 745         case SHM_LOCK:
 746         case SHM_UNLOCK:
 747         {
 748                 struct file *uninitialized_var(shm_file);
 749
 750                 lru_add_drain_all();  /* drain pagevecs to lru lists */
 751
 752                 shp = shm_lock_check(ns, shmid);
 753                 if (IS_ERR(shp)) {
 754                         err = PTR_ERR(shp);
 755                         goto out;
 756                 }
 757
 758                 audit_ipc_obj(&(shp->shm_perm));
 759
 760                 if (!capable(CAP_IPC_LOCK)) {
 761                         uid_t euid = current_euid();
 762                         err = -EPERM;
 763                         if (euid != shp->shm_perm.uid &&
 764                             euid != shp->shm_perm.cuid)
 765                                 goto out_unlock;
 766                         if (cmd == SHM_LOCK &&
 767                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 768                                 goto out_unlock;
 769                 }
 770
 771                 err = security_shm_shmctl(shp, cmd);
 772                 if (err)
 773                         goto out_unlock;
 774
 775                 if(cmd==SHM_LOCK) {
 776                         struct user_struct *user = current_user();
 777                         if (!is_file_hugepages(shp->shm_file)) {
 778                                 err = shmem_lock(shp->shm_file, 1, user);
 779                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 780                                         shp->shm_perm.mode |= SHM_LOCKED;
 781                                         shp->mlock_user = user;
 782                                 }
 783                         }
 784                 } else if (!is_file_hugepages(shp->shm_file)) {
 785                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 786                         shp->shm_perm.mode &= ~SHM_LOCKED;
 787                         shp->mlock_user = NULL;
 788                 }
 789                 shm_unlock(shp);
 790                 goto out;
 791         }
 792         case IPC_RMID:
 793         case IPC_SET:
 794                 err = shmctl_down(ns, shmid, cmd, buf, version);
 795                 return err;
 796         default:
 797                 return -EINVAL;
 798         }
 799
 800 out_unlock:
 801         shm_unlock(shp);
 802 out:
 803         return err;
 804 }
 805
 806 /*
 807  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 808  *
 809  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 810  * "raddr" thing points to kernel space, and there has to be a wrapper around
 811  * this.
 812  */
 813 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 814 {
 815         struct shmid_kernel *shp;
 816         unsigned long addr;
 817         unsigned long size;
 818         struct file * file;
 819         int    err;
 820         unsigned long flags;
 821         unsigned long prot;
 822         int acc_mode;
 823         unsigned long user_addr;
 824         struct ipc_namespace *ns;
 825         struct shm_file_data *sfd;
 826         struct path path;
 827         fmode_t f_mode;
 828
 829         err = -EINVAL;
 830         if (shmid < 0)
 831                 goto out;
 832         else if ((addr = (ulong)shmaddr)) {
 833                 if (addr & (SHMLBA-1)) {
 834                         if (shmflg & SHM_RND)
 835                                 addr &= ~(SHMLBA-1);       /* round down */
 836                         else
 837 #ifndef __ARCH_FORCE_SHMLBA
 838                                 if (addr & ~PAGE_MASK)
 839 #endif
 840                                         goto out;
 841                 }
 842                 flags = MAP_SHARED | MAP_FIXED;
 843         } else {
 844                 if ((shmflg & SHM_REMAP))
 845                         goto out;
 846
 847                 flags = MAP_SHARED;
 848         }
 849
 850         if (shmflg & SHM_RDONLY) {
 851                 prot = PROT_READ;
 852                 acc_mode = S_IRUGO;
 853                 f_mode = FMODE_READ;
 854         } else {
 855                 prot = PROT_READ | PROT_WRITE;
 856                 acc_mode = S_IRUGO | S_IWUGO;
 857                 f_mode = FMODE_READ | FMODE_WRITE;
 858         }
 859         if (shmflg & SHM_EXEC) {
 860                 prot |= PROT_EXEC;
 861                 acc_mode |= S_IXUGO;
 862         }
 863
 864         /*
 865          * We cannot rely on the fs check since SYSV IPC does have an
 866          * additional creator id...
 867          */
 868         ns = current->nsproxy->ipc_ns;
 869         shp = shm_lock_check(ns, shmid);
 870         if (IS_ERR(shp)) {
 871                 err = PTR_ERR(shp);
 872                 goto out;
 873         }
 874
 875         err = -EACCES;
 876         if (ipcperms(&shp->shm_perm, acc_mode))
 877                 goto out_unlock;
 878
 879         err = security_shm_shmat(shp, shmaddr, shmflg);
 880         if (err)
 881                 goto out_unlock;
 882
 883         path.dentry = dget(shp->shm_file->f_path.dentry);
 884         path.mnt    = shp->shm_file->f_path.mnt;
 885         shp->shm_nattch++;
 886         size = i_size_read(path.dentry->d_inode);
 887         shm_unlock(shp);
 888
 889         err = -ENOMEM;
 890         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 891         if (!sfd)
 892                 goto out_put_dentry;
 893
 894         file = alloc_file(path.mnt, path.dentry, f_mode,
 895                         is_file_hugepages(shp->shm_file) ?
 896                                 &shm_file_operations_huge :
 897                                 &shm_file_operations);
 898         if (!file)
 899                 goto out_free;
 900         ima_counts_get(file);
 901
 902         file->private_data = sfd;
 903         file->f_mapping = shp->shm_file->f_mapping;
 904         sfd->id = shp->shm_perm.id;
 905         sfd->ns = get_ipc_ns(ns);
 906         sfd->file = shp->shm_file;
 907         sfd->vm_ops = NULL;
 908
 909         down_write(&current->mm->mmap_sem);
 910         if (addr && !(shmflg & SHM_REMAP)) {
 911                 err = -EINVAL;
 912                 if (find_vma_intersection(current->mm, addr, addr + size))
 913                         goto invalid;
 914                 /*
 915                  * If shm segment goes below stack, make sure there is some
 916                  * space left for the stack to grow (at least 4 pages).
 917                  */
 918                 if (addr < current->mm->start_stack &&
 919                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 920                         goto invalid;
 921         }
 922
 923         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 924         *raddr = user_addr;
 925         err = 0;
 926         if (IS_ERR_VALUE(user_addr))
 927                 err = (long)user_addr;
 928 invalid:
 929         up_write(&current->mm->mmap_sem);
 930
 931         fput(file);
 932
 933 out_nattch:
 934         down_write(&shm_ids(ns).rw_mutex);
 935         shp = shm_lock(ns, shmid);
 936         BUG_ON(IS_ERR(shp));
 937         shp->shm_nattch--;
 938         if(shp->shm_nattch == 0 &&
 939            shp->shm_perm.mode & SHM_DEST)
 940                 shm_destroy(ns, shp);
 941         else
 942                 shm_unlock(shp);
 943         up_write(&shm_ids(ns).rw_mutex);
 944
 945 out:
 946         return err;
 947
 948 out_unlock:
 949         shm_unlock(shp);
 950         goto out;
 951
 952 out_free:
 953         kfree(sfd);
 954 out_put_dentry:
 955         dput(path.dentry);
 956         goto out_nattch;
 957 }
 958
 959 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 960 {
 961         unsigned long ret;
 962         long err;
 963
 964         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 965         if (err)
 966                 return err;
 967         force_successful_syscall_return();
 968         return (long)ret;
 969 }
 970
 971 /*
 972  * detach and kill segment if marked destroyed.
 973  * The work is done in shm_close.
 974  */
 975 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 976 {
 977         struct mm_struct *mm = current->mm;
 978         struct vm_area_struct *vma;
 979         unsigned long addr = (unsigned long)shmaddr;
 980         int retval = -EINVAL;
 981 #ifdef CONFIG_MMU
 982         loff_t size = 0;
 983         struct vm_area_struct *next;
 984 #endif
 985
 986         if (addr & ~PAGE_MASK)
 987                 return retval;
 988
 989         down_write(&mm->mmap_sem);
 990
 991         /*
 992          * This function tries to be smart and unmap shm segments that
 993          * were modified by partial mlock or munmap calls:
 994          * - It first determines the size of the shm segment that should be
 995          *   unmapped: It searches for a vma that is backed by shm and that
 996          *   started at address shmaddr. It records it's size and then unmaps
 997          *   it.
 998          * - Then it unmaps all shm vmas that started at shmaddr and that
 999          *   are within the initially determined size.
1000          * Errors from do_munmap are ignored: the function only fails if
1001          * it's called with invalid parameters or if it's called to unmap
1002          * a part of a vma. Both calls in this function are for full vmas,
1003          * the parameters are directly copied from the vma itself and always
1004          * valid - therefore do_munmap cannot fail. (famous last words?)
1005          */
1006         /*
1007          * If it had been mremap()'d, the starting address would not
1008          * match the usual checks anyway. So assume all vma's are
1009          * above the starting address given.
1010          */
1011         vma = find_vma(mm, addr);
1012
1013 #ifdef CONFIG_MMU
1014         while (vma) {
1015                 next = vma->vm_next;
1016
1017                 /*
1018                  * Check if the starting address would match, i.e. it's
1019                  * a fragment created by mprotect() and/or munmap(), or it
1020                  * otherwise it starts at this address with no hassles.
1021                  */
1022                 if ((vma->vm_ops == &shm_vm_ops) &&
1023                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1024
1025
1026                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1027                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1028                         /*
1029                          * We discovered the size of the shm segment, so
1030                          * break out of here and fall through to the next
1031                          * loop that uses the size information to stop
1032                          * searching for matching vma's.
1033                          */
1034                         retval = 0;
1035                         vma = next;
1036                         break;
1037                 }
1038                 vma = next;
1039         }
1040
1041         /*
1042          * We need look no further than the maximum address a fragment
1043          * could possibly have landed at. Also cast things to loff_t to
1044          * prevent overflows and make comparisions vs. equal-width types.
1045          */
1046         size = PAGE_ALIGN(size);
1047         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1048                 next = vma->vm_next;
1049
1050                 /* finding a matching vma now does not alter retval */
1051                 if ((vma->vm_ops == &shm_vm_ops) &&
1052                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1053
1054                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1055                 vma = next;
1056         }
1057
1058 #else /* CONFIG_MMU */
1059         /* under NOMMU conditions, the exact address to be destroyed must be
1060          * given */
1061         retval = -EINVAL;
1062         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1063                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1064                 retval = 0;
1065         }
1066
1067 #endif
1068
1069         up_write(&mm->mmap_sem);
1070         return retval;
1071 }
1072
1073 #ifdef CONFIG_PROC_FS
1074 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1075 {
1076         struct shmid_kernel *shp = it;
1077
1078 #if BITS_PER_LONG <= 32
1079 #define SIZE_SPEC "%10lu"
1080 #else
1081 #define SIZE_SPEC "%21lu"
1082 #endif
1083
1084         return seq_printf(s,
1085                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1086                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1087                           shp->shm_perm.key,
1088                           shp->shm_perm.id,
1089                           shp->shm_perm.mode,
1090                           shp->shm_segsz,
1091                           shp->shm_cprid,
1092                           shp->shm_lprid,
1093                           shp->shm_nattch,
1094                           shp->shm_perm.uid,
1095                           shp->shm_perm.gid,
1096                           shp->shm_perm.cuid,
1097                           shp->shm_perm.cgid,
1098                           shp->shm_atim,
1099                           shp->shm_dtim,
1100                           shp->shm_ctim);
1101 }
1102 #endif