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
  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_unlock(shp)                 \
  63         ipc_unlock(&(shp)->shm_perm)
  64 #define shm_buildid(id, seq)    ipc_buildid(id, seq)
  65
  66 static int newseg(struct ipc_namespace *, struct ipc_params *);
  67 static void shm_open(struct vm_area_struct *vma);
  68 static void shm_close(struct vm_area_struct *vma);
  69 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  70 #ifdef CONFIG_PROC_FS
  71 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  72 #endif
  73
  74 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
  75 {
  76         ns->ids[IPC_SHM_IDS] = ids;
  77         ns->shm_ctlmax = SHMMAX;
  78         ns->shm_ctlall = SHMALL;
  79         ns->shm_ctlmni = SHMMNI;
  80         ns->shm_tot = 0;
  81         ipc_init_ids(ids);
  82 }
  83
  84 /*
  85  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  86  * Only shm_ids.rw_mutex remains locked on exit.
  87  */
  88 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
  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 int shm_init_ns(struct ipc_namespace *ns)
 100 {
 101         struct ipc_ids *ids;
 102
 103         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
 104         if (ids == NULL)
 105                 return -ENOMEM;
 106
 107         __shm_init_ns(ns, ids);
 108         return 0;
 109 }
 110
 111 void shm_exit_ns(struct ipc_namespace *ns)
 112 {
 113         struct shmid_kernel *shp;
 114         struct kern_ipc_perm *perm;
 115         int next_id;
 116         int total, in_use;
 117
 118         down_write(&shm_ids(ns).rw_mutex);
 119
 120         in_use = shm_ids(ns).in_use;
 121
 122         for (total = 0, next_id = 0; total < in_use; next_id++) {
 123                 perm = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 124                 if (perm == NULL)
 125                         continue;
 126                 ipc_lock_by_ptr(perm);
 127                 shp = container_of(perm, struct shmid_kernel, shm_perm);
 128                 do_shm_rmid(ns, shp);
 129                 total++;
 130         }
 131         up_write(&shm_ids(ns).rw_mutex);
 132
 133         kfree(ns->ids[IPC_SHM_IDS]);
 134         ns->ids[IPC_SHM_IDS] = NULL;
 135 }
 136
 137 void __init shm_init (void)
 138 {
 139         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
 140         ipc_init_proc_interface("sysvipc/shm",
 141                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 142                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 143 }
 144
 145 /*
 146  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
 147  * is held to protect access to the idr tree.
 148  */
 149 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
 150                                                 int id)
 151 {
 152         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
 153
 154         if (IS_ERR(ipcp))
 155                 return (struct shmid_kernel *)ipcp;
 156
 157         return container_of(ipcp, struct shmid_kernel, shm_perm);
 158 }
 159
 160 static inline struct shmid_kernel *shm_lock_check_down(
 161                                                 struct ipc_namespace *ns,
 162                                                 int id)
 163 {
 164         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
 165
 166         if (IS_ERR(ipcp))
 167                 return (struct shmid_kernel *)ipcp;
 168
 169         return container_of(ipcp, struct shmid_kernel, shm_perm);
 170 }
 171
 172 /*
 173  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 174  * is not held.
 175  */
 176 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 177 {
 178         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 179
 180         if (IS_ERR(ipcp))
 181                 return (struct shmid_kernel *)ipcp;
 182
 183         return container_of(ipcp, struct shmid_kernel, shm_perm);
 184 }
 185
 186 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 187                                                 int id)
 188 {
 189         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 190
 191         if (IS_ERR(ipcp))
 192                 return (struct shmid_kernel *)ipcp;
 193
 194         return container_of(ipcp, struct shmid_kernel, shm_perm);
 195 }
 196
 197 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 198 {
 199         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 200 }
 201
 202 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 203 {
 204         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 205 }
 206
 207
 208
 209 /* This is called by fork, once for every shm attach. */
 210 static void shm_open(struct vm_area_struct *vma)
 211 {
 212         struct file *file = vma->vm_file;
 213         struct shm_file_data *sfd = shm_file_data(file);
 214         struct shmid_kernel *shp;
 215
 216         shp = shm_lock(sfd->ns, sfd->id);
 217         BUG_ON(IS_ERR(shp));
 218         shp->shm_atim = get_seconds();
 219         shp->shm_lprid = task_tgid_vnr(current);
 220         shp->shm_nattch++;
 221         shm_unlock(shp);
 222 }
 223
 224 /*
 225  * shm_destroy - free the struct shmid_kernel
 226  *
 227  * @ns: namespace
 228  * @shp: struct to free
 229  *
 230  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 231  * but returns with shp unlocked and freed.
 232  */
 233 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 234 {
 235         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 236         shm_rmid(ns, shp);
 237         shm_unlock(shp);
 238         if (!is_file_hugepages(shp->shm_file))
 239                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 240         else
 241                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 242                                                 shp->mlock_user);
 243         fput (shp->shm_file);
 244         security_shm_free(shp);
 245         ipc_rcu_putref(shp);
 246 }
 247
 248 /*
 249  * remove the attach descriptor vma.
 250  * free memory for segment if it is marked destroyed.
 251  * The descriptor has already been removed from the current->mm->mmap list
 252  * and will later be kfree()d.
 253  */
 254 static void shm_close(struct vm_area_struct *vma)
 255 {
 256         struct file * file = vma->vm_file;
 257         struct shm_file_data *sfd = shm_file_data(file);
 258         struct shmid_kernel *shp;
 259         struct ipc_namespace *ns = sfd->ns;
 260
 261         down_write(&shm_ids(ns).rw_mutex);
 262         /* remove from the list of attaches of the shm segment */
 263         shp = shm_lock_down(ns, sfd->id);
 264         BUG_ON(IS_ERR(shp));
 265         shp->shm_lprid = task_tgid_vnr(current);
 266         shp->shm_dtim = get_seconds();
 267         shp->shm_nattch--;
 268         if(shp->shm_nattch == 0 &&
 269            shp->shm_perm.mode & SHM_DEST)
 270                 shm_destroy(ns, shp);
 271         else
 272                 shm_unlock(shp);
 273         up_write(&shm_ids(ns).rw_mutex);
 274 }
 275
 276 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 277 {
 278         struct file *file = vma->vm_file;
 279         struct shm_file_data *sfd = shm_file_data(file);
 280
 281         return sfd->vm_ops->fault(vma, vmf);
 282 }
 283
 284 #ifdef CONFIG_NUMA
 285 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 286 {
 287         struct file *file = vma->vm_file;
 288         struct shm_file_data *sfd = shm_file_data(file);
 289         int err = 0;
 290         if (sfd->vm_ops->set_policy)
 291                 err = sfd->vm_ops->set_policy(vma, new);
 292         return err;
 293 }
 294
 295 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 296                                         unsigned long addr)
 297 {
 298         struct file *file = vma->vm_file;
 299         struct shm_file_data *sfd = shm_file_data(file);
 300         struct mempolicy *pol = NULL;
 301
 302         if (sfd->vm_ops->get_policy)
 303                 pol = sfd->vm_ops->get_policy(vma, addr);
 304         else if (vma->vm_policy)
 305                 pol = vma->vm_policy;
 306         else
 307                 pol = current->mempolicy;
 308         return pol;
 309 }
 310 #endif
 311
 312 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 313 {
 314         struct shm_file_data *sfd = shm_file_data(file);
 315         int ret;
 316
 317         ret = sfd->file->f_op->mmap(sfd->file, vma);
 318         if (ret != 0)
 319                 return ret;
 320         sfd->vm_ops = vma->vm_ops;
 321 #ifdef CONFIG_MMU
 322         BUG_ON(!sfd->vm_ops->fault);
 323 #endif
 324         vma->vm_ops = &shm_vm_ops;
 325         shm_open(vma);
 326
 327         return ret;
 328 }
 329
 330 static int shm_release(struct inode *ino, struct file *file)
 331 {
 332         struct shm_file_data *sfd = shm_file_data(file);
 333
 334         put_ipc_ns(sfd->ns);
 335         shm_file_data(file) = NULL;
 336         kfree(sfd);
 337         return 0;
 338 }
 339
 340 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 341 {
 342         int (*fsync) (struct file *, struct dentry *, int datasync);
 343         struct shm_file_data *sfd = shm_file_data(file);
 344         int ret = -EINVAL;
 345
 346         fsync = sfd->file->f_op->fsync;
 347         if (fsync)
 348                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 349         return ret;
 350 }
 351
 352 static unsigned long shm_get_unmapped_area(struct file *file,
 353         unsigned long addr, unsigned long len, unsigned long pgoff,
 354         unsigned long flags)
 355 {
 356         struct shm_file_data *sfd = shm_file_data(file);
 357         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 358 }
 359
 360 int is_file_shm_hugepages(struct file *file)
 361 {
 362         int ret = 0;
 363
 364         if (file->f_op == &shm_file_operations) {
 365                 struct shm_file_data *sfd;
 366                 sfd = shm_file_data(file);
 367                 ret = is_file_hugepages(sfd->file);
 368         }
 369         return ret;
 370 }
 371
 372 static const struct file_operations shm_file_operations = {
 373         .mmap           = shm_mmap,
 374         .fsync          = shm_fsync,
 375         .release        = shm_release,
 376         .get_unmapped_area      = shm_get_unmapped_area,
 377 };
 378
 379 static struct vm_operations_struct shm_vm_ops = {
 380         .open   = shm_open,     /* callback for a new vm-area open */
 381         .close  = shm_close,    /* callback for when the vm-area is released */
 382         .fault  = shm_fault,
 383 #if defined(CONFIG_NUMA)
 384         .set_policy = shm_set_policy,
 385         .get_policy = shm_get_policy,
 386 #endif
 387 };
 388
 389 /**
 390  * newseg - Create a new shared memory segment
 391  * @ns: namespace
 392  * @params: ptr to the structure that contains key, size and shmflg
 393  *
 394  * Called with shm_ids.rw_mutex held as a writer.
 395  */
 396
 397 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 398 {
 399         key_t key = params->key;
 400         int shmflg = params->flg;
 401         size_t size = params->u.size;
 402         int error;
 403         struct shmid_kernel *shp;
 404         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 405         struct file * file;
 406         char name[13];
 407         int id;
 408
 409         if (size < SHMMIN || size > ns->shm_ctlmax)
 410                 return -EINVAL;
 411
 412         if (ns->shm_tot + numpages > ns->shm_ctlall)
 413                 return -ENOSPC;
 414
 415         shp = ipc_rcu_alloc(sizeof(*shp));
 416         if (!shp)
 417                 return -ENOMEM;
 418
 419         shp->shm_perm.key = key;
 420         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 421         shp->mlock_user = NULL;
 422
 423         shp->shm_perm.security = NULL;
 424         error = security_shm_alloc(shp);
 425         if (error) {
 426                 ipc_rcu_putref(shp);
 427                 return error;
 428         }
 429
 430         sprintf (name, "SYSV%08x", key);
 431         if (shmflg & SHM_HUGETLB) {
 432                 /* hugetlb_file_setup takes care of mlock user accounting */
 433                 file = hugetlb_file_setup(name, size);
 434                 shp->mlock_user = current->user;
 435         } else {
 436                 int acctflag = VM_ACCOUNT;
 437                 /*
 438                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 439                  * if it's asked for.
 440                  */
 441                 if  ((shmflg & SHM_NORESERVE) &&
 442                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 443                         acctflag = 0;
 444                 file = shmem_file_setup(name, size, acctflag);
 445         }
 446         error = PTR_ERR(file);
 447         if (IS_ERR(file))
 448                 goto no_file;
 449
 450         id = shm_addid(ns, shp);
 451         if (id < 0) {
 452                 error = id;
 453                 goto no_id;
 454         }
 455
 456         shp->shm_cprid = task_tgid_vnr(current);
 457         shp->shm_lprid = 0;
 458         shp->shm_atim = shp->shm_dtim = 0;
 459         shp->shm_ctim = get_seconds();
 460         shp->shm_segsz = size;
 461         shp->shm_nattch = 0;
 462         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
 463         shp->shm_file = file;
 464         /*
 465          * shmid gets reported as "inode#" in /proc/pid/maps.
 466          * proc-ps tools use this. Changing this will break them.
 467          */
 468         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 469
 470         ns->shm_tot += numpages;
 471         error = shp->shm_perm.id;
 472         shm_unlock(shp);
 473         return error;
 474
 475 no_id:
 476         fput(file);
 477 no_file:
 478         security_shm_free(shp);
 479         ipc_rcu_putref(shp);
 480         return error;
 481 }
 482
 483 /*
 484  * Called with shm_ids.rw_mutex and ipcp locked.
 485  */
 486 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 487 {
 488         struct shmid_kernel *shp;
 489
 490         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 491         return security_shm_associate(shp, shmflg);
 492 }
 493
 494 /*
 495  * Called with shm_ids.rw_mutex and ipcp locked.
 496  */
 497 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 498                                 struct ipc_params *params)
 499 {
 500         struct shmid_kernel *shp;
 501
 502         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 503         if (shp->shm_segsz < params->u.size)
 504                 return -EINVAL;
 505
 506         return 0;
 507 }
 508
 509 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 510 {
 511         struct ipc_namespace *ns;
 512         struct ipc_ops shm_ops;
 513         struct ipc_params shm_params;
 514
 515         ns = current->nsproxy->ipc_ns;
 516
 517         shm_ops.getnew = newseg;
 518         shm_ops.associate = shm_security;
 519         shm_ops.more_checks = shm_more_checks;
 520
 521         shm_params.key = key;
 522         shm_params.flg = shmflg;
 523         shm_params.u.size = size;
 524
 525         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 526 }
 527
 528 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 529 {
 530         switch(version) {
 531         case IPC_64:
 532                 return copy_to_user(buf, in, sizeof(*in));
 533         case IPC_OLD:
 534             {
 535                 struct shmid_ds out;
 536
 537                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 538                 out.shm_segsz   = in->shm_segsz;
 539                 out.shm_atime   = in->shm_atime;
 540                 out.shm_dtime   = in->shm_dtime;
 541                 out.shm_ctime   = in->shm_ctime;
 542                 out.shm_cpid    = in->shm_cpid;
 543                 out.shm_lpid    = in->shm_lpid;
 544                 out.shm_nattch  = in->shm_nattch;
 545
 546                 return copy_to_user(buf, &out, sizeof(out));
 547             }
 548         default:
 549                 return -EINVAL;
 550         }
 551 }
 552
 553 struct shm_setbuf {
 554         uid_t   uid;
 555         gid_t   gid;
 556         mode_t  mode;
 557 };
 558
 559 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 560 {
 561         switch(version) {
 562         case IPC_64:
 563             {
 564                 struct shmid64_ds tbuf;
 565
 566                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 567                         return -EFAULT;
 568
 569                 out->uid        = tbuf.shm_perm.uid;
 570                 out->gid        = tbuf.shm_perm.gid;
 571                 out->mode       = tbuf.shm_perm.mode;
 572
 573                 return 0;
 574             }
 575         case IPC_OLD:
 576             {
 577                 struct shmid_ds tbuf_old;
 578
 579                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 580                         return -EFAULT;
 581
 582                 out->uid        = tbuf_old.shm_perm.uid;
 583                 out->gid        = tbuf_old.shm_perm.gid;
 584                 out->mode       = tbuf_old.shm_perm.mode;
 585
 586                 return 0;
 587             }
 588         default:
 589                 return -EINVAL;
 590         }
 591 }
 592
 593 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 594 {
 595         switch(version) {
 596         case IPC_64:
 597                 return copy_to_user(buf, in, sizeof(*in));
 598         case IPC_OLD:
 599             {
 600                 struct shminfo out;
 601
 602                 if(in->shmmax > INT_MAX)
 603                         out.shmmax = INT_MAX;
 604                 else
 605                         out.shmmax = (int)in->shmmax;
 606
 607                 out.shmmin      = in->shmmin;
 608                 out.shmmni      = in->shmmni;
 609                 out.shmseg      = in->shmseg;
 610                 out.shmall      = in->shmall;
 611
 612                 return copy_to_user(buf, &out, sizeof(out));
 613             }
 614         default:
 615                 return -EINVAL;
 616         }
 617 }
 618
 619 /*
 620  * Called with shm_ids.rw_mutex held as a reader
 621  */
 622 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 623                 unsigned long *swp)
 624 {
 625         int next_id;
 626         int total, in_use;
 627
 628         *rss = 0;
 629         *swp = 0;
 630
 631         in_use = shm_ids(ns).in_use;
 632
 633         for (total = 0, next_id = 0; total < in_use; next_id++) {
 634                 struct shmid_kernel *shp;
 635                 struct inode *inode;
 636
 637                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 638                 if (shp == NULL)
 639                         continue;
 640
 641                 inode = shp->shm_file->f_path.dentry->d_inode;
 642
 643                 if (is_file_hugepages(shp->shm_file)) {
 644                         struct address_space *mapping = inode->i_mapping;
 645                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 646                 } else {
 647                         struct shmem_inode_info *info = SHMEM_I(inode);
 648                         spin_lock(&info->lock);
 649                         *rss += inode->i_mapping->nrpages;
 650                         *swp += info->swapped;
 651                         spin_unlock(&info->lock);
 652                 }
 653
 654                 total++;
 655         }
 656 }
 657
 658 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 659 {
 660         struct shm_setbuf setbuf;
 661         struct shmid_kernel *shp;
 662         int err, version;
 663         struct ipc_namespace *ns;
 664
 665         if (cmd < 0 || shmid < 0) {
 666                 err = -EINVAL;
 667                 goto out;
 668         }
 669
 670         version = ipc_parse_version(&cmd);
 671         ns = current->nsproxy->ipc_ns;
 672
 673         switch (cmd) { /* replace with proc interface ? */
 674         case IPC_INFO:
 675         {
 676                 struct shminfo64 shminfo;
 677
 678                 err = security_shm_shmctl(NULL, cmd);
 679                 if (err)
 680                         return err;
 681
 682                 memset(&shminfo,0,sizeof(shminfo));
 683                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 684                 shminfo.shmmax = ns->shm_ctlmax;
 685                 shminfo.shmall = ns->shm_ctlall;
 686
 687                 shminfo.shmmin = SHMMIN;
 688                 if(copy_shminfo_to_user (buf, &shminfo, version))
 689                         return -EFAULT;
 690
 691                 down_read(&shm_ids(ns).rw_mutex);
 692                 err = ipc_get_maxid(&shm_ids(ns));
 693                 up_read(&shm_ids(ns).rw_mutex);
 694
 695                 if(err<0)
 696                         err = 0;
 697                 goto out;
 698         }
 699         case SHM_INFO:
 700         {
 701                 struct shm_info shm_info;
 702
 703                 err = security_shm_shmctl(NULL, cmd);
 704                 if (err)
 705                         return err;
 706
 707                 memset(&shm_info,0,sizeof(shm_info));
 708                 down_read(&shm_ids(ns).rw_mutex);
 709                 shm_info.used_ids = shm_ids(ns).in_use;
 710                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 711                 shm_info.shm_tot = ns->shm_tot;
 712                 shm_info.swap_attempts = 0;
 713                 shm_info.swap_successes = 0;
 714                 err = ipc_get_maxid(&shm_ids(ns));
 715                 up_read(&shm_ids(ns).rw_mutex);
 716                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 717                         err = -EFAULT;
 718                         goto out;
 719                 }
 720
 721                 err = err < 0 ? 0 : err;
 722                 goto out;
 723         }
 724         case SHM_STAT:
 725         case IPC_STAT:
 726         {
 727                 struct shmid64_ds tbuf;
 728                 int result;
 729
 730                 if (!buf) {
 731                         err = -EFAULT;
 732                         goto out;
 733                 }
 734
 735                 if (cmd == SHM_STAT) {
 736                         shp = shm_lock(ns, shmid);
 737                         if (IS_ERR(shp)) {
 738                                 err = PTR_ERR(shp);
 739                                 goto out;
 740                         }
 741                         result = shp->shm_perm.id;
 742                 } else {
 743                         shp = shm_lock_check(ns, shmid);
 744                         if (IS_ERR(shp)) {
 745                                 err = PTR_ERR(shp);
 746                                 goto out;
 747                         }
 748                         result = 0;
 749                 }
 750                 err=-EACCES;
 751                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 752                         goto out_unlock;
 753                 err = security_shm_shmctl(shp, cmd);
 754                 if (err)
 755                         goto out_unlock;
 756                 memset(&tbuf, 0, sizeof(tbuf));
 757                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 758                 tbuf.shm_segsz  = shp->shm_segsz;
 759                 tbuf.shm_atime  = shp->shm_atim;
 760                 tbuf.shm_dtime  = shp->shm_dtim;
 761                 tbuf.shm_ctime  = shp->shm_ctim;
 762                 tbuf.shm_cpid   = shp->shm_cprid;
 763                 tbuf.shm_lpid   = shp->shm_lprid;
 764                 tbuf.shm_nattch = shp->shm_nattch;
 765                 shm_unlock(shp);
 766                 if(copy_shmid_to_user (buf, &tbuf, version))
 767                         err = -EFAULT;
 768                 else
 769                         err = result;
 770                 goto out;
 771         }
 772         case SHM_LOCK:
 773         case SHM_UNLOCK:
 774         {
 775                 shp = shm_lock_check(ns, shmid);
 776                 if (IS_ERR(shp)) {
 777                         err = PTR_ERR(shp);
 778                         goto out;
 779                 }
 780
 781                 err = audit_ipc_obj(&(shp->shm_perm));
 782                 if (err)
 783                         goto out_unlock;
 784
 785                 if (!capable(CAP_IPC_LOCK)) {
 786                         err = -EPERM;
 787                         if (current->euid != shp->shm_perm.uid &&
 788                             current->euid != shp->shm_perm.cuid)
 789                                 goto out_unlock;
 790                         if (cmd == SHM_LOCK &&
 791                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 792                                 goto out_unlock;
 793                 }
 794
 795                 err = security_shm_shmctl(shp, cmd);
 796                 if (err)
 797                         goto out_unlock;
 798
 799                 if(cmd==SHM_LOCK) {
 800                         struct user_struct * user = current->user;
 801                         if (!is_file_hugepages(shp->shm_file)) {
 802                                 err = shmem_lock(shp->shm_file, 1, user);
 803                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 804                                         shp->shm_perm.mode |= SHM_LOCKED;
 805                                         shp->mlock_user = user;
 806                                 }
 807                         }
 808                 } else if (!is_file_hugepages(shp->shm_file)) {
 809                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 810                         shp->shm_perm.mode &= ~SHM_LOCKED;
 811                         shp->mlock_user = NULL;
 812                 }
 813                 shm_unlock(shp);
 814                 goto out;
 815         }
 816         case IPC_RMID:
 817         {
 818                 /*
 819                  *      We cannot simply remove the file. The SVID states
 820                  *      that the block remains until the last person
 821                  *      detaches from it, then is deleted. A shmat() on
 822                  *      an RMID segment is legal in older Linux and if
 823                  *      we change it apps break...
 824                  *
 825                  *      Instead we set a destroyed flag, and then blow
 826                  *      the name away when the usage hits zero.
 827                  */
 828                 down_write(&shm_ids(ns).rw_mutex);
 829                 shp = shm_lock_check_down(ns, shmid);
 830                 if (IS_ERR(shp)) {
 831                         err = PTR_ERR(shp);
 832                         goto out_up;
 833                 }
 834
 835                 err = audit_ipc_obj(&(shp->shm_perm));
 836                 if (err)
 837                         goto out_unlock_up;
 838
 839                 if (current->euid != shp->shm_perm.uid &&
 840                     current->euid != shp->shm_perm.cuid &&
 841                     !capable(CAP_SYS_ADMIN)) {
 842                         err=-EPERM;
 843                         goto out_unlock_up;
 844                 }
 845
 846                 err = security_shm_shmctl(shp, cmd);
 847                 if (err)
 848                         goto out_unlock_up;
 849
 850                 do_shm_rmid(ns, shp);
 851                 up_write(&shm_ids(ns).rw_mutex);
 852                 goto out;
 853         }
 854
 855         case IPC_SET:
 856         {
 857                 if (!buf) {
 858                         err = -EFAULT;
 859                         goto out;
 860                 }
 861
 862                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 863                         err = -EFAULT;
 864                         goto out;
 865                 }
 866                 down_write(&shm_ids(ns).rw_mutex);
 867                 shp = shm_lock_check_down(ns, shmid);
 868                 if (IS_ERR(shp)) {
 869                         err = PTR_ERR(shp);
 870                         goto out_up;
 871                 }
 872                 err = audit_ipc_obj(&(shp->shm_perm));
 873                 if (err)
 874                         goto out_unlock_up;
 875                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
 876                 if (err)
 877                         goto out_unlock_up;
 878                 err=-EPERM;
 879                 if (current->euid != shp->shm_perm.uid &&
 880                     current->euid != shp->shm_perm.cuid &&
 881                     !capable(CAP_SYS_ADMIN)) {
 882                         goto out_unlock_up;
 883                 }
 884
 885                 err = security_shm_shmctl(shp, cmd);
 886                 if (err)
 887                         goto out_unlock_up;
 888
 889                 shp->shm_perm.uid = setbuf.uid;
 890                 shp->shm_perm.gid = setbuf.gid;
 891                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 892                         | (setbuf.mode & S_IRWXUGO);
 893                 shp->shm_ctim = get_seconds();
 894                 break;
 895         }
 896
 897         default:
 898                 err = -EINVAL;
 899                 goto out;
 900         }
 901
 902         err = 0;
 903 out_unlock_up:
 904         shm_unlock(shp);
 905 out_up:
 906         up_write(&shm_ids(ns).rw_mutex);
 907         goto out;
 908 out_unlock:
 909         shm_unlock(shp);
 910 out:
 911         return err;
 912 }
 913
 914 /*
 915  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 916  *
 917  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 918  * "raddr" thing points to kernel space, and there has to be a wrapper around
 919  * this.
 920  */
 921 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 922 {
 923         struct shmid_kernel *shp;
 924         unsigned long addr;
 925         unsigned long size;
 926         struct file * file;
 927         int    err;
 928         unsigned long flags;
 929         unsigned long prot;
 930         int acc_mode;
 931         unsigned long user_addr;
 932         struct ipc_namespace *ns;
 933         struct shm_file_data *sfd;
 934         struct path path;
 935         mode_t f_mode;
 936
 937         err = -EINVAL;
 938         if (shmid < 0)
 939                 goto out;
 940         else if ((addr = (ulong)shmaddr)) {
 941                 if (addr & (SHMLBA-1)) {
 942                         if (shmflg & SHM_RND)
 943                                 addr &= ~(SHMLBA-1);       /* round down */
 944                         else
 945 #ifndef __ARCH_FORCE_SHMLBA
 946                                 if (addr & ~PAGE_MASK)
 947 #endif
 948                                         goto out;
 949                 }
 950                 flags = MAP_SHARED | MAP_FIXED;
 951         } else {
 952                 if ((shmflg & SHM_REMAP))
 953                         goto out;
 954
 955                 flags = MAP_SHARED;
 956         }
 957
 958         if (shmflg & SHM_RDONLY) {
 959                 prot = PROT_READ;
 960                 acc_mode = S_IRUGO;
 961                 f_mode = FMODE_READ;
 962         } else {
 963                 prot = PROT_READ | PROT_WRITE;
 964                 acc_mode = S_IRUGO | S_IWUGO;
 965                 f_mode = FMODE_READ | FMODE_WRITE;
 966         }
 967         if (shmflg & SHM_EXEC) {
 968                 prot |= PROT_EXEC;
 969                 acc_mode |= S_IXUGO;
 970         }
 971
 972         /*
 973          * We cannot rely on the fs check since SYSV IPC does have an
 974          * additional creator id...
 975          */
 976         ns = current->nsproxy->ipc_ns;
 977         shp = shm_lock_check(ns, shmid);
 978         if (IS_ERR(shp)) {
 979                 err = PTR_ERR(shp);
 980                 goto out;
 981         }
 982
 983         err = -EACCES;
 984         if (ipcperms(&shp->shm_perm, acc_mode))
 985                 goto out_unlock;
 986
 987         err = security_shm_shmat(shp, shmaddr, shmflg);
 988         if (err)
 989                 goto out_unlock;
 990
 991         path.dentry = dget(shp->shm_file->f_path.dentry);
 992         path.mnt    = shp->shm_file->f_path.mnt;
 993         shp->shm_nattch++;
 994         size = i_size_read(path.dentry->d_inode);
 995         shm_unlock(shp);
 996
 997         err = -ENOMEM;
 998         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 999         if (!sfd)
1000                 goto out_put_dentry;
1001
1002         err = -ENOMEM;
1003
1004         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
1005         if (!file)
1006                 goto out_free;
1007
1008         file->private_data = sfd;
1009         file->f_mapping = shp->shm_file->f_mapping;
1010         sfd->id = shp->shm_perm.id;
1011         sfd->ns = get_ipc_ns(ns);
1012         sfd->file = shp->shm_file;
1013         sfd->vm_ops = NULL;
1014
1015         down_write(&current->mm->mmap_sem);
1016         if (addr && !(shmflg & SHM_REMAP)) {
1017                 err = -EINVAL;
1018                 if (find_vma_intersection(current->mm, addr, addr + size))
1019                         goto invalid;
1020                 /*
1021                  * If shm segment goes below stack, make sure there is some
1022                  * space left for the stack to grow (at least 4 pages).
1023                  */
1024                 if (addr < current->mm->start_stack &&
1025                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1026                         goto invalid;
1027         }
1028
1029         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1030         *raddr = user_addr;
1031         err = 0;
1032         if (IS_ERR_VALUE(user_addr))
1033                 err = (long)user_addr;
1034 invalid:
1035         up_write(&current->mm->mmap_sem);
1036
1037         fput(file);
1038
1039 out_nattch:
1040         down_write(&shm_ids(ns).rw_mutex);
1041         shp = shm_lock_down(ns, shmid);
1042         BUG_ON(IS_ERR(shp));
1043         shp->shm_nattch--;
1044         if(shp->shm_nattch == 0 &&
1045            shp->shm_perm.mode & SHM_DEST)
1046                 shm_destroy(ns, shp);
1047         else
1048                 shm_unlock(shp);
1049         up_write(&shm_ids(ns).rw_mutex);
1050
1051 out:
1052         return err;
1053
1054 out_unlock:
1055         shm_unlock(shp);
1056         goto out;
1057
1058 out_free:
1059         kfree(sfd);
1060 out_put_dentry:
1061         dput(path.dentry);
1062         goto out_nattch;
1063 }
1064
1065 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1066 {
1067         unsigned long ret;
1068         long err;
1069
1070         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1071         if (err)
1072                 return err;
1073         force_successful_syscall_return();
1074         return (long)ret;
1075 }
1076
1077 /*
1078  * detach and kill segment if marked destroyed.
1079  * The work is done in shm_close.
1080  */
1081 asmlinkage long sys_shmdt(char __user *shmaddr)
1082 {
1083         struct mm_struct *mm = current->mm;
1084         struct vm_area_struct *vma, *next;
1085         unsigned long addr = (unsigned long)shmaddr;
1086         loff_t size = 0;
1087         int retval = -EINVAL;
1088
1089         if (addr & ~PAGE_MASK)
1090                 return retval;
1091
1092         down_write(&mm->mmap_sem);
1093
1094         /*
1095          * This function tries to be smart and unmap shm segments that
1096          * were modified by partial mlock or munmap calls:
1097          * - It first determines the size of the shm segment that should be
1098          *   unmapped: It searches for a vma that is backed by shm and that
1099          *   started at address shmaddr. It records it's size and then unmaps
1100          *   it.
1101          * - Then it unmaps all shm vmas that started at shmaddr and that
1102          *   are within the initially determined size.
1103          * Errors from do_munmap are ignored: the function only fails if
1104          * it's called with invalid parameters or if it's called to unmap
1105          * a part of a vma. Both calls in this function are for full vmas,
1106          * the parameters are directly copied from the vma itself and always
1107          * valid - therefore do_munmap cannot fail. (famous last words?)
1108          */
1109         /*
1110          * If it had been mremap()'d, the starting address would not
1111          * match the usual checks anyway. So assume all vma's are
1112          * above the starting address given.
1113          */
1114         vma = find_vma(mm, addr);
1115
1116         while (vma) {
1117                 next = vma->vm_next;
1118
1119                 /*
1120                  * Check if the starting address would match, i.e. it's
1121                  * a fragment created by mprotect() and/or munmap(), or it
1122                  * otherwise it starts at this address with no hassles.
1123                  */
1124                 if ((vma->vm_ops == &shm_vm_ops) &&
1125                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1126
1127
1128                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1129                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1130                         /*
1131                          * We discovered the size of the shm segment, so
1132                          * break out of here and fall through to the next
1133                          * loop that uses the size information to stop
1134                          * searching for matching vma's.
1135                          */
1136                         retval = 0;
1137                         vma = next;
1138                         break;
1139                 }
1140                 vma = next;
1141         }
1142
1143         /*
1144          * We need look no further than the maximum address a fragment
1145          * could possibly have landed at. Also cast things to loff_t to
1146          * prevent overflows and make comparisions vs. equal-width types.
1147          */
1148         size = PAGE_ALIGN(size);
1149         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1150                 next = vma->vm_next;
1151
1152                 /* finding a matching vma now does not alter retval */
1153                 if ((vma->vm_ops == &shm_vm_ops) &&
1154                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1155
1156                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1157                 vma = next;
1158         }
1159
1160         up_write(&mm->mmap_sem);
1161         return retval;
1162 }
1163
1164 #ifdef CONFIG_PROC_FS
1165 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1166 {
1167         struct shmid_kernel *shp = it;
1168         char *format;
1169
1170 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1171 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1172
1173         if (sizeof(size_t) <= sizeof(int))
1174                 format = SMALL_STRING;
1175         else
1176                 format = BIG_STRING;
1177         return seq_printf(s, format,
1178                           shp->shm_perm.key,
1179                           shp->shm_perm.id,
1180                           shp->shm_perm.mode,
1181                           shp->shm_segsz,
1182                           shp->shm_cprid,
1183                           shp->shm_lprid,
1184                           shp->shm_nattch,
1185                           shp->shm_perm.uid,
1186                           shp->shm_perm.gid,
1187                           shp->shm_perm.cuid,
1188                           shp->shm_perm.cgid,
1189                           shp->shm_atim,
1190                           shp->shm_dtim,
1191                           shp->shm_ctim);
1192 }
1193 #endif