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/mutex.h>
  39 #include <linux/nsproxy.h>
  40 #include <linux/mount.h>
  41
  42 #include <asm/uaccess.h>
  43
  44 #include "util.h"
  45
  46 struct shm_file_data {
  47         int id;
  48         struct ipc_namespace *ns;
  49         struct file *file;
  50         const struct vm_operations_struct *vm_ops;
  51 };
  52
  53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  54
  55 static const struct file_operations shm_file_operations;
  56 static struct vm_operations_struct shm_vm_ops;
  57
  58 static struct ipc_ids init_shm_ids;
  59
  60 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
  61
  62 #define shm_lock(ns, id)                \
  63         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
  64 #define shm_unlock(shp)                 \
  65         ipc_unlock(&(shp)->shm_perm)
  66 #define shm_get(ns, id)                 \
  67         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
  68 #define shm_buildid(ns, id, seq)        \
  69         ipc_buildid(&shm_ids(ns), id, seq)
  70
  71 static int newseg(struct ipc_namespace *, struct ipc_params *);
  72 static void shm_open(struct vm_area_struct *vma);
  73 static void shm_close(struct vm_area_struct *vma);
  74 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  75 #ifdef CONFIG_PROC_FS
  76 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  77 #endif
  78
  79 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  80 {
  81         ns->ids[IPC_SHM_IDS] = ids;
  82         ns->shm_ctlmax = SHMMAX;
  83         ns->shm_ctlall = SHMALL;
  84         ns->shm_ctlmni = SHMMNI;
  85         ns->shm_tot = 0;
  86         ipc_init_ids(ids);
  87 }
  88
  89 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  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 int shm_init_ns(struct ipc_namespace *ns)
 101 {
 102         struct ipc_ids *ids;
 103
 104         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 105         if (ids == NULL)
 106                 return -ENOMEM;
 107
 108         __shm_init_ns(ns, ids);
 109         return 0;
 110 }
 111
 112 void shm_exit_ns(struct ipc_namespace *ns)
 113 {
 114         struct shmid_kernel *shp;
 115         int next_id;
 116         int total, in_use;
 117
 118         mutex_lock(&shm_ids(ns).mutex);
 119
 120         in_use = shm_ids(ns).in_use;
 121
 122         for (total = 0, next_id = 0; total < in_use; next_id++) {
 123                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 124                 if (shp == NULL)
 125                         continue;
 126                 ipc_lock_by_ptr(&shp->shm_perm);
 127                 do_shm_rmid(ns, shp);
 128                 total++;
 129         }
 130         mutex_unlock(&shm_ids(ns).mutex);
 131
 132         kfree(ns->ids[IPC_SHM_IDS]);
 133         ns->ids[IPC_SHM_IDS] = NULL;
 134 }
 135
 136 void __init shm_init (void)
 137 {
 138         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 139         ipc_init_proc_interface("sysvipc/shm",
 140                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 141                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 142 }
 143
 144 static inline int shm_checkid(struct ipc_namespace *ns,
 145                 struct shmid_kernel *s, int id)
 146 {
 147         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
 148                 return -EIDRM;
 149         return 0;
 150 }
 151
 152 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 153 {
 154         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 155 }
 156
 157 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 158 {
 159         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 160 }
 161
 162
 163
 164 /* This is called by fork, once for every shm attach. */
 165 static void shm_open(struct vm_area_struct *vma)
 166 {
 167         struct file *file = vma->vm_file;
 168         struct shm_file_data *sfd = shm_file_data(file);
 169         struct shmid_kernel *shp;
 170
 171         shp = shm_lock(sfd->ns, sfd->id);
 172         BUG_ON(!shp);
 173         shp->shm_atim = get_seconds();
 174         shp->shm_lprid = task_tgid_vnr(current);
 175         shp->shm_nattch++;
 176         shm_unlock(shp);
 177 }
 178
 179 /*
 180  * shm_destroy - free the struct shmid_kernel
 181  *
 182  * @shp: struct to free
 183  *
 184  * It has to be called with shp and shm_ids.mutex locked,
 185  * but returns with shp unlocked and freed.
 186  */
 187 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 188 {
 189         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 190         shm_rmid(ns, shp);
 191         shm_unlock(shp);
 192         if (!is_file_hugepages(shp->shm_file))
 193                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 194         else
 195                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 196                                                 shp->mlock_user);
 197         fput (shp->shm_file);
 198         security_shm_free(shp);
 199         ipc_rcu_putref(shp);
 200 }
 201
 202 /*
 203  * remove the attach descriptor vma.
 204  * free memory for segment if it is marked destroyed.
 205  * The descriptor has already been removed from the current->mm->mmap list
 206  * and will later be kfree()d.
 207  */
 208 static void shm_close(struct vm_area_struct *vma)
 209 {
 210         struct file * file = vma->vm_file;
 211         struct shm_file_data *sfd = shm_file_data(file);
 212         struct shmid_kernel *shp;
 213         struct ipc_namespace *ns = sfd->ns;
 214
 215         mutex_lock(&shm_ids(ns).mutex);
 216         /* remove from the list of attaches of the shm segment */
 217         shp = shm_lock(ns, sfd->id);
 218         BUG_ON(!shp);
 219         shp->shm_lprid = task_tgid_vnr(current);
 220         shp->shm_dtim = get_seconds();
 221         shp->shm_nattch--;
 222         if(shp->shm_nattch == 0 &&
 223            shp->shm_perm.mode & SHM_DEST)
 224                 shm_destroy(ns, shp);
 225         else
 226                 shm_unlock(shp);
 227         mutex_unlock(&shm_ids(ns).mutex);
 228 }
 229
 230 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 231 {
 232         struct file *file = vma->vm_file;
 233         struct shm_file_data *sfd = shm_file_data(file);
 234
 235         return sfd->vm_ops->fault(vma, vmf);
 236 }
 237
 238 #ifdef CONFIG_NUMA
 239 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 240 {
 241         struct file *file = vma->vm_file;
 242         struct shm_file_data *sfd = shm_file_data(file);
 243         int err = 0;
 244         if (sfd->vm_ops->set_policy)
 245                 err = sfd->vm_ops->set_policy(vma, new);
 246         return err;
 247 }
 248
 249 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 250                                         unsigned long addr)
 251 {
 252         struct file *file = vma->vm_file;
 253         struct shm_file_data *sfd = shm_file_data(file);
 254         struct mempolicy *pol = NULL;
 255
 256         if (sfd->vm_ops->get_policy)
 257                 pol = sfd->vm_ops->get_policy(vma, addr);
 258         else if (vma->vm_policy)
 259                 pol = vma->vm_policy;
 260         else
 261                 pol = current->mempolicy;
 262         return pol;
 263 }
 264 #endif
 265
 266 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 267 {
 268         struct shm_file_data *sfd = shm_file_data(file);
 269         int ret;
 270
 271         ret = sfd->file->f_op->mmap(sfd->file, vma);
 272         if (ret != 0)
 273                 return ret;
 274         sfd->vm_ops = vma->vm_ops;
 275 #ifdef CONFIG_MMU
 276         BUG_ON(!sfd->vm_ops->fault);
 277 #endif
 278         vma->vm_ops = &shm_vm_ops;
 279         shm_open(vma);
 280
 281         return ret;
 282 }
 283
 284 static int shm_release(struct inode *ino, struct file *file)
 285 {
 286         struct shm_file_data *sfd = shm_file_data(file);
 287
 288         put_ipc_ns(sfd->ns);
 289         shm_file_data(file) = NULL;
 290         kfree(sfd);
 291         return 0;
 292 }
 293
 294 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 295 {
 296         int (*fsync) (struct file *, struct dentry *, int datasync);
 297         struct shm_file_data *sfd = shm_file_data(file);
 298         int ret = -EINVAL;
 299
 300         fsync = sfd->file->f_op->fsync;
 301         if (fsync)
 302                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 303         return ret;
 304 }
 305
 306 static unsigned long shm_get_unmapped_area(struct file *file,
 307         unsigned long addr, unsigned long len, unsigned long pgoff,
 308         unsigned long flags)
 309 {
 310         struct shm_file_data *sfd = shm_file_data(file);
 311         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 312 }
 313
 314 int is_file_shm_hugepages(struct file *file)
 315 {
 316         int ret = 0;
 317
 318         if (file->f_op == &shm_file_operations) {
 319                 struct shm_file_data *sfd;
 320                 sfd = shm_file_data(file);
 321                 ret = is_file_hugepages(sfd->file);
 322         }
 323         return ret;
 324 }
 325
 326 static const struct file_operations shm_file_operations = {
 327         .mmap           = shm_mmap,
 328         .fsync          = shm_fsync,
 329         .release        = shm_release,
 330         .get_unmapped_area      = shm_get_unmapped_area,
 331 };
 332
 333 static struct vm_operations_struct shm_vm_ops = {
 334         .open   = shm_open,     /* callback for a new vm-area open */
 335         .close  = shm_close,    /* callback for when the vm-area is released */
 336         .fault  = shm_fault,
 337 #if defined(CONFIG_NUMA)
 338         .set_policy = shm_set_policy,
 339         .get_policy = shm_get_policy,
 340 #endif
 341 };
 342
 343 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 344 {
 345         key_t key = params->key;
 346         int shmflg = params->flg;
 347         size_t size = params->u.size;
 348         int error;
 349         struct shmid_kernel *shp;
 350         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 351         struct file * file;
 352         char name[13];
 353         int id;
 354
 355         if (size < SHMMIN || size > ns->shm_ctlmax)
 356                 return -EINVAL;
 357
 358         if (ns->shm_tot + numpages > ns->shm_ctlall)
 359                 return -ENOSPC;
 360
 361         shp = ipc_rcu_alloc(sizeof(*shp));
 362         if (!shp)
 363                 return -ENOMEM;
 364
 365         shp->shm_perm.key = key;
 366         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 367         shp->mlock_user = NULL;
 368
 369         shp->shm_perm.security = NULL;
 370         error = security_shm_alloc(shp);
 371         if (error) {
 372                 ipc_rcu_putref(shp);
 373                 return error;
 374         }
 375
 376         sprintf (name, "SYSV%08x", key);
 377         if (shmflg & SHM_HUGETLB) {
 378                 /* hugetlb_file_setup takes care of mlock user accounting */
 379                 file = hugetlb_file_setup(name, size);
 380                 shp->mlock_user = current->user;
 381         } else {
 382                 int acctflag = VM_ACCOUNT;
 383                 /*
 384                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 385                  * if it's asked for.
 386                  */
 387                 if  ((shmflg & SHM_NORESERVE) &&
 388                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 389                         acctflag = 0;
 390                 file = shmem_file_setup(name, size, acctflag);
 391         }
 392         error = PTR_ERR(file);
 393         if (IS_ERR(file))
 394                 goto no_file;
 395
 396         error = -ENOSPC;
 397         id = shm_addid(ns, shp);
 398         if(id == -1)
 399                 goto no_id;
 400
 401         shp->shm_cprid = task_tgid_vnr(current);
 402         shp->shm_lprid = 0;
 403         shp->shm_atim = shp->shm_dtim = 0;
 404         shp->shm_ctim = get_seconds();
 405         shp->shm_segsz = size;
 406         shp->shm_nattch = 0;
 407         shp->shm_perm.id = shm_buildid(ns, id, shp->shm_perm.seq);
 408         shp->shm_file = file;
 409         /*
 410          * shmid gets reported as "inode#" in /proc/pid/maps.
 411          * proc-ps tools use this. Changing this will break them.
 412          */
 413         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 414
 415         ns->shm_tot += numpages;
 416         error = shp->shm_perm.id;
 417         shm_unlock(shp);
 418         return error;
 419
 420 no_id:
 421         fput(file);
 422 no_file:
 423         security_shm_free(shp);
 424         ipc_rcu_putref(shp);
 425         return error;
 426 }
 427
 428 static inline int shm_security(void *shp, int shmflg)
 429 {
 430         return security_shm_associate((struct shmid_kernel *) shp, shmflg);
 431 }
 432
 433 static inline int shm_more_checks(void *shp, struct ipc_params *params)
 434 {
 435         if (((struct shmid_kernel *)shp)->shm_segsz < params->u.size)
 436                 return -EINVAL;
 437
 438         return 0;
 439 }
 440
 441 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 442 {
 443         struct ipc_namespace *ns;
 444         struct ipc_ops shm_ops;
 445         struct ipc_params shm_params;
 446
 447         ns = current->nsproxy->ipc_ns;
 448
 449         shm_ops.getnew = newseg;
 450         shm_ops.associate = shm_security;
 451         shm_ops.more_checks = shm_more_checks;
 452
 453         shm_params.key = key;
 454         shm_params.flg = shmflg;
 455         shm_params.u.size = size;
 456
 457         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 458 }
 459
 460 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 461 {
 462         switch(version) {
 463         case IPC_64:
 464                 return copy_to_user(buf, in, sizeof(*in));
 465         case IPC_OLD:
 466             {
 467                 struct shmid_ds out;
 468
 469                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 470                 out.shm_segsz   = in->shm_segsz;
 471                 out.shm_atime   = in->shm_atime;
 472                 out.shm_dtime   = in->shm_dtime;
 473                 out.shm_ctime   = in->shm_ctime;
 474                 out.shm_cpid    = in->shm_cpid;
 475                 out.shm_lpid    = in->shm_lpid;
 476                 out.shm_nattch  = in->shm_nattch;
 477
 478                 return copy_to_user(buf, &out, sizeof(out));
 479             }
 480         default:
 481                 return -EINVAL;
 482         }
 483 }
 484
 485 struct shm_setbuf {
 486         uid_t   uid;
 487         gid_t   gid;
 488         mode_t  mode;
 489 };
 490
 491 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 492 {
 493         switch(version) {
 494         case IPC_64:
 495             {
 496                 struct shmid64_ds tbuf;
 497
 498                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 499                         return -EFAULT;
 500
 501                 out->uid        = tbuf.shm_perm.uid;
 502                 out->gid        = tbuf.shm_perm.gid;
 503                 out->mode       = tbuf.shm_perm.mode;
 504
 505                 return 0;
 506             }
 507         case IPC_OLD:
 508             {
 509                 struct shmid_ds tbuf_old;
 510
 511                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 512                         return -EFAULT;
 513
 514                 out->uid        = tbuf_old.shm_perm.uid;
 515                 out->gid        = tbuf_old.shm_perm.gid;
 516                 out->mode       = tbuf_old.shm_perm.mode;
 517
 518                 return 0;
 519             }
 520         default:
 521                 return -EINVAL;
 522         }
 523 }
 524
 525 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 526 {
 527         switch(version) {
 528         case IPC_64:
 529                 return copy_to_user(buf, in, sizeof(*in));
 530         case IPC_OLD:
 531             {
 532                 struct shminfo out;
 533
 534                 if(in->shmmax > INT_MAX)
 535                         out.shmmax = INT_MAX;
 536                 else
 537                         out.shmmax = (int)in->shmmax;
 538
 539                 out.shmmin      = in->shmmin;
 540                 out.shmmni      = in->shmmni;
 541                 out.shmseg      = in->shmseg;
 542                 out.shmall      = in->shmall;
 543
 544                 return copy_to_user(buf, &out, sizeof(out));
 545             }
 546         default:
 547                 return -EINVAL;
 548         }
 549 }
 550
 551 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 552                 unsigned long *swp)
 553 {
 554         int next_id;
 555         int total, in_use;
 556
 557         *rss = 0;
 558         *swp = 0;
 559
 560         in_use = shm_ids(ns).in_use;
 561
 562         for (total = 0, next_id = 0; total < in_use; next_id++) {
 563                 struct shmid_kernel *shp;
 564                 struct inode *inode;
 565
 566                 shp = shm_get(ns, next_id);
 567                 if (shp == NULL)
 568                         continue;
 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                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 575                 } else {
 576                         struct shmem_inode_info *info = SHMEM_I(inode);
 577                         spin_lock(&info->lock);
 578                         *rss += inode->i_mapping->nrpages;
 579                         *swp += info->swapped;
 580                         spin_unlock(&info->lock);
 581                 }
 582
 583                 total++;
 584         }
 585 }
 586
 587 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 588 {
 589         struct shm_setbuf setbuf;
 590         struct shmid_kernel *shp;
 591         int err, version;
 592         struct ipc_namespace *ns;
 593
 594         if (cmd < 0 || shmid < 0) {
 595                 err = -EINVAL;
 596                 goto out;
 597         }
 598
 599         version = ipc_parse_version(&cmd);
 600         ns = current->nsproxy->ipc_ns;
 601
 602         switch (cmd) { /* replace with proc interface ? */
 603         case IPC_INFO:
 604         {
 605                 struct shminfo64 shminfo;
 606
 607                 err = security_shm_shmctl(NULL, cmd);
 608                 if (err)
 609                         return err;
 610
 611                 memset(&shminfo,0,sizeof(shminfo));
 612                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 613                 shminfo.shmmax = ns->shm_ctlmax;
 614                 shminfo.shmall = ns->shm_ctlall;
 615
 616                 shminfo.shmmin = SHMMIN;
 617                 if(copy_shminfo_to_user (buf, &shminfo, version))
 618                         return -EFAULT;
 619                 /* reading a integer is always atomic */
 620                 err = ipc_get_maxid(&shm_ids(ns));
 621                 if(err<0)
 622                         err = 0;
 623                 goto out;
 624         }
 625         case SHM_INFO:
 626         {
 627                 struct shm_info shm_info;
 628
 629                 err = security_shm_shmctl(NULL, cmd);
 630                 if (err)
 631                         return err;
 632
 633                 memset(&shm_info,0,sizeof(shm_info));
 634                 mutex_lock(&shm_ids(ns).mutex);
 635                 shm_info.used_ids = shm_ids(ns).in_use;
 636                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 637                 shm_info.shm_tot = ns->shm_tot;
 638                 shm_info.swap_attempts = 0;
 639                 shm_info.swap_successes = 0;
 640                 err = ipc_get_maxid(&shm_ids(ns));
 641                 mutex_unlock(&shm_ids(ns).mutex);
 642                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 643                         err = -EFAULT;
 644                         goto out;
 645                 }
 646
 647                 err = err < 0 ? 0 : err;
 648                 goto out;
 649         }
 650         case SHM_STAT:
 651         case IPC_STAT:
 652         {
 653                 struct shmid64_ds tbuf;
 654                 int result;
 655                 memset(&tbuf, 0, sizeof(tbuf));
 656                 shp = shm_lock(ns, shmid);
 657                 if(shp==NULL) {
 658                         err = -EINVAL;
 659                         goto out;
 660                 } else if (cmd == SHM_STAT) {
 661                         result = shp->shm_perm.id;
 662                 } else {
 663                         err = shm_checkid(ns, shp,shmid);
 664                         if(err)
 665                                 goto out_unlock;
 666                         result = 0;
 667                 }
 668                 err=-EACCES;
 669                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 670                         goto out_unlock;
 671                 err = security_shm_shmctl(shp, cmd);
 672                 if (err)
 673                         goto out_unlock;
 674                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 675                 tbuf.shm_segsz  = shp->shm_segsz;
 676                 tbuf.shm_atime  = shp->shm_atim;
 677                 tbuf.shm_dtime  = shp->shm_dtim;
 678                 tbuf.shm_ctime  = shp->shm_ctim;
 679                 tbuf.shm_cpid   = shp->shm_cprid;
 680                 tbuf.shm_lpid   = shp->shm_lprid;
 681                 tbuf.shm_nattch = shp->shm_nattch;
 682                 shm_unlock(shp);
 683                 if(copy_shmid_to_user (buf, &tbuf, version))
 684                         err = -EFAULT;
 685                 else
 686                         err = result;
 687                 goto out;
 688         }
 689         case SHM_LOCK:
 690         case SHM_UNLOCK:
 691         {
 692                 shp = shm_lock(ns, shmid);
 693                 if(shp==NULL) {
 694                         err = -EINVAL;
 695                         goto out;
 696                 }
 697                 err = shm_checkid(ns, shp,shmid);
 698                 if(err)
 699                         goto out_unlock;
 700
 701                 err = audit_ipc_obj(&(shp->shm_perm));
 702                 if (err)
 703                         goto out_unlock;
 704
 705                 if (!capable(CAP_IPC_LOCK)) {
 706                         err = -EPERM;
 707                         if (current->euid != shp->shm_perm.uid &&
 708                             current->euid != shp->shm_perm.cuid)
 709                                 goto out_unlock;
 710                         if (cmd == SHM_LOCK &&
 711                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 712                                 goto out_unlock;
 713                 }
 714
 715                 err = security_shm_shmctl(shp, cmd);
 716                 if (err)
 717                         goto out_unlock;
 718
 719                 if(cmd==SHM_LOCK) {
 720                         struct user_struct * user = current->user;
 721                         if (!is_file_hugepages(shp->shm_file)) {
 722                                 err = shmem_lock(shp->shm_file, 1, user);
 723                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 724                                         shp->shm_perm.mode |= SHM_LOCKED;
 725                                         shp->mlock_user = user;
 726                                 }
 727                         }
 728                 } else if (!is_file_hugepages(shp->shm_file)) {
 729                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 730                         shp->shm_perm.mode &= ~SHM_LOCKED;
 731                         shp->mlock_user = NULL;
 732                 }
 733                 shm_unlock(shp);
 734                 goto out;
 735         }
 736         case IPC_RMID:
 737         {
 738                 /*
 739                  *      We cannot simply remove the file. The SVID states
 740                  *      that the block remains until the last person
 741                  *      detaches from it, then is deleted. A shmat() on
 742                  *      an RMID segment is legal in older Linux and if
 743                  *      we change it apps break...
 744                  *
 745                  *      Instead we set a destroyed flag, and then blow
 746                  *      the name away when the usage hits zero.
 747                  */
 748                 mutex_lock(&shm_ids(ns).mutex);
 749                 shp = shm_lock(ns, shmid);
 750                 err = -EINVAL;
 751                 if (shp == NULL)
 752                         goto out_up;
 753                 err = shm_checkid(ns, shp, shmid);
 754                 if(err)
 755                         goto out_unlock_up;
 756
 757                 err = audit_ipc_obj(&(shp->shm_perm));
 758                 if (err)
 759                         goto out_unlock_up;
 760
 761                 if (current->euid != shp->shm_perm.uid &&
 762                     current->euid != shp->shm_perm.cuid &&
 763                     !capable(CAP_SYS_ADMIN)) {
 764                         err=-EPERM;
 765                         goto out_unlock_up;
 766                 }
 767
 768                 err = security_shm_shmctl(shp, cmd);
 769                 if (err)
 770                         goto out_unlock_up;
 771
 772                 do_shm_rmid(ns, shp);
 773                 mutex_unlock(&shm_ids(ns).mutex);
 774                 goto out;
 775         }
 776
 777         case IPC_SET:
 778         {
 779                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 780                         err = -EFAULT;
 781                         goto out;
 782                 }
 783                 mutex_lock(&shm_ids(ns).mutex);
 784                 shp = shm_lock(ns, shmid);
 785                 err=-EINVAL;
 786                 if(shp==NULL)
 787                         goto out_up;
 788                 err = shm_checkid(ns, shp,shmid);
 789                 if(err)
 790                         goto out_unlock_up;
 791                 err = audit_ipc_obj(&(shp->shm_perm));
 792                 if (err)
 793                         goto out_unlock_up;
 794                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 795                 if (err)
 796                         goto out_unlock_up;
 797                 err=-EPERM;
 798                 if (current->euid != shp->shm_perm.uid &&
 799                     current->euid != shp->shm_perm.cuid &&
 800                     !capable(CAP_SYS_ADMIN)) {
 801                         goto out_unlock_up;
 802                 }
 803
 804                 err = security_shm_shmctl(shp, cmd);
 805                 if (err)
 806                         goto out_unlock_up;
 807
 808                 shp->shm_perm.uid = setbuf.uid;
 809                 shp->shm_perm.gid = setbuf.gid;
 810                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 811                         | (setbuf.mode & S_IRWXUGO);
 812                 shp->shm_ctim = get_seconds();
 813                 break;
 814         }
 815
 816         default:
 817                 err = -EINVAL;
 818                 goto out;
 819         }
 820
 821         err = 0;
 822 out_unlock_up:
 823         shm_unlock(shp);
 824 out_up:
 825         mutex_unlock(&shm_ids(ns).mutex);
 826         goto out;
 827 out_unlock:
 828         shm_unlock(shp);
 829 out:
 830         return err;
 831 }
 832
 833 /*
 834  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 835  *
 836  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 837  * "raddr" thing points to kernel space, and there has to be a wrapper around
 838  * this.
 839  */
 840 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 841 {
 842         struct shmid_kernel *shp;
 843         unsigned long addr;
 844         unsigned long size;
 845         struct file * file;
 846         int    err;
 847         unsigned long flags;
 848         unsigned long prot;
 849         int acc_mode;
 850         unsigned long user_addr;
 851         struct ipc_namespace *ns;
 852         struct shm_file_data *sfd;
 853         struct path path;
 854         mode_t f_mode;
 855
 856         err = -EINVAL;
 857         if (shmid < 0)
 858                 goto out;
 859         else if ((addr = (ulong)shmaddr)) {
 860                 if (addr & (SHMLBA-1)) {
 861                         if (shmflg & SHM_RND)
 862                                 addr &= ~(SHMLBA-1);       /* round down */
 863                         else
 864 #ifndef __ARCH_FORCE_SHMLBA
 865                                 if (addr & ~PAGE_MASK)
 866 #endif
 867                                         goto out;
 868                 }
 869                 flags = MAP_SHARED | MAP_FIXED;
 870         } else {
 871                 if ((shmflg & SHM_REMAP))
 872                         goto out;
 873
 874                 flags = MAP_SHARED;
 875         }
 876
 877         if (shmflg & SHM_RDONLY) {
 878                 prot = PROT_READ;
 879                 acc_mode = S_IRUGO;
 880                 f_mode = FMODE_READ;
 881         } else {
 882                 prot = PROT_READ | PROT_WRITE;
 883                 acc_mode = S_IRUGO | S_IWUGO;
 884                 f_mode = FMODE_READ | FMODE_WRITE;
 885         }
 886         if (shmflg & SHM_EXEC) {
 887                 prot |= PROT_EXEC;
 888                 acc_mode |= S_IXUGO;
 889         }
 890
 891         /*
 892          * We cannot rely on the fs check since SYSV IPC does have an
 893          * additional creator id...
 894          */
 895         ns = current->nsproxy->ipc_ns;
 896         shp = shm_lock(ns, shmid);
 897         if(shp == NULL)
 898                 goto out;
 899
 900         err = shm_checkid(ns, shp,shmid);
 901         if (err)
 902                 goto out_unlock;
 903
 904         err = -EACCES;
 905         if (ipcperms(&shp->shm_perm, acc_mode))
 906                 goto out_unlock;
 907
 908         err = security_shm_shmat(shp, shmaddr, shmflg);
 909         if (err)
 910                 goto out_unlock;
 911
 912         path.dentry = dget(shp->shm_file->f_path.dentry);
 913         path.mnt    = shp->shm_file->f_path.mnt;
 914         shp->shm_nattch++;
 915         size = i_size_read(path.dentry->d_inode);
 916         shm_unlock(shp);
 917
 918         err = -ENOMEM;
 919         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 920         if (!sfd)
 921                 goto out_put_dentry;
 922
 923         err = -ENOMEM;
 924
 925         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 926         if (!file)
 927                 goto out_free;
 928
 929         file->private_data = sfd;
 930         file->f_mapping = shp->shm_file->f_mapping;
 931         sfd->id = shp->shm_perm.id;
 932         sfd->ns = get_ipc_ns(ns);
 933         sfd->file = shp->shm_file;
 934         sfd->vm_ops = NULL;
 935
 936         down_write(&current->mm->mmap_sem);
 937         if (addr && !(shmflg & SHM_REMAP)) {
 938                 err = -EINVAL;
 939                 if (find_vma_intersection(current->mm, addr, addr + size))
 940                         goto invalid;
 941                 /*
 942                  * If shm segment goes below stack, make sure there is some
 943                  * space left for the stack to grow (at least 4 pages).
 944                  */
 945                 if (addr < current->mm->start_stack &&
 946                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 947                         goto invalid;
 948         }
 949
 950         user_addr = do_mmap (file, addr, size, prot, flags, 0);
 951         *raddr = user_addr;
 952         err = 0;
 953         if (IS_ERR_VALUE(user_addr))
 954                 err = (long)user_addr;
 955 invalid:
 956         up_write(&current->mm->mmap_sem);
 957
 958         fput(file);
 959
 960 out_nattch:
 961         mutex_lock(&shm_ids(ns).mutex);
 962         shp = shm_lock(ns, shmid);
 963         BUG_ON(!shp);
 964         shp->shm_nattch--;
 965         if(shp->shm_nattch == 0 &&
 966            shp->shm_perm.mode & SHM_DEST)
 967                 shm_destroy(ns, shp);
 968         else
 969                 shm_unlock(shp);
 970         mutex_unlock(&shm_ids(ns).mutex);
 971
 972 out:
 973         return err;
 974
 975 out_unlock:
 976         shm_unlock(shp);
 977         goto out;
 978
 979 out_free:
 980         kfree(sfd);
 981 out_put_dentry:
 982         dput(path.dentry);
 983         goto out_nattch;
 984 }
 985
 986 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 987 {
 988         unsigned long ret;
 989         long err;
 990
 991         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 992         if (err)
 993                 return err;
 994         force_successful_syscall_return();
 995         return (long)ret;
 996 }
 997
 998 /*
 999  * detach and kill segment if marked destroyed.
1000  * The work is done in shm_close.
1001  */
1002 asmlinkage long sys_shmdt(char __user *shmaddr)
1003 {
1004         struct mm_struct *mm = current->mm;
1005         struct vm_area_struct *vma, *next;
1006         unsigned long addr = (unsigned long)shmaddr;
1007         loff_t size = 0;
1008         int retval = -EINVAL;
1009
1010         if (addr & ~PAGE_MASK)
1011                 return retval;
1012
1013         down_write(&mm->mmap_sem);
1014
1015         /*
1016          * This function tries to be smart and unmap shm segments that
1017          * were modified by partial mlock or munmap calls:
1018          * - It first determines the size of the shm segment that should be
1019          *   unmapped: It searches for a vma that is backed by shm and that
1020          *   started at address shmaddr. It records it's size and then unmaps
1021          *   it.
1022          * - Then it unmaps all shm vmas that started at shmaddr and that
1023          *   are within the initially determined size.
1024          * Errors from do_munmap are ignored: the function only fails if
1025          * it's called with invalid parameters or if it's called to unmap
1026          * a part of a vma. Both calls in this function are for full vmas,
1027          * the parameters are directly copied from the vma itself and always
1028          * valid - therefore do_munmap cannot fail. (famous last words?)
1029          */
1030         /*
1031          * If it had been mremap()'d, the starting address would not
1032          * match the usual checks anyway. So assume all vma's are
1033          * above the starting address given.
1034          */
1035         vma = find_vma(mm, addr);
1036
1037         while (vma) {
1038                 next = vma->vm_next;
1039
1040                 /*
1041                  * Check if the starting address would match, i.e. it's
1042                  * a fragment created by mprotect() and/or munmap(), or it
1043                  * otherwise it starts at this address with no hassles.
1044                  */
1045                 if ((vma->vm_ops == &shm_vm_ops) &&
1046                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1047
1048
1049                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1050                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051                         /*
1052                          * We discovered the size of the shm segment, so
1053                          * break out of here and fall through to the next
1054                          * loop that uses the size information to stop
1055                          * searching for matching vma's.
1056                          */
1057                         retval = 0;
1058                         vma = next;
1059                         break;
1060                 }
1061                 vma = next;
1062         }
1063
1064         /*
1065          * We need look no further than the maximum address a fragment
1066          * could possibly have landed at. Also cast things to loff_t to
1067          * prevent overflows and make comparisions vs. equal-width types.
1068          */
1069         size = PAGE_ALIGN(size);
1070         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1071                 next = vma->vm_next;
1072
1073                 /* finding a matching vma now does not alter retval */
1074                 if ((vma->vm_ops == &shm_vm_ops) &&
1075                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1076
1077                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1078                 vma = next;
1079         }
1080
1081         up_write(&mm->mmap_sem);
1082         return retval;
1083 }
1084
1085 #ifdef CONFIG_PROC_FS
1086 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1087 {
1088         struct shmid_kernel *shp = it;
1089         char *format;
1090
1091 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1092 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1093
1094         if (sizeof(size_t) <= sizeof(int))
1095                 format = SMALL_STRING;
1096         else
1097                 format = BIG_STRING;
1098         return seq_printf(s, format,
1099                           shp->shm_perm.key,
1100                           shp->shm_perm.id,
1101                           shp->shm_perm.mode,
1102                           shp->shm_segsz,
1103                           shp->shm_cprid,
1104                           shp->shm_lprid,
1105                           shp->shm_nattch,
1106                           shp->shm_perm.uid,
1107                           shp->shm_perm.gid,
1108                           shp->shm_perm.cuid,
1109                           shp->shm_perm.cgid,
1110                           shp->shm_atim,
1111                           shp->shm_dtim,
1112                           shp->shm_ctim);
1113 }
1114 #endif