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_tot = 0;
  78         ipc_init_ids(&shm_ids(ns));
  79 }
  80
  81 /*
  82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  83  * Only shm_ids.rw_mutex remains locked on exit.
  84  */
  85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  86 {
  87         struct shmid_kernel *shp;
  88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  89
  90         if (shp->shm_nattch){
  91                 shp->shm_perm.mode |= SHM_DEST;
  92                 /* Do not find it any more */
  93                 shp->shm_perm.key = IPC_PRIVATE;
  94                 shm_unlock(shp);
  95         } else
  96                 shm_destroy(ns, shp);
  97 }
  98
  99 #ifdef CONFIG_IPC_NS
 100 void shm_exit_ns(struct ipc_namespace *ns)
 101 {
 102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 103         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 104 }
 105 #endif
 106
 107 void __init shm_init (void)
 108 {
 109         shm_init_ns(&init_ipc_ns);
 110         ipc_init_proc_interface("sysvipc/shm",
 111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 113 }
 114
 115 /*
 116  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 117  * is not necessarily held.
 118  */
 119 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 120 {
 121         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 122
 123         if (IS_ERR(ipcp))
 124                 return (struct shmid_kernel *)ipcp;
 125
 126         return container_of(ipcp, struct shmid_kernel, shm_perm);
 127 }
 128
 129 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 130                                                 int id)
 131 {
 132         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 133
 134         if (IS_ERR(ipcp))
 135                 return (struct shmid_kernel *)ipcp;
 136
 137         return container_of(ipcp, struct shmid_kernel, shm_perm);
 138 }
 139
 140 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 141 {
 142         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 143 }
 144
 145
 146 /* This is called by fork, once for every shm attach. */
 147 static void shm_open(struct vm_area_struct *vma)
 148 {
 149         struct file *file = vma->vm_file;
 150         struct shm_file_data *sfd = shm_file_data(file);
 151         struct shmid_kernel *shp;
 152
 153         shp = shm_lock(sfd->ns, sfd->id);
 154         BUG_ON(IS_ERR(shp));
 155         shp->shm_atim = get_seconds();
 156         shp->shm_lprid = task_tgid_vnr(current);
 157         shp->shm_nattch++;
 158         shm_unlock(shp);
 159 }
 160
 161 /*
 162  * shm_destroy - free the struct shmid_kernel
 163  *
 164  * @ns: namespace
 165  * @shp: struct to free
 166  *
 167  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 168  * but returns with shp unlocked and freed.
 169  */
 170 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 171 {
 172         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 173         shm_rmid(ns, shp);
 174         shm_unlock(shp);
 175         if (!is_file_hugepages(shp->shm_file))
 176                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 177         else if (shp->mlock_user)
 178                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 179                                                 shp->mlock_user);
 180         fput (shp->shm_file);
 181         security_shm_free(shp);
 182         ipc_rcu_putref(shp);
 183 }
 184
 185 /*
 186  * remove the attach descriptor vma.
 187  * free memory for segment if it is marked destroyed.
 188  * The descriptor has already been removed from the current->mm->mmap list
 189  * and will later be kfree()d.
 190  */
 191 static void shm_close(struct vm_area_struct *vma)
 192 {
 193         struct file * file = vma->vm_file;
 194         struct shm_file_data *sfd = shm_file_data(file);
 195         struct shmid_kernel *shp;
 196         struct ipc_namespace *ns = sfd->ns;
 197
 198         down_write(&shm_ids(ns).rw_mutex);
 199         /* remove from the list of attaches of the shm segment */
 200         shp = shm_lock(ns, sfd->id);
 201         BUG_ON(IS_ERR(shp));
 202         shp->shm_lprid = task_tgid_vnr(current);
 203         shp->shm_dtim = get_seconds();
 204         shp->shm_nattch--;
 205         if(shp->shm_nattch == 0 &&
 206            shp->shm_perm.mode & SHM_DEST)
 207                 shm_destroy(ns, shp);
 208         else
 209                 shm_unlock(shp);
 210         up_write(&shm_ids(ns).rw_mutex);
 211 }
 212
 213 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 214 {
 215         struct file *file = vma->vm_file;
 216         struct shm_file_data *sfd = shm_file_data(file);
 217
 218         return sfd->vm_ops->fault(vma, vmf);
 219 }
 220
 221 #ifdef CONFIG_NUMA
 222 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 223 {
 224         struct file *file = vma->vm_file;
 225         struct shm_file_data *sfd = shm_file_data(file);
 226         int err = 0;
 227         if (sfd->vm_ops->set_policy)
 228                 err = sfd->vm_ops->set_policy(vma, new);
 229         return err;
 230 }
 231
 232 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 233                                         unsigned long addr)
 234 {
 235         struct file *file = vma->vm_file;
 236         struct shm_file_data *sfd = shm_file_data(file);
 237         struct mempolicy *pol = NULL;
 238
 239         if (sfd->vm_ops->get_policy)
 240                 pol = sfd->vm_ops->get_policy(vma, addr);
 241         else if (vma->vm_policy)
 242                 pol = vma->vm_policy;
 243
 244         return pol;
 245 }
 246 #endif
 247
 248 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 249 {
 250         struct shm_file_data *sfd = shm_file_data(file);
 251         int ret;
 252
 253         ret = sfd->file->f_op->mmap(sfd->file, vma);
 254         if (ret != 0)
 255                 return ret;
 256         sfd->vm_ops = vma->vm_ops;
 257 #ifdef CONFIG_MMU
 258         BUG_ON(!sfd->vm_ops->fault);
 259 #endif
 260         vma->vm_ops = &shm_vm_ops;
 261         shm_open(vma);
 262
 263         return ret;
 264 }
 265
 266 static int shm_release(struct inode *ino, struct file *file)
 267 {
 268         struct shm_file_data *sfd = shm_file_data(file);
 269
 270         put_ipc_ns(sfd->ns);
 271         shm_file_data(file) = NULL;
 272         kfree(sfd);
 273         return 0;
 274 }
 275
 276 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 277 {
 278         int (*fsync) (struct file *, struct dentry *, int datasync);
 279         struct shm_file_data *sfd = shm_file_data(file);
 280         int ret = -EINVAL;
 281
 282         fsync = sfd->file->f_op->fsync;
 283         if (fsync)
 284                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 285         return ret;
 286 }
 287
 288 static unsigned long shm_get_unmapped_area(struct file *file,
 289         unsigned long addr, unsigned long len, unsigned long pgoff,
 290         unsigned long flags)
 291 {
 292         struct shm_file_data *sfd = shm_file_data(file);
 293         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 294                                                 pgoff, flags);
 295 }
 296
 297 static const struct file_operations shm_file_operations = {
 298         .mmap           = shm_mmap,
 299         .fsync          = shm_fsync,
 300         .release        = shm_release,
 301 #ifndef CONFIG_MMU
 302         .get_unmapped_area      = shm_get_unmapped_area,
 303 #endif
 304 };
 305
 306 static const struct file_operations shm_file_operations_huge = {
 307         .mmap           = shm_mmap,
 308         .fsync          = shm_fsync,
 309         .release        = shm_release,
 310         .get_unmapped_area      = shm_get_unmapped_area,
 311 };
 312
 313 int is_file_shm_hugepages(struct file *file)
 314 {
 315         return file->f_op == &shm_file_operations_huge;
 316 }
 317
 318 static const struct vm_operations_struct shm_vm_ops = {
 319         .open   = shm_open,     /* callback for a new vm-area open */
 320         .close  = shm_close,    /* callback for when the vm-area is released */
 321         .fault  = shm_fault,
 322 #if defined(CONFIG_NUMA)
 323         .set_policy = shm_set_policy,
 324         .get_policy = shm_get_policy,
 325 #endif
 326 };
 327
 328 /**
 329  * newseg - Create a new shared memory segment
 330  * @ns: namespace
 331  * @params: ptr to the structure that contains key, size and shmflg
 332  *
 333  * Called with shm_ids.rw_mutex held as a writer.
 334  */
 335
 336 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 337 {
 338         key_t key = params->key;
 339         int shmflg = params->flg;
 340         size_t size = params->u.size;
 341         int error;
 342         struct shmid_kernel *shp;
 343         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 344         struct file * file;
 345         char name[13];
 346         int id;
 347         int acctflag = 0;
 348
 349         if (size < SHMMIN || size > ns->shm_ctlmax)
 350                 return -EINVAL;
 351
 352         if (ns->shm_tot + numpages > ns->shm_ctlall)
 353                 return -ENOSPC;
 354
 355         shp = ipc_rcu_alloc(sizeof(*shp));
 356         if (!shp)
 357                 return -ENOMEM;
 358
 359         shp->shm_perm.key = key;
 360         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 361         shp->mlock_user = NULL;
 362
 363         shp->shm_perm.security = NULL;
 364         error = security_shm_alloc(shp);
 365         if (error) {
 366                 ipc_rcu_putref(shp);
 367                 return error;
 368         }
 369
 370         sprintf (name, "SYSV%08x", key);
 371         if (shmflg & SHM_HUGETLB) {
 372                 /* hugetlb_file_setup applies strict accounting */
 373                 if (shmflg & SHM_NORESERVE)
 374                         acctflag = VM_NORESERVE;
 375                 file = hugetlb_file_setup(name, size, acctflag,
 376                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
 377         } else {
 378                 /*
 379                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 380                  * if it's asked for.
 381                  */
 382                 if  ((shmflg & SHM_NORESERVE) &&
 383                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 384                         acctflag = VM_NORESERVE;
 385                 file = shmem_file_setup(name, size, acctflag);
 386         }
 387         error = PTR_ERR(file);
 388         if (IS_ERR(file))
 389                 goto no_file;
 390
 391         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 392         if (id < 0) {
 393                 error = id;
 394                 goto no_id;
 395         }
 396
 397         shp->shm_cprid = task_tgid_vnr(current);
 398         shp->shm_lprid = 0;
 399         shp->shm_atim = shp->shm_dtim = 0;
 400         shp->shm_ctim = get_seconds();
 401         shp->shm_segsz = size;
 402         shp->shm_nattch = 0;
 403         shp->shm_file = file;
 404         /*
 405          * shmid gets reported as "inode#" in /proc/pid/maps.
 406          * proc-ps tools use this. Changing this will break them.
 407          */
 408         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 409
 410         ns->shm_tot += numpages;
 411         error = shp->shm_perm.id;
 412         shm_unlock(shp);
 413         return error;
 414
 415 no_id:
 416         if (is_file_hugepages(file) && shp->mlock_user)
 417                 user_shm_unlock(size, shp->mlock_user);
 418         fput(file);
 419 no_file:
 420         security_shm_free(shp);
 421         ipc_rcu_putref(shp);
 422         return error;
 423 }
 424
 425 /*
 426  * Called with shm_ids.rw_mutex and ipcp locked.
 427  */
 428 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 429 {
 430         struct shmid_kernel *shp;
 431
 432         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 433         return security_shm_associate(shp, shmflg);
 434 }
 435
 436 /*
 437  * Called with shm_ids.rw_mutex and ipcp locked.
 438  */
 439 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 440                                 struct ipc_params *params)
 441 {
 442         struct shmid_kernel *shp;
 443
 444         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 445         if (shp->shm_segsz < params->u.size)
 446                 return -EINVAL;
 447
 448         return 0;
 449 }
 450
 451 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 452 {
 453         struct ipc_namespace *ns;
 454         struct ipc_ops shm_ops;
 455         struct ipc_params shm_params;
 456
 457         ns = current->nsproxy->ipc_ns;
 458
 459         shm_ops.getnew = newseg;
 460         shm_ops.associate = shm_security;
 461         shm_ops.more_checks = shm_more_checks;
 462
 463         shm_params.key = key;
 464         shm_params.flg = shmflg;
 465         shm_params.u.size = size;
 466
 467         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 468 }
 469
 470 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 471 {
 472         switch(version) {
 473         case IPC_64:
 474                 return copy_to_user(buf, in, sizeof(*in));
 475         case IPC_OLD:
 476             {
 477                 struct shmid_ds out;
 478
 479                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 480                 out.shm_segsz   = in->shm_segsz;
 481                 out.shm_atime   = in->shm_atime;
 482                 out.shm_dtime   = in->shm_dtime;
 483                 out.shm_ctime   = in->shm_ctime;
 484                 out.shm_cpid    = in->shm_cpid;
 485                 out.shm_lpid    = in->shm_lpid;
 486                 out.shm_nattch  = in->shm_nattch;
 487
 488                 return copy_to_user(buf, &out, sizeof(out));
 489             }
 490         default:
 491                 return -EINVAL;
 492         }
 493 }
 494
 495 static inline unsigned long
 496 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 497 {
 498         switch(version) {
 499         case IPC_64:
 500                 if (copy_from_user(out, buf, sizeof(*out)))
 501                         return -EFAULT;
 502                 return 0;
 503         case IPC_OLD:
 504             {
 505                 struct shmid_ds tbuf_old;
 506
 507                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 508                         return -EFAULT;
 509
 510                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 511                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 512                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 513
 514                 return 0;
 515             }
 516         default:
 517                 return -EINVAL;
 518         }
 519 }
 520
 521 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 522 {
 523         switch(version) {
 524         case IPC_64:
 525                 return copy_to_user(buf, in, sizeof(*in));
 526         case IPC_OLD:
 527             {
 528                 struct shminfo out;
 529
 530                 if(in->shmmax > INT_MAX)
 531                         out.shmmax = INT_MAX;
 532                 else
 533                         out.shmmax = (int)in->shmmax;
 534
 535                 out.shmmin      = in->shmmin;
 536                 out.shmmni      = in->shmmni;
 537                 out.shmseg      = in->shmseg;
 538                 out.shmall      = in->shmall;
 539
 540                 return copy_to_user(buf, &out, sizeof(out));
 541             }
 542         default:
 543                 return -EINVAL;
 544         }
 545 }
 546
 547 /*
 548  * Called with shm_ids.rw_mutex held as a reader
 549  */
 550 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 551                 unsigned long *swp)
 552 {
 553         int next_id;
 554         int total, in_use;
 555
 556         *rss = 0;
 557         *swp = 0;
 558
 559         in_use = shm_ids(ns).in_use;
 560
 561         for (total = 0, next_id = 0; total < in_use; next_id++) {
 562                 struct kern_ipc_perm *ipc;
 563                 struct shmid_kernel *shp;
 564                 struct inode *inode;
 565
 566                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 567                 if (ipc == NULL)
 568                         continue;
 569                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 570
 571                 inode = shp->shm_file->f_path.dentry->d_inode;
 572
 573                 if (is_file_hugepages(shp->shm_file)) {
 574                         struct address_space *mapping = inode->i_mapping;
 575                         struct hstate *h = hstate_file(shp->shm_file);
 576                         *rss += pages_per_huge_page(h) * mapping->nrpages;
 577                 } else {
 578 #ifdef CONFIG_SHMEM
 579                         struct shmem_inode_info *info = SHMEM_I(inode);
 580                         spin_lock(&info->lock);
 581                         *rss += inode->i_mapping->nrpages;
 582                         *swp += info->swapped;
 583                         spin_unlock(&info->lock);
 584 #else
 585                         *rss += inode->i_mapping->nrpages;
 586 #endif
 587                 }
 588
 589                 total++;
 590         }
 591 }
 592
 593 /*
 594  * This function handles some shmctl commands which require the rw_mutex
 595  * to be held in write mode.
 596  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 597  */
 598 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 599                        struct shmid_ds __user *buf, int version)
 600 {
 601         struct kern_ipc_perm *ipcp;
 602         struct shmid64_ds shmid64;
 603         struct shmid_kernel *shp;
 604         int err;
 605
 606         if (cmd == IPC_SET) {
 607                 if (copy_shmid_from_user(&shmid64, buf, version))
 608                         return -EFAULT;
 609         }
 610
 611         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 612         if (IS_ERR(ipcp))
 613                 return PTR_ERR(ipcp);
 614
 615         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 616
 617         err = security_shm_shmctl(shp, cmd);
 618         if (err)
 619                 goto out_unlock;
 620         switch (cmd) {
 621         case IPC_RMID:
 622                 do_shm_rmid(ns, ipcp);
 623                 goto out_up;
 624         case IPC_SET:
 625                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 626                 shp->shm_ctim = get_seconds();
 627                 break;
 628         default:
 629                 err = -EINVAL;
 630         }
 631 out_unlock:
 632         shm_unlock(shp);
 633 out_up:
 634         up_write(&shm_ids(ns).rw_mutex);
 635         return err;
 636 }
 637
 638 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 639 {
 640         struct shmid_kernel *shp;
 641         int err, version;
 642         struct ipc_namespace *ns;
 643
 644         if (cmd < 0 || shmid < 0) {
 645                 err = -EINVAL;
 646                 goto out;
 647         }
 648
 649         version = ipc_parse_version(&cmd);
 650         ns = current->nsproxy->ipc_ns;
 651
 652         switch (cmd) { /* replace with proc interface ? */
 653         case IPC_INFO:
 654         {
 655                 struct shminfo64 shminfo;
 656
 657                 err = security_shm_shmctl(NULL, cmd);
 658                 if (err)
 659                         return err;
 660
 661                 memset(&shminfo, 0, sizeof(shminfo));
 662                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 663                 shminfo.shmmax = ns->shm_ctlmax;
 664                 shminfo.shmall = ns->shm_ctlall;
 665
 666                 shminfo.shmmin = SHMMIN;
 667                 if(copy_shminfo_to_user (buf, &shminfo, version))
 668                         return -EFAULT;
 669
 670                 down_read(&shm_ids(ns).rw_mutex);
 671                 err = ipc_get_maxid(&shm_ids(ns));
 672                 up_read(&shm_ids(ns).rw_mutex);
 673
 674                 if(err<0)
 675                         err = 0;
 676                 goto out;
 677         }
 678         case SHM_INFO:
 679         {
 680                 struct shm_info shm_info;
 681
 682                 err = security_shm_shmctl(NULL, cmd);
 683                 if (err)
 684                         return err;
 685
 686                 memset(&shm_info, 0, sizeof(shm_info));
 687                 down_read(&shm_ids(ns).rw_mutex);
 688                 shm_info.used_ids = shm_ids(ns).in_use;
 689                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 690                 shm_info.shm_tot = ns->shm_tot;
 691                 shm_info.swap_attempts = 0;
 692                 shm_info.swap_successes = 0;
 693                 err = ipc_get_maxid(&shm_ids(ns));
 694                 up_read(&shm_ids(ns).rw_mutex);
 695                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 696                         err = -EFAULT;
 697                         goto out;
 698                 }
 699
 700                 err = err < 0 ? 0 : err;
 701                 goto out;
 702         }
 703         case SHM_STAT:
 704         case IPC_STAT:
 705         {
 706                 struct shmid64_ds tbuf;
 707                 int result;
 708
 709                 if (cmd == SHM_STAT) {
 710                         shp = shm_lock(ns, shmid);
 711                         if (IS_ERR(shp)) {
 712                                 err = PTR_ERR(shp);
 713                                 goto out;
 714                         }
 715                         result = shp->shm_perm.id;
 716                 } else {
 717                         shp = shm_lock_check(ns, shmid);
 718                         if (IS_ERR(shp)) {
 719                                 err = PTR_ERR(shp);
 720                                 goto out;
 721                         }
 722                         result = 0;
 723                 }
 724                 err = -EACCES;
 725                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 726                         goto out_unlock;
 727                 err = security_shm_shmctl(shp, cmd);
 728                 if (err)
 729                         goto out_unlock;
 730                 memset(&tbuf, 0, sizeof(tbuf));
 731                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 732                 tbuf.shm_segsz  = shp->shm_segsz;
 733                 tbuf.shm_atime  = shp->shm_atim;
 734                 tbuf.shm_dtime  = shp->shm_dtim;
 735                 tbuf.shm_ctime  = shp->shm_ctim;
 736                 tbuf.shm_cpid   = shp->shm_cprid;
 737                 tbuf.shm_lpid   = shp->shm_lprid;
 738                 tbuf.shm_nattch = shp->shm_nattch;
 739                 shm_unlock(shp);
 740                 if(copy_shmid_to_user (buf, &tbuf, version))
 741                         err = -EFAULT;
 742                 else
 743                         err = result;
 744                 goto out;
 745         }
 746         case SHM_LOCK:
 747         case SHM_UNLOCK:
 748         {
 749                 struct file *uninitialized_var(shm_file);
 750
 751                 lru_add_drain_all();  /* drain pagevecs to lru lists */
 752
 753                 shp = shm_lock_check(ns, shmid);
 754                 if (IS_ERR(shp)) {
 755                         err = PTR_ERR(shp);
 756                         goto out;
 757                 }
 758
 759                 audit_ipc_obj(&(shp->shm_perm));
 760
 761                 if (!capable(CAP_IPC_LOCK)) {
 762                         uid_t euid = current_euid();
 763                         err = -EPERM;
 764                         if (euid != shp->shm_perm.uid &&
 765                             euid != shp->shm_perm.cuid)
 766                                 goto out_unlock;
 767                         if (cmd == SHM_LOCK &&
 768                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 769                                 goto out_unlock;
 770                 }
 771
 772                 err = security_shm_shmctl(shp, cmd);
 773                 if (err)
 774                         goto out_unlock;
 775
 776                 if(cmd==SHM_LOCK) {
 777                         struct user_struct *user = current_user();
 778                         if (!is_file_hugepages(shp->shm_file)) {
 779                                 err = shmem_lock(shp->shm_file, 1, user);
 780                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 781                                         shp->shm_perm.mode |= SHM_LOCKED;
 782                                         shp->mlock_user = user;
 783                                 }
 784                         }
 785                 } else if (!is_file_hugepages(shp->shm_file)) {
 786                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 787                         shp->shm_perm.mode &= ~SHM_LOCKED;
 788                         shp->mlock_user = NULL;
 789                 }
 790                 shm_unlock(shp);
 791                 goto out;
 792         }
 793         case IPC_RMID:
 794         case IPC_SET:
 795                 err = shmctl_down(ns, shmid, cmd, buf, version);
 796                 return err;
 797         default:
 798                 return -EINVAL;
 799         }
 800
 801 out_unlock:
 802         shm_unlock(shp);
 803 out:
 804         return err;
 805 }
 806
 807 /*
 808  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 809  *
 810  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 811  * "raddr" thing points to kernel space, and there has to be a wrapper around
 812  * this.
 813  */
 814 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 815 {
 816         struct shmid_kernel *shp;
 817         unsigned long addr;
 818         unsigned long size;
 819         struct file * file;
 820         int    err;
 821         unsigned long flags;
 822         unsigned long prot;
 823         int acc_mode;
 824         unsigned long user_addr;
 825         struct ipc_namespace *ns;
 826         struct shm_file_data *sfd;
 827         struct path path;
 828         fmode_t f_mode;
 829
 830         err = -EINVAL;
 831         if (shmid < 0)
 832                 goto out;
 833         else if ((addr = (ulong)shmaddr)) {
 834                 if (addr & (SHMLBA-1)) {
 835                         if (shmflg & SHM_RND)
 836                                 addr &= ~(SHMLBA-1);       /* round down */
 837                         else
 838 #ifndef __ARCH_FORCE_SHMLBA
 839                                 if (addr & ~PAGE_MASK)
 840 #endif
 841                                         goto out;
 842                 }
 843                 flags = MAP_SHARED | MAP_FIXED;
 844         } else {
 845                 if ((shmflg & SHM_REMAP))
 846                         goto out;
 847
 848                 flags = MAP_SHARED;
 849         }
 850
 851         if (shmflg & SHM_RDONLY) {
 852                 prot = PROT_READ;
 853                 acc_mode = S_IRUGO;
 854                 f_mode = FMODE_READ;
 855         } else {
 856                 prot = PROT_READ | PROT_WRITE;
 857                 acc_mode = S_IRUGO | S_IWUGO;
 858                 f_mode = FMODE_READ | FMODE_WRITE;
 859         }
 860         if (shmflg & SHM_EXEC) {
 861                 prot |= PROT_EXEC;
 862                 acc_mode |= S_IXUGO;
 863         }
 864
 865         /*
 866          * We cannot rely on the fs check since SYSV IPC does have an
 867          * additional creator id...
 868          */
 869         ns = current->nsproxy->ipc_ns;
 870         shp = shm_lock_check(ns, shmid);
 871         if (IS_ERR(shp)) {
 872                 err = PTR_ERR(shp);
 873                 goto out;
 874         }
 875
 876         err = -EACCES;
 877         if (ipcperms(&shp->shm_perm, acc_mode))
 878                 goto out_unlock;
 879
 880         err = security_shm_shmat(shp, shmaddr, shmflg);
 881         if (err)
 882                 goto out_unlock;
 883
 884         path = shp->shm_file->f_path;
 885         path_get(&path);
 886         shp->shm_nattch++;
 887         size = i_size_read(path.dentry->d_inode);
 888         shm_unlock(shp);
 889
 890         err = -ENOMEM;
 891         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 892         if (!sfd)
 893                 goto out_put_dentry;
 894
 895         file = alloc_file(&path, f_mode,
 896                           is_file_hugepages(shp->shm_file) ?
 897                                 &shm_file_operations_huge :
 898                                 &shm_file_operations);
 899         if (!file)
 900                 goto out_free;
 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         path_put(&path);
 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