err.no Git - linux-2.6/blob - 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 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(&ns->ids[IPC_SHM_IDS]);
  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 }
 104 #endif
 105
 106 void __init shm_init (void)
 107 {
 108         shm_init_ns(&init_ipc_ns);
 109         ipc_init_proc_interface("sysvipc/shm",
 110                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 111                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 112 }
 113
 114 /*
 115  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
 116  * is held to protect access to the idr tree.
 117  */
 118 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
 119                                                 int id)
 120 {
 121         struct kern_ipc_perm *ipcp = ipc_lock_down(&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_down(
 130                                                 struct ipc_namespace *ns,
 131                                                 int id)
 132 {
 133         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&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 /*
 142  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 143  * is not held.
 144  */
 145 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 146 {
 147         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 148
 149         if (IS_ERR(ipcp))
 150                 return (struct shmid_kernel *)ipcp;
 151
 152         return container_of(ipcp, struct shmid_kernel, shm_perm);
 153 }
 154
 155 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 156                                                 int id)
 157 {
 158         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 159
 160         if (IS_ERR(ipcp))
 161                 return (struct shmid_kernel *)ipcp;
 162
 163         return container_of(ipcp, struct shmid_kernel, shm_perm);
 164 }
 165
 166 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 167 {
 168         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 169 }
 170
 171
 172 /* This is called by fork, once for every shm attach. */
 173 static void shm_open(struct vm_area_struct *vma)
 174 {
 175         struct file *file = vma->vm_file;
 176         struct shm_file_data *sfd = shm_file_data(file);
 177         struct shmid_kernel *shp;
 178
 179         shp = shm_lock(sfd->ns, sfd->id);
 180         BUG_ON(IS_ERR(shp));
 181         shp->shm_atim = get_seconds();
 182         shp->shm_lprid = task_tgid_vnr(current);
 183         shp->shm_nattch++;
 184         shm_unlock(shp);
 185 }
 186
 187 /*
 188  * shm_destroy - free the struct shmid_kernel
 189  *
 190  * @ns: namespace
 191  * @shp: struct to free
 192  *
 193  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 194  * but returns with shp unlocked and freed.
 195  */
 196 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 197 {
 198         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 199         shm_rmid(ns, shp);
 200         shm_unlock(shp);
 201         if (!is_file_hugepages(shp->shm_file))
 202                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 203         else
 204                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 205                                                 shp->mlock_user);
 206         fput (shp->shm_file);
 207         security_shm_free(shp);
 208         ipc_rcu_putref(shp);
 209 }
 210
 211 /*
 212  * remove the attach descriptor vma.
 213  * free memory for segment if it is marked destroyed.
 214  * The descriptor has already been removed from the current->mm->mmap list
 215  * and will later be kfree()d.
 216  */
 217 static void shm_close(struct vm_area_struct *vma)
 218 {
 219         struct file * file = vma->vm_file;
 220         struct shm_file_data *sfd = shm_file_data(file);
 221         struct shmid_kernel *shp;
 222         struct ipc_namespace *ns = sfd->ns;
 223
 224         down_write(&shm_ids(ns).rw_mutex);
 225         /* remove from the list of attaches of the shm segment */
 226         shp = shm_lock_down(ns, sfd->id);
 227         BUG_ON(IS_ERR(shp));
 228         shp->shm_lprid = task_tgid_vnr(current);
 229         shp->shm_dtim = get_seconds();
 230         shp->shm_nattch--;
 231         if(shp->shm_nattch == 0 &&
 232            shp->shm_perm.mode & SHM_DEST)
 233                 shm_destroy(ns, shp);
 234         else
 235                 shm_unlock(shp);
 236         up_write(&shm_ids(ns).rw_mutex);
 237 }
 238
 239 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 240 {
 241         struct file *file = vma->vm_file;
 242         struct shm_file_data *sfd = shm_file_data(file);
 243
 244         return sfd->vm_ops->fault(vma, vmf);
 245 }
 246
 247 #ifdef CONFIG_NUMA
 248 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 249 {
 250         struct file *file = vma->vm_file;
 251         struct shm_file_data *sfd = shm_file_data(file);
 252         int err = 0;
 253         if (sfd->vm_ops->set_policy)
 254                 err = sfd->vm_ops->set_policy(vma, new);
 255         return err;
 256 }
 257
 258 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 259                                         unsigned long addr)
 260 {
 261         struct file *file = vma->vm_file;
 262         struct shm_file_data *sfd = shm_file_data(file);
 263         struct mempolicy *pol = NULL;
 264
 265         if (sfd->vm_ops->get_policy)
 266                 pol = sfd->vm_ops->get_policy(vma, addr);
 267         else if (vma->vm_policy)
 268                 pol = vma->vm_policy;
 269
 270         return pol;
 271 }
 272 #endif
 273
 274 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 275 {
 276         struct shm_file_data *sfd = shm_file_data(file);
 277         int ret;
 278
 279         ret = sfd->file->f_op->mmap(sfd->file, vma);
 280         if (ret != 0)
 281                 return ret;
 282         sfd->vm_ops = vma->vm_ops;
 283 #ifdef CONFIG_MMU
 284         BUG_ON(!sfd->vm_ops->fault);
 285 #endif
 286         vma->vm_ops = &shm_vm_ops;
 287         shm_open(vma);
 288
 289         return ret;
 290 }
 291
 292 static int shm_release(struct inode *ino, struct file *file)
 293 {
 294         struct shm_file_data *sfd = shm_file_data(file);
 295
 296         put_ipc_ns(sfd->ns);
 297         shm_file_data(file) = NULL;
 298         kfree(sfd);
 299         return 0;
 300 }
 301
 302 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 303 {
 304         int (*fsync) (struct file *, struct dentry *, int datasync);
 305         struct shm_file_data *sfd = shm_file_data(file);
 306         int ret = -EINVAL;
 307
 308         fsync = sfd->file->f_op->fsync;
 309         if (fsync)
 310                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 311         return ret;
 312 }
 313
 314 static unsigned long shm_get_unmapped_area(struct file *file,
 315         unsigned long addr, unsigned long len, unsigned long pgoff,
 316         unsigned long flags)
 317 {
 318         struct shm_file_data *sfd = shm_file_data(file);
 319         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 320 }
 321
 322 int is_file_shm_hugepages(struct file *file)
 323 {
 324         int ret = 0;
 325
 326         if (file->f_op == &shm_file_operations) {
 327                 struct shm_file_data *sfd;
 328                 sfd = shm_file_data(file);
 329                 ret = is_file_hugepages(sfd->file);
 330         }
 331         return ret;
 332 }
 333
 334 static const struct file_operations shm_file_operations = {
 335         .mmap           = shm_mmap,
 336         .fsync          = shm_fsync,
 337         .release        = shm_release,
 338         .get_unmapped_area      = shm_get_unmapped_area,
 339 };
 340
 341 static struct vm_operations_struct shm_vm_ops = {
 342         .open   = shm_open,     /* callback for a new vm-area open */
 343         .close  = shm_close,    /* callback for when the vm-area is released */
 344         .fault  = shm_fault,
 345 #if defined(CONFIG_NUMA)
 346         .set_policy = shm_set_policy,
 347         .get_policy = shm_get_policy,
 348 #endif
 349 };
 350
 351 /**
 352  * newseg - Create a new shared memory segment
 353  * @ns: namespace
 354  * @params: ptr to the structure that contains key, size and shmflg
 355  *
 356  * Called with shm_ids.rw_mutex held as a writer.
 357  */
 358
 359 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 360 {
 361         key_t key = params->key;
 362         int shmflg = params->flg;
 363         size_t size = params->u.size;
 364         int error;
 365         struct shmid_kernel *shp;
 366         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 367         struct file * file;
 368         char name[13];
 369         int id;
 370
 371         if (size < SHMMIN || size > ns->shm_ctlmax)
 372                 return -EINVAL;
 373
 374         if (ns->shm_tot + numpages > ns->shm_ctlall)
 375                 return -ENOSPC;
 376
 377         shp = ipc_rcu_alloc(sizeof(*shp));
 378         if (!shp)
 379                 return -ENOMEM;
 380
 381         shp->shm_perm.key = key;
 382         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 383         shp->mlock_user = NULL;
 384
 385         shp->shm_perm.security = NULL;
 386         error = security_shm_alloc(shp);
 387         if (error) {
 388                 ipc_rcu_putref(shp);
 389                 return error;
 390         }
 391
 392         sprintf (name, "SYSV%08x", key);
 393         if (shmflg & SHM_HUGETLB) {
 394                 /* hugetlb_file_setup takes care of mlock user accounting */
 395                 file = hugetlb_file_setup(name, size);
 396                 shp->mlock_user = current->user;
 397         } else {
 398                 int acctflag = VM_ACCOUNT;
 399                 /*
 400                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 401                  * if it's asked for.
 402                  */
 403                 if  ((shmflg & SHM_NORESERVE) &&
 404                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 405                         acctflag = 0;
 406                 file = shmem_file_setup(name, size, acctflag);
 407         }
 408         error = PTR_ERR(file);
 409         if (IS_ERR(file))
 410                 goto no_file;
 411
 412         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 413         if (id < 0) {
 414                 error = id;
 415                 goto no_id;
 416         }
 417
 418         shp->shm_cprid = task_tgid_vnr(current);
 419         shp->shm_lprid = 0;
 420         shp->shm_atim = shp->shm_dtim = 0;
 421         shp->shm_ctim = get_seconds();
 422         shp->shm_segsz = size;
 423         shp->shm_nattch = 0;
 424         shp->shm_file = file;
 425         /*
 426          * shmid gets reported as "inode#" in /proc/pid/maps.
 427          * proc-ps tools use this. Changing this will break them.
 428          */
 429         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 430
 431         ns->shm_tot += numpages;
 432         error = shp->shm_perm.id;
 433         shm_unlock(shp);
 434         return error;
 435
 436 no_id:
 437         fput(file);
 438 no_file:
 439         security_shm_free(shp);
 440         ipc_rcu_putref(shp);
 441         return error;
 442 }
 443
 444 /*
 445  * Called with shm_ids.rw_mutex and ipcp locked.
 446  */
 447 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 448 {
 449         struct shmid_kernel *shp;
 450
 451         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 452         return security_shm_associate(shp, shmflg);
 453 }
 454
 455 /*
 456  * Called with shm_ids.rw_mutex and ipcp locked.
 457  */
 458 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 459                                 struct ipc_params *params)
 460 {
 461         struct shmid_kernel *shp;
 462
 463         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 464         if (shp->shm_segsz < params->u.size)
 465                 return -EINVAL;
 466
 467         return 0;
 468 }
 469
 470 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 471 {
 472         struct ipc_namespace *ns;
 473         struct ipc_ops shm_ops;
 474         struct ipc_params shm_params;
 475
 476         ns = current->nsproxy->ipc_ns;
 477
 478         shm_ops.getnew = newseg;
 479         shm_ops.associate = shm_security;
 480         shm_ops.more_checks = shm_more_checks;
 481
 482         shm_params.key = key;
 483         shm_params.flg = shmflg;
 484         shm_params.u.size = size;
 485
 486         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 487 }
 488
 489 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 490 {
 491         switch(version) {
 492         case IPC_64:
 493                 return copy_to_user(buf, in, sizeof(*in));
 494         case IPC_OLD:
 495             {
 496                 struct shmid_ds out;
 497
 498                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 499                 out.shm_segsz   = in->shm_segsz;
 500                 out.shm_atime   = in->shm_atime;
 501                 out.shm_dtime   = in->shm_dtime;
 502                 out.shm_ctime   = in->shm_ctime;
 503                 out.shm_cpid    = in->shm_cpid;
 504                 out.shm_lpid    = in->shm_lpid;
 505                 out.shm_nattch  = in->shm_nattch;
 506
 507                 return copy_to_user(buf, &out, sizeof(out));
 508             }
 509         default:
 510                 return -EINVAL;
 511         }
 512 }
 513
 514 static inline unsigned long
 515 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 516 {
 517         switch(version) {
 518         case IPC_64:
 519                 if (copy_from_user(out, buf, sizeof(*out)))
 520                         return -EFAULT;
 521                 return 0;
 522         case IPC_OLD:
 523             {
 524                 struct shmid_ds tbuf_old;
 525
 526                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 527                         return -EFAULT;
 528
 529                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 530                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 531                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 532
 533                 return 0;
 534             }
 535         default:
 536                 return -EINVAL;
 537         }
 538 }
 539
 540 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 541 {
 542         switch(version) {
 543         case IPC_64:
 544                 return copy_to_user(buf, in, sizeof(*in));
 545         case IPC_OLD:
 546             {
 547                 struct shminfo out;
 548
 549                 if(in->shmmax > INT_MAX)
 550                         out.shmmax = INT_MAX;
 551                 else
 552                         out.shmmax = (int)in->shmmax;
 553
 554                 out.shmmin      = in->shmmin;
 555                 out.shmmni      = in->shmmni;
 556                 out.shmseg      = in->shmseg;
 557                 out.shmall      = in->shmall;
 558
 559                 return copy_to_user(buf, &out, sizeof(out));
 560             }
 561         default:
 562                 return -EINVAL;
 563         }
 564 }
 565
 566 /*
 567  * Called with shm_ids.rw_mutex held as a reader
 568  */
 569 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 570                 unsigned long *swp)
 571 {
 572         int next_id;
 573         int total, in_use;
 574
 575         *rss = 0;
 576         *swp = 0;
 577
 578         in_use = shm_ids(ns).in_use;
 579
 580         for (total = 0, next_id = 0; total < in_use; next_id++) {
 581                 struct shmid_kernel *shp;
 582                 struct inode *inode;
 583
 584                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 585                 if (shp == NULL)
 586                         continue;
 587
 588                 inode = shp->shm_file->f_path.dentry->d_inode;
 589
 590                 if (is_file_hugepages(shp->shm_file)) {
 591                         struct address_space *mapping = inode->i_mapping;
 592                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 593                 } else {
 594                         struct shmem_inode_info *info = SHMEM_I(inode);
 595                         spin_lock(&info->lock);
 596                         *rss += inode->i_mapping->nrpages;
 597                         *swp += info->swapped;
 598                         spin_unlock(&info->lock);
 599                 }
 600
 601                 total++;
 602         }
 603 }
 604
 605 /*
 606  * This function handles some shmctl commands which require the rw_mutex
 607  * to be held in write mode.
 608  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 609  */
 610 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 611                        struct shmid_ds __user *buf, int version)
 612 {
 613         struct kern_ipc_perm *ipcp;
 614         struct shmid64_ds shmid64;
 615         struct shmid_kernel *shp;
 616         int err;
 617
 618         if (cmd == IPC_SET) {
 619                 if (copy_shmid_from_user(&shmid64, buf, version))
 620                         return -EFAULT;
 621         }
 622
 623         down_write(&shm_ids(ns).rw_mutex);
 624         shp = shm_lock_check_down(ns, shmid);
 625         if (IS_ERR(shp)) {
 626                 err = PTR_ERR(shp);
 627                 goto out_up;
 628         }
 629
 630         ipcp = &shp->shm_perm;
 631
 632         err = audit_ipc_obj(ipcp);
 633         if (err)
 634                 goto out_unlock;
 635
 636         if (cmd == IPC_SET) {
 637                 err = audit_ipc_set_perm(0, shmid64.shm_perm.uid,
 638                                          shmid64.shm_perm.gid,
 639                                          shmid64.shm_perm.mode);
 640                 if (err)
 641                         goto out_unlock;
 642         }
 643
 644         if (current->euid != ipcp->uid &&
 645             current->euid != ipcp->cuid &&
 646             !capable(CAP_SYS_ADMIN)) {
 647                 err = -EPERM;
 648                 goto out_unlock;
 649         }
 650
 651         err = security_shm_shmctl(shp, cmd);
 652         if (err)
 653                 goto out_unlock;
 654         switch (cmd) {
 655         case IPC_RMID:
 656                 do_shm_rmid(ns, ipcp);
 657                 goto out_up;
 658         case IPC_SET:
 659                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 660                 shp->shm_ctim = get_seconds();
 661                 break;
 662         default:
 663                 err = -EINVAL;
 664         }
 665 out_unlock:
 666         shm_unlock(shp);
 667 out_up:
 668         up_write(&shm_ids(ns).rw_mutex);
 669         return err;
 670 }
 671
 672 asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
 673 {
 674         struct shmid_kernel *shp;
 675         int err, version;
 676         struct ipc_namespace *ns;
 677
 678         if (cmd < 0 || shmid < 0) {
 679                 err = -EINVAL;
 680                 goto out;
 681         }
 682
 683         version = ipc_parse_version(&cmd);
 684         ns = current->nsproxy->ipc_ns;
 685
 686         switch (cmd) { /* replace with proc interface ? */
 687         case IPC_INFO:
 688         {
 689                 struct shminfo64 shminfo;
 690
 691                 err = security_shm_shmctl(NULL, cmd);
 692                 if (err)
 693                         return err;
 694
 695                 memset(&shminfo,0,sizeof(shminfo));
 696                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 697                 shminfo.shmmax = ns->shm_ctlmax;
 698                 shminfo.shmall = ns->shm_ctlall;
 699
 700                 shminfo.shmmin = SHMMIN;
 701                 if(copy_shminfo_to_user (buf, &shminfo, version))
 702                         return -EFAULT;
 703
 704                 down_read(&shm_ids(ns).rw_mutex);
 705                 err = ipc_get_maxid(&shm_ids(ns));
 706                 up_read(&shm_ids(ns).rw_mutex);
 707
 708                 if(err<0)
 709                         err = 0;
 710                 goto out;
 711         }
 712         case SHM_INFO:
 713         {
 714                 struct shm_info shm_info;
 715
 716                 err = security_shm_shmctl(NULL, cmd);
 717                 if (err)
 718                         return err;
 719
 720                 memset(&shm_info,0,sizeof(shm_info));
 721                 down_read(&shm_ids(ns).rw_mutex);
 722                 shm_info.used_ids = shm_ids(ns).in_use;
 723                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 724                 shm_info.shm_tot = ns->shm_tot;
 725                 shm_info.swap_attempts = 0;
 726                 shm_info.swap_successes = 0;
 727                 err = ipc_get_maxid(&shm_ids(ns));
 728                 up_read(&shm_ids(ns).rw_mutex);
 729                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 730                         err = -EFAULT;
 731                         goto out;
 732                 }
 733
 734                 err = err < 0 ? 0 : err;
 735                 goto out;
 736         }
 737         case SHM_STAT:
 738         case IPC_STAT:
 739         {
 740                 struct shmid64_ds tbuf;
 741                 int result;
 742
 743                 if (!buf) {
 744                         err = -EFAULT;
 745                         goto out;
 746                 }
 747
 748                 if (cmd == SHM_STAT) {
 749                         shp = shm_lock(ns, shmid);
 750                         if (IS_ERR(shp)) {
 751                                 err = PTR_ERR(shp);
 752                                 goto out;
 753                         }
 754                         result = shp->shm_perm.id;
 755                 } else {
 756                         shp = shm_lock_check(ns, shmid);
 757                         if (IS_ERR(shp)) {
 758                                 err = PTR_ERR(shp);
 759                                 goto out;
 760                         }
 761                         result = 0;
 762                 }
 763                 err=-EACCES;
 764                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 765                         goto out_unlock;
 766                 err = security_shm_shmctl(shp, cmd);
 767                 if (err)
 768                         goto out_unlock;
 769                 memset(&tbuf, 0, sizeof(tbuf));
 770                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 771                 tbuf.shm_segsz  = shp->shm_segsz;
 772                 tbuf.shm_atime  = shp->shm_atim;
 773                 tbuf.shm_dtime  = shp->shm_dtim;
 774                 tbuf.shm_ctime  = shp->shm_ctim;
 775                 tbuf.shm_cpid   = shp->shm_cprid;
 776                 tbuf.shm_lpid   = shp->shm_lprid;
 777                 tbuf.shm_nattch = shp->shm_nattch;
 778                 shm_unlock(shp);
 779                 if(copy_shmid_to_user (buf, &tbuf, version))
 780                         err = -EFAULT;
 781                 else
 782                         err = result;
 783                 goto out;
 784         }
 785         case SHM_LOCK:
 786         case SHM_UNLOCK:
 787         {
 788                 shp = shm_lock_check(ns, shmid);
 789                 if (IS_ERR(shp)) {
 790                         err = PTR_ERR(shp);
 791                         goto out;
 792                 }
 793
 794                 err = audit_ipc_obj(&(shp->shm_perm));
 795                 if (err)
 796                         goto out_unlock;
 797
 798                 if (!capable(CAP_IPC_LOCK)) {
 799                         err = -EPERM;
 800                         if (current->euid != shp->shm_perm.uid &&
 801                             current->euid != shp->shm_perm.cuid)
 802                                 goto out_unlock;
 803                         if (cmd == SHM_LOCK &&
 804                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 805                                 goto out_unlock;
 806                 }
 807
 808                 err = security_shm_shmctl(shp, cmd);
 809                 if (err)
 810                         goto out_unlock;
 811
 812                 if(cmd==SHM_LOCK) {
 813                         struct user_struct * user = current->user;
 814                         if (!is_file_hugepages(shp->shm_file)) {
 815                                 err = shmem_lock(shp->shm_file, 1, user);
 816                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 817                                         shp->shm_perm.mode |= SHM_LOCKED;
 818                                         shp->mlock_user = user;
 819                                 }
 820                         }
 821                 } else if (!is_file_hugepages(shp->shm_file)) {
 822                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 823                         shp->shm_perm.mode &= ~SHM_LOCKED;
 824                         shp->mlock_user = NULL;
 825                 }
 826                 shm_unlock(shp);
 827                 goto out;
 828         }
 829         case IPC_RMID:
 830         case IPC_SET:
 831                 err = shmctl_down(ns, shmid, cmd, buf, version);
 832                 return err;
 833         default:
 834                 return -EINVAL;
 835         }
 836
 837 out_unlock:
 838         shm_unlock(shp);
 839 out:
 840         return err;
 841 }
 842
 843 /*
 844  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 845  *
 846  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 847  * "raddr" thing points to kernel space, and there has to be a wrapper around
 848  * this.
 849  */
 850 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 851 {
 852         struct shmid_kernel *shp;
 853         unsigned long addr;
 854         unsigned long size;
 855         struct file * file;
 856         int    err;
 857         unsigned long flags;
 858         unsigned long prot;
 859         int acc_mode;
 860         unsigned long user_addr;
 861         struct ipc_namespace *ns;
 862         struct shm_file_data *sfd;
 863         struct path path;
 864         mode_t f_mode;
 865
 866         err = -EINVAL;
 867         if (shmid < 0)
 868                 goto out;
 869         else if ((addr = (ulong)shmaddr)) {
 870                 if (addr & (SHMLBA-1)) {
 871                         if (shmflg & SHM_RND)
 872                                 addr &= ~(SHMLBA-1);       /* round down */
 873                         else
 874 #ifndef __ARCH_FORCE_SHMLBA
 875                                 if (addr & ~PAGE_MASK)
 876 #endif
 877                                         goto out;
 878                 }
 879                 flags = MAP_SHARED | MAP_FIXED;
 880         } else {
 881                 if ((shmflg & SHM_REMAP))
 882                         goto out;
 883
 884                 flags = MAP_SHARED;
 885         }
 886
 887         if (shmflg & SHM_RDONLY) {
 888                 prot = PROT_READ;
 889                 acc_mode = S_IRUGO;
 890                 f_mode = FMODE_READ;
 891         } else {
 892                 prot = PROT_READ | PROT_WRITE;
 893                 acc_mode = S_IRUGO | S_IWUGO;
 894                 f_mode = FMODE_READ | FMODE_WRITE;
 895         }
 896         if (shmflg & SHM_EXEC) {
 897                 prot |= PROT_EXEC;
 898                 acc_mode |= S_IXUGO;
 899         }
 900
 901         /*
 902          * We cannot rely on the fs check since SYSV IPC does have an
 903          * additional creator id...
 904          */
 905         ns = current->nsproxy->ipc_ns;
 906         shp = shm_lock_check(ns, shmid);
 907         if (IS_ERR(shp)) {
 908                 err = PTR_ERR(shp);
 909                 goto out;
 910         }
 911
 912         err = -EACCES;
 913         if (ipcperms(&shp->shm_perm, acc_mode))
 914                 goto out_unlock;
 915
 916         err = security_shm_shmat(shp, shmaddr, shmflg);
 917         if (err)
 918                 goto out_unlock;
 919
 920         path.dentry = dget(shp->shm_file->f_path.dentry);
 921         path.mnt    = shp->shm_file->f_path.mnt;
 922         shp->shm_nattch++;
 923         size = i_size_read(path.dentry->d_inode);
 924         shm_unlock(shp);
 925
 926         err = -ENOMEM;
 927         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 928         if (!sfd)
 929                 goto out_put_dentry;
 930
 931         err = -ENOMEM;
 932
 933         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 934         if (!file)
 935                 goto out_free;
 936
 937         file->private_data = sfd;
 938         file->f_mapping = shp->shm_file->f_mapping;
 939         sfd->id = shp->shm_perm.id;
 940         sfd->ns = get_ipc_ns(ns);
 941         sfd->file = shp->shm_file;
 942         sfd->vm_ops = NULL;
 943
 944         down_write(&current->mm->mmap_sem);
 945         if (addr && !(shmflg & SHM_REMAP)) {
 946                 err = -EINVAL;
 947                 if (find_vma_intersection(current->mm, addr, addr + size))
 948                         goto invalid;
 949                 /*
 950                  * If shm segment goes below stack, make sure there is some
 951                  * space left for the stack to grow (at least 4 pages).
 952                  */
 953                 if (addr < current->mm->start_stack &&
 954                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 955                         goto invalid;
 956         }
 957
 958         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 959         *raddr = user_addr;
 960         err = 0;
 961         if (IS_ERR_VALUE(user_addr))
 962                 err = (long)user_addr;
 963 invalid:
 964         up_write(&current->mm->mmap_sem);
 965
 966         fput(file);
 967
 968 out_nattch:
 969         down_write(&shm_ids(ns).rw_mutex);
 970         shp = shm_lock_down(ns, shmid);
 971         BUG_ON(IS_ERR(shp));
 972         shp->shm_nattch--;
 973         if(shp->shm_nattch == 0 &&
 974            shp->shm_perm.mode & SHM_DEST)
 975                 shm_destroy(ns, shp);
 976         else
 977                 shm_unlock(shp);
 978         up_write(&shm_ids(ns).rw_mutex);
 979
 980 out:
 981         return err;
 982
 983 out_unlock:
 984         shm_unlock(shp);
 985         goto out;
 986
 987 out_free:
 988         kfree(sfd);
 989 out_put_dentry:
 990         dput(path.dentry);
 991         goto out_nattch;
 992 }
 993
 994 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 995 {
 996         unsigned long ret;
 997         long err;
 998
 999         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1000         if (err)
1001                 return err;
1002         force_successful_syscall_return();
1003         return (long)ret;
1004 }
1005
1006 /*
1007  * detach and kill segment if marked destroyed.
1008  * The work is done in shm_close.
1009  */
1010 asmlinkage long sys_shmdt(char __user *shmaddr)
1011 {
1012         struct mm_struct *mm = current->mm;
1013         struct vm_area_struct *vma, *next;
1014         unsigned long addr = (unsigned long)shmaddr;
1015         loff_t size = 0;
1016         int retval = -EINVAL;
1017
1018         if (addr & ~PAGE_MASK)
1019                 return retval;
1020
1021         down_write(&mm->mmap_sem);
1022
1023         /*
1024          * This function tries to be smart and unmap shm segments that
1025          * were modified by partial mlock or munmap calls:
1026          * - It first determines the size of the shm segment that should be
1027          *   unmapped: It searches for a vma that is backed by shm and that
1028          *   started at address shmaddr. It records it's size and then unmaps
1029          *   it.
1030          * - Then it unmaps all shm vmas that started at shmaddr and that
1031          *   are within the initially determined size.
1032          * Errors from do_munmap are ignored: the function only fails if
1033          * it's called with invalid parameters or if it's called to unmap
1034          * a part of a vma. Both calls in this function are for full vmas,
1035          * the parameters are directly copied from the vma itself and always
1036          * valid - therefore do_munmap cannot fail. (famous last words?)
1037          */
1038         /*
1039          * If it had been mremap()'d, the starting address would not
1040          * match the usual checks anyway. So assume all vma's are
1041          * above the starting address given.
1042          */
1043         vma = find_vma(mm, addr);
1044
1045         while (vma) {
1046                 next = vma->vm_next;
1047
1048                 /*
1049                  * Check if the starting address would match, i.e. it's
1050                  * a fragment created by mprotect() and/or munmap(), or it
1051                  * otherwise it starts at this address with no hassles.
1052                  */
1053                 if ((vma->vm_ops == &shm_vm_ops) &&
1054                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1055
1056
1057                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1058                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1059                         /*
1060                          * We discovered the size of the shm segment, so
1061                          * break out of here and fall through to the next
1062                          * loop that uses the size information to stop
1063                          * searching for matching vma's.
1064                          */
1065                         retval = 0;
1066                         vma = next;
1067                         break;
1068                 }
1069                 vma = next;
1070         }
1071
1072         /*
1073          * We need look no further than the maximum address a fragment
1074          * could possibly have landed at. Also cast things to loff_t to
1075          * prevent overflows and make comparisions vs. equal-width types.
1076          */
1077         size = PAGE_ALIGN(size);
1078         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1079                 next = vma->vm_next;
1080
1081                 /* finding a matching vma now does not alter retval */
1082                 if ((vma->vm_ops == &shm_vm_ops) &&
1083                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1084
1085                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1086                 vma = next;
1087         }
1088
1089         up_write(&mm->mmap_sem);
1090         return retval;
1091 }
1092
1093 #ifdef CONFIG_PROC_FS
1094 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1095 {
1096         struct shmid_kernel *shp = it;
1097         char *format;
1098
1099 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1100 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1101
1102         if (sizeof(size_t) <= sizeof(int))
1103                 format = SMALL_STRING;
1104         else
1105                 format = BIG_STRING;
1106         return seq_printf(s, format,
1107                           shp->shm_perm.key,
1108                           shp->shm_perm.id,
1109                           shp->shm_perm.mode,
1110                           shp->shm_segsz,
1111                           shp->shm_cprid,
1112                           shp->shm_lprid,
1113                           shp->shm_nattch,
1114                           shp->shm_perm.uid,
1115                           shp->shm_perm.gid,
1116                           shp->shm_perm.cuid,
1117                           shp->shm_perm.cgid,
1118                           shp->shm_atim,
1119                           shp->shm_dtim,
1120                           shp->shm_ctim);
1121 }
1122 #endif