ipc/util.c

   1 /*
   2  * linux/ipc/util.c
   3  * Copyright (C) 1992 Krishna Balasubramanian
   4  *
   5  * Sep 1997 - Call suser() last after "normal" permission checks so we
   6  *            get BSD style process accounting right.
   7  *            Occurs in several places in the IPC code.
   8  *            Chris Evans, <chris@ferret.lmh.ox.ac.uk>
   9  * Nov 1999 - ipc helper functions, unified SMP locking
  10  *            Manfred Spraul <manfred@colorfullife.com>
  11  * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
  12  *            Mingming Cao <cmm@us.ibm.com>
  13  * Mar 2006 - support for audit of ipc object properties
  14  *            Dustin Kirkland <dustin.kirkland@us.ibm.com>
  15  * Jun 2006 - namespaces ssupport
  16  *            OpenVZ, SWsoft Inc.
  17  *            Pavel Emelianov <xemul@openvz.org>
  18  */
  19
  20 #include <linux/mm.h>
  21 #include <linux/shm.h>
  22 #include <linux/init.h>
  23 #include <linux/msg.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/capability.h>
  27 #include <linux/highuid.h>
  28 #include <linux/security.h>
  29 #include <linux/rcupdate.h>
  30 #include <linux/workqueue.h>
  31 #include <linux/seq_file.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/audit.h>
  34 #include <linux/nsproxy.h>
  35
  36 #include <asm/unistd.h>
  37
  38 #include "util.h"
  39
  40 struct ipc_proc_iface {
  41         const char *path;
  42         const char *header;
  43         int ids;
  44         int (*show)(struct seq_file *, void *);
  45 };
  46
  47 struct ipc_namespace init_ipc_ns = {
  48         .kref = {
  49                 .refcount       = ATOMIC_INIT(2),
  50         },
  51 };
  52
  53 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
  54 {
  55         int err;
  56         struct ipc_namespace *ns;
  57
  58         err = -ENOMEM;
  59         ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
  60         if (ns == NULL)
  61                 goto err_mem;
  62
  63         err = sem_init_ns(ns);
  64         if (err)
  65                 goto err_sem;
  66         err = msg_init_ns(ns);
  67         if (err)
  68                 goto err_msg;
  69         err = shm_init_ns(ns);
  70         if (err)
  71                 goto err_shm;
  72
  73         kref_init(&ns->kref);
  74         return ns;
  75
  76 err_shm:
  77         msg_exit_ns(ns);
  78 err_msg:
  79         sem_exit_ns(ns);
  80 err_sem:
  81         kfree(ns);
  82 err_mem:
  83         return ERR_PTR(err);
  84 }
  85
  86 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
  87 {
  88         struct ipc_namespace *new_ns;
  89
  90         BUG_ON(!ns);
  91         get_ipc_ns(ns);
  92
  93         if (!(flags & CLONE_NEWIPC))
  94                 return ns;
  95
  96         new_ns = clone_ipc_ns(ns);
  97
  98         put_ipc_ns(ns);
  99         return new_ns;
 100 }
 101
 102 void free_ipc_ns(struct kref *kref)
 103 {
 104         struct ipc_namespace *ns;
 105
 106         ns = container_of(kref, struct ipc_namespace, kref);
 107         sem_exit_ns(ns);
 108         msg_exit_ns(ns);
 109         shm_exit_ns(ns);
 110         kfree(ns);
 111 }
 112
 113 /**
 114  *      ipc_init        -       initialise IPC subsystem
 115  *
 116  *      The various system5 IPC resources (semaphores, messages and shared
 117  *      memory) are initialised
 118  */
 119
 120 static int __init ipc_init(void)
 121 {
 122         sem_init();
 123         msg_init();
 124         shm_init();
 125         return 0;
 126 }
 127 __initcall(ipc_init);
 128
 129 /**
 130  *      ipc_init_ids            -       initialise IPC identifiers
 131  *      @ids: Identifier set
 132  *      @size: Number of identifiers
 133  *
 134  *      Given a size for the ipc identifier range (limited below IPCMNI)
 135  *      set up the sequence range to use then allocate and initialise the
 136  *      array itself.
 137  */
 138
 139 void ipc_init_ids(struct ipc_ids* ids, int size)
 140 {
 141         int i;
 142
 143         mutex_init(&ids->mutex);
 144
 145         if(size > IPCMNI)
 146                 size = IPCMNI;
 147         ids->in_use = 0;
 148         ids->max_id = -1;
 149         ids->seq = 0;
 150         {
 151                 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
 152                 if(seq_limit > USHRT_MAX)
 153                         ids->seq_max = USHRT_MAX;
 154                  else
 155                         ids->seq_max = seq_limit;
 156         }
 157
 158         ids->entries = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*size +
 159                                      sizeof(struct ipc_id_ary));
 160
 161         if(ids->entries == NULL) {
 162                 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
 163                 size = 0;
 164                 ids->entries = &ids->nullentry;
 165         }
 166         ids->entries->size = size;
 167         for(i=0;i<size;i++)
 168                 ids->entries->p[i] = NULL;
 169 }
 170
 171 #ifdef CONFIG_PROC_FS
 172 static const struct file_operations sysvipc_proc_fops;
 173 /**
 174  *      ipc_init_proc_interface -  Create a proc interface for sysipc types using a seq_file interface.
 175  *      @path: Path in procfs
 176  *      @header: Banner to be printed at the beginning of the file.
 177  *      @ids: ipc id table to iterate.
 178  *      @show: show routine.
 179  */
 180 void __init ipc_init_proc_interface(const char *path, const char *header,
 181                 int ids, int (*show)(struct seq_file *, void *))
 182 {
 183         struct proc_dir_entry *pde;
 184         struct ipc_proc_iface *iface;
 185
 186         iface = kmalloc(sizeof(*iface), GFP_KERNEL);
 187         if (!iface)
 188                 return;
 189         iface->path     = path;
 190         iface->header   = header;
 191         iface->ids      = ids;
 192         iface->show     = show;
 193
 194         pde = create_proc_entry(path,
 195                                 S_IRUGO,        /* world readable */
 196                                 NULL            /* parent dir */);
 197         if (pde) {
 198                 pde->data = iface;
 199                 pde->proc_fops = &sysvipc_proc_fops;
 200         } else {
 201                 kfree(iface);
 202         }
 203 }
 204 #endif
 205
 206 /**
 207  *      ipc_findkey     -       find a key in an ipc identifier set
 208  *      @ids: Identifier set
 209  *      @key: The key to find
 210  *
 211  *      Requires ipc_ids.mutex locked.
 212  *      Returns the identifier if found or -1 if not.
 213  */
 214
 215 int ipc_findkey(struct ipc_ids* ids, key_t key)
 216 {
 217         int id;
 218         struct kern_ipc_perm* p;
 219         int max_id = ids->max_id;
 220
 221         /*
 222          * rcu_dereference() is not needed here
 223          * since ipc_ids.mutex is held
 224          */
 225         for (id = 0; id <= max_id; id++) {
 226                 p = ids->entries->p[id];
 227                 if(p==NULL)
 228                         continue;
 229                 if (key == p->key)
 230                         return id;
 231         }
 232         return -1;
 233 }
 234
 235 /*
 236  * Requires ipc_ids.mutex locked
 237  */
 238 static int grow_ary(struct ipc_ids* ids, int newsize)
 239 {
 240         struct ipc_id_ary* new;
 241         struct ipc_id_ary* old;
 242         int i;
 243         int size = ids->entries->size;
 244
 245         if(newsize > IPCMNI)
 246                 newsize = IPCMNI;
 247         if(newsize <= size)
 248                 return newsize;
 249
 250         new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
 251                             sizeof(struct ipc_id_ary));
 252         if(new == NULL)
 253                 return size;
 254         new->size = newsize;
 255         memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
 256         for(i=size;i<newsize;i++) {
 257                 new->p[i] = NULL;
 258         }
 259         old = ids->entries;
 260
 261         /*
 262          * Use rcu_assign_pointer() to make sure the memcpyed contents
 263          * of the new array are visible before the new array becomes visible.
 264          */
 265         rcu_assign_pointer(ids->entries, new);
 266
 267         __ipc_fini_ids(ids, old);
 268         return newsize;
 269 }
 270
 271 /**
 272  *      ipc_addid       -       add an IPC identifier
 273  *      @ids: IPC identifier set
 274  *      @new: new IPC permission set
 275  *      @size: new size limit for the id array
 276  *
 277  *      Add an entry 'new' to the IPC arrays. The permissions object is
 278  *      initialised and the first free entry is set up and the id assigned
 279  *      is returned. The list is returned in a locked state on success.
 280  *      On failure the list is not locked and -1 is returned.
 281  *
 282  *      Called with ipc_ids.mutex held.
 283  */
 284
 285 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
 286 {
 287         int id;
 288
 289         size = grow_ary(ids,size);
 290
 291         /*
 292          * rcu_dereference()() is not needed here since
 293          * ipc_ids.mutex is held
 294          */
 295         for (id = 0; id < size; id++) {
 296                 if(ids->entries->p[id] == NULL)
 297                         goto found;
 298         }
 299         return -1;
 300 found:
 301         ids->in_use++;
 302         if (id > ids->max_id)
 303                 ids->max_id = id;
 304
 305         new->cuid = new->uid = current->euid;
 306         new->gid = new->cgid = current->egid;
 307
 308         new->seq = ids->seq++;
 309         if(ids->seq > ids->seq_max)
 310                 ids->seq = 0;
 311
 312         spin_lock_init(&new->lock);
 313         new->deleted = 0;
 314         rcu_read_lock();
 315         spin_lock(&new->lock);
 316         ids->entries->p[id] = new;
 317         return id;
 318 }
 319
 320 /**
 321  *      ipc_rmid        -       remove an IPC identifier
 322  *      @ids: identifier set
 323  *      @id: Identifier to remove
 324  *
 325  *      The identifier must be valid, and in use. The kernel will panic if
 326  *      fed an invalid identifier. The entry is removed and internal
 327  *      variables recomputed. The object associated with the identifier
 328  *      is returned.
 329  *      ipc_ids.mutex and the spinlock for this ID is hold before this function
 330  *      is called, and remain locked on the exit.
 331  */
 332
 333 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
 334 {
 335         struct kern_ipc_perm* p;
 336         int lid = id % SEQ_MULTIPLIER;
 337         BUG_ON(lid >= ids->entries->size);
 338
 339         /*
 340          * do not need a rcu_dereference()() here to force ordering
 341          * on Alpha, since the ipc_ids.mutex is held.
 342          */
 343         p = ids->entries->p[lid];
 344         ids->entries->p[lid] = NULL;
 345         BUG_ON(p==NULL);
 346         ids->in_use--;
 347
 348         if (lid == ids->max_id) {
 349                 do {
 350                         lid--;
 351                         if(lid == -1)
 352                                 break;
 353                 } while (ids->entries->p[lid] == NULL);
 354                 ids->max_id = lid;
 355         }
 356         p->deleted = 1;
 357         return p;
 358 }
 359
 360 /**
 361  *      ipc_alloc       -       allocate ipc space
 362  *      @size: size desired
 363  *
 364  *      Allocate memory from the appropriate pools and return a pointer to it.
 365  *      NULL is returned if the allocation fails
 366  */
 367
 368 void* ipc_alloc(int size)
 369 {
 370         void* out;
 371         if(size > PAGE_SIZE)
 372                 out = vmalloc(size);
 373         else
 374                 out = kmalloc(size, GFP_KERNEL);
 375         return out;
 376 }
 377
 378 /**
 379  *      ipc_free        -       free ipc space
 380  *      @ptr: pointer returned by ipc_alloc
 381  *      @size: size of block
 382  *
 383  *      Free a block created with ipc_alloc(). The caller must know the size
 384  *      used in the allocation call.
 385  */
 386
 387 void ipc_free(void* ptr, int size)
 388 {
 389         if(size > PAGE_SIZE)
 390                 vfree(ptr);
 391         else
 392                 kfree(ptr);
 393 }
 394
 395 /*
 396  * rcu allocations:
 397  * There are three headers that are prepended to the actual allocation:
 398  * - during use: ipc_rcu_hdr.
 399  * - during the rcu grace period: ipc_rcu_grace.
 400  * - [only if vmalloc]: ipc_rcu_sched.
 401  * Their lifetime doesn't overlap, thus the headers share the same memory.
 402  * Unlike a normal union, they are right-aligned, thus some container_of
 403  * forward/backward casting is necessary:
 404  */
 405 struct ipc_rcu_hdr
 406 {
 407         int refcount;
 408         int is_vmalloc;
 409         void *data[0];
 410 };
 411
 412
 413 struct ipc_rcu_grace
 414 {
 415         struct rcu_head rcu;
 416         /* "void *" makes sure alignment of following data is sane. */
 417         void *data[0];
 418 };
 419
 420 struct ipc_rcu_sched
 421 {
 422         struct work_struct work;
 423         /* "void *" makes sure alignment of following data is sane. */
 424         void *data[0];
 425 };
 426
 427 #define HDRLEN_KMALLOC          (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
 428                                         sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
 429 #define HDRLEN_VMALLOC          (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
 430                                         sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
 431
 432 static inline int rcu_use_vmalloc(int size)
 433 {
 434         /* Too big for a single page? */
 435         if (HDRLEN_KMALLOC + size > PAGE_SIZE)
 436                 return 1;
 437         return 0;
 438 }
 439
 440 /**
 441  *      ipc_rcu_alloc   -       allocate ipc and rcu space
 442  *      @size: size desired
 443  *
 444  *      Allocate memory for the rcu header structure +  the object.
 445  *      Returns the pointer to the object.
 446  *      NULL is returned if the allocation fails.
 447  */
 448
 449 void* ipc_rcu_alloc(int size)
 450 {
 451         void* out;
 452         /*
 453          * We prepend the allocation with the rcu struct, and
 454          * workqueue if necessary (for vmalloc).
 455          */
 456         if (rcu_use_vmalloc(size)) {
 457                 out = vmalloc(HDRLEN_VMALLOC + size);
 458                 if (out) {
 459                         out += HDRLEN_VMALLOC;
 460                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
 461                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 462                 }
 463         } else {
 464                 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
 465                 if (out) {
 466                         out += HDRLEN_KMALLOC;
 467                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
 468                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 469                 }
 470         }
 471
 472         return out;
 473 }
 474
 475 void ipc_rcu_getref(void *ptr)
 476 {
 477         container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
 478 }
 479
 480 static void ipc_do_vfree(struct work_struct *work)
 481 {
 482         vfree(container_of(work, struct ipc_rcu_sched, work));
 483 }
 484
 485 /**
 486  * ipc_schedule_free - free ipc + rcu space
 487  * @head: RCU callback structure for queued work
 488  *
 489  * Since RCU callback function is called in bh,
 490  * we need to defer the vfree to schedule_work().
 491  */
 492 static void ipc_schedule_free(struct rcu_head *head)
 493 {
 494         struct ipc_rcu_grace *grace =
 495                 container_of(head, struct ipc_rcu_grace, rcu);
 496         struct ipc_rcu_sched *sched =
 497                         container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
 498
 499         INIT_WORK(&sched->work, ipc_do_vfree);
 500         schedule_work(&sched->work);
 501 }
 502
 503 /**
 504  * ipc_immediate_free - free ipc + rcu space
 505  * @head: RCU callback structure that contains pointer to be freed
 506  *
 507  * Free from the RCU callback context.
 508  */
 509 static void ipc_immediate_free(struct rcu_head *head)
 510 {
 511         struct ipc_rcu_grace *free =
 512                 container_of(head, struct ipc_rcu_grace, rcu);
 513         kfree(free);
 514 }
 515
 516 void ipc_rcu_putref(void *ptr)
 517 {
 518         if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
 519                 return;
 520
 521         if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
 522                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 523                                 ipc_schedule_free);
 524         } else {
 525                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 526                                 ipc_immediate_free);
 527         }
 528 }
 529
 530 /**
 531  *      ipcperms        -       check IPC permissions
 532  *      @ipcp: IPC permission set
 533  *      @flag: desired permission set.
 534  *
 535  *      Check user, group, other permissions for access
 536  *      to ipc resources. return 0 if allowed
 537  */
 538
 539 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
 540 {       /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
 541         int requested_mode, granted_mode, err;
 542
 543         if (unlikely((err = audit_ipc_obj(ipcp))))
 544                 return err;
 545         requested_mode = (flag >> 6) | (flag >> 3) | flag;
 546         granted_mode = ipcp->mode;
 547         if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
 548                 granted_mode >>= 6;
 549         else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
 550                 granted_mode >>= 3;
 551         /* is there some bit set in requested_mode but not in granted_mode? */
 552         if ((requested_mode & ~granted_mode & 0007) &&
 553             !capable(CAP_IPC_OWNER))
 554                 return -1;
 555
 556         return security_ipc_permission(ipcp, flag);
 557 }
 558
 559 /*
 560  * Functions to convert between the kern_ipc_perm structure and the
 561  * old/new ipc_perm structures
 562  */
 563
 564 /**
 565  *      kernel_to_ipc64_perm    -       convert kernel ipc permissions to user
 566  *      @in: kernel permissions
 567  *      @out: new style IPC permissions
 568  *
 569  *      Turn the kernel object @in into a set of permissions descriptions
 570  *      for returning to userspace (@out).
 571  */
 572
 573
 574 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
 575 {
 576         out->key        = in->key;
 577         out->uid        = in->uid;
 578         out->gid        = in->gid;
 579         out->cuid       = in->cuid;
 580         out->cgid       = in->cgid;
 581         out->mode       = in->mode;
 582         out->seq        = in->seq;
 583 }
 584
 585 /**
 586  *      ipc64_perm_to_ipc_perm  -       convert old ipc permissions to new
 587  *      @in: new style IPC permissions
 588  *      @out: old style IPC permissions
 589  *
 590  *      Turn the new style permissions object @in into a compatibility
 591  *      object and store it into the @out pointer.
 592  */
 593
 594 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
 595 {
 596         out->key        = in->key;
 597         SET_UID(out->uid, in->uid);
 598         SET_GID(out->gid, in->gid);
 599         SET_UID(out->cuid, in->cuid);
 600         SET_GID(out->cgid, in->cgid);
 601         out->mode       = in->mode;
 602         out->seq        = in->seq;
 603 }
 604
 605 /*
 606  * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
 607  * is called with shm_ids.mutex locked.  Since grow_ary() is also called with
 608  * shm_ids.mutex down(for Shared Memory), there is no need to add read
 609  * barriers here to gurantee the writes in grow_ary() are seen in order
 610  * here (for Alpha).
 611  *
 612  * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
 613  * if in the future ipc_get() is used by other places without ipc_ids.mutex
 614  * down, then ipc_get() needs read memery barriers as ipc_lock() does.
 615  */
 616 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
 617 {
 618         struct kern_ipc_perm* out;
 619         int lid = id % SEQ_MULTIPLIER;
 620         if(lid >= ids->entries->size)
 621                 return NULL;
 622         out = ids->entries->p[lid];
 623         return out;
 624 }
 625
 626 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
 627 {
 628         struct kern_ipc_perm* out;
 629         int lid = id % SEQ_MULTIPLIER;
 630         struct ipc_id_ary* entries;
 631
 632         rcu_read_lock();
 633         entries = rcu_dereference(ids->entries);
 634         if(lid >= entries->size) {
 635                 rcu_read_unlock();
 636                 return NULL;
 637         }
 638         out = entries->p[lid];
 639         if(out == NULL) {
 640                 rcu_read_unlock();
 641                 return NULL;
 642         }
 643         spin_lock(&out->lock);
 644
 645         /* ipc_rmid() may have already freed the ID while ipc_lock
 646          * was spinning: here verify that the structure is still valid
 647          */
 648         if (out->deleted) {
 649                 spin_unlock(&out->lock);
 650                 rcu_read_unlock();
 651                 return NULL;
 652         }
 653         return out;
 654 }
 655
 656 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
 657 {
 658         rcu_read_lock();
 659         spin_lock(&perm->lock);
 660 }
 661
 662 void ipc_unlock(struct kern_ipc_perm* perm)
 663 {
 664         spin_unlock(&perm->lock);
 665         rcu_read_unlock();
 666 }
 667
 668 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
 669 {
 670         return SEQ_MULTIPLIER*seq + id;
 671 }
 672
 673 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
 674 {
 675         if(uid/SEQ_MULTIPLIER != ipcp->seq)
 676                 return 1;
 677         return 0;
 678 }
 679
 680 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
 681
 682
 683 /**
 684  *      ipc_parse_version       -       IPC call version
 685  *      @cmd: pointer to command
 686  *
 687  *      Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
 688  *      The @cmd value is turned from an encoding command and version into
 689  *      just the command code.
 690  */
 691
 692 int ipc_parse_version (int *cmd)
 693 {
 694         if (*cmd & IPC_64) {
 695                 *cmd ^= IPC_64;
 696                 return IPC_64;
 697         } else {
 698                 return IPC_OLD;
 699         }
 700 }
 701
 702 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
 703
 704 #ifdef CONFIG_PROC_FS
 705 struct ipc_proc_iter {
 706         struct ipc_namespace *ns;
 707         struct ipc_proc_iface *iface;
 708 };
 709
 710 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
 711 {
 712         struct ipc_proc_iter *iter = s->private;
 713         struct ipc_proc_iface *iface = iter->iface;
 714         struct kern_ipc_perm *ipc = it;
 715         loff_t p;
 716         struct ipc_ids *ids;
 717
 718         ids = iter->ns->ids[iface->ids];
 719
 720         /* If we had an ipc id locked before, unlock it */
 721         if (ipc && ipc != SEQ_START_TOKEN)
 722                 ipc_unlock(ipc);
 723
 724         /*
 725          * p = *pos - 1 (because id 0 starts at position 1)
 726          *          + 1 (because we increment the position by one)
 727          */
 728         for (p = *pos; p <= ids->max_id; p++) {
 729                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 730                         *pos = p + 1;
 731                         return ipc;
 732                 }
 733         }
 734
 735         /* Out of range - return NULL to terminate iteration */
 736         return NULL;
 737 }
 738
 739 /*
 740  * File positions: pos 0 -> header, pos n -> ipc id + 1.
 741  * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
 742  */
 743 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 744 {
 745         struct ipc_proc_iter *iter = s->private;
 746         struct ipc_proc_iface *iface = iter->iface;
 747         struct kern_ipc_perm *ipc;
 748         loff_t p;
 749         struct ipc_ids *ids;
 750
 751         ids = iter->ns->ids[iface->ids];
 752
 753         /*
 754          * Take the lock - this will be released by the corresponding
 755          * call to stop().
 756          */
 757         mutex_lock(&ids->mutex);
 758
 759         /* pos < 0 is invalid */
 760         if (*pos < 0)
 761                 return NULL;
 762
 763         /* pos == 0 means header */
 764         if (*pos == 0)
 765                 return SEQ_START_TOKEN;
 766
 767         /* Find the (pos-1)th ipc */
 768         for (p = *pos - 1; p <= ids->max_id; p++) {
 769                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 770                         *pos = p + 1;
 771                         return ipc;
 772                 }
 773         }
 774         return NULL;
 775 }
 776
 777 static void sysvipc_proc_stop(struct seq_file *s, void *it)
 778 {
 779         struct kern_ipc_perm *ipc = it;
 780         struct ipc_proc_iter *iter = s->private;
 781         struct ipc_proc_iface *iface = iter->iface;
 782         struct ipc_ids *ids;
 783
 784         /* If we had a locked segment, release it */
 785         if (ipc && ipc != SEQ_START_TOKEN)
 786                 ipc_unlock(ipc);
 787
 788         ids = iter->ns->ids[iface->ids];
 789         /* Release the lock we took in start() */
 790         mutex_unlock(&ids->mutex);
 791 }
 792
 793 static int sysvipc_proc_show(struct seq_file *s, void *it)
 794 {
 795         struct ipc_proc_iter *iter = s->private;
 796         struct ipc_proc_iface *iface = iter->iface;
 797
 798         if (it == SEQ_START_TOKEN)
 799                 return seq_puts(s, iface->header);
 800
 801         return iface->show(s, it);
 802 }
 803
 804 static struct seq_operations sysvipc_proc_seqops = {
 805         .start = sysvipc_proc_start,
 806         .stop  = sysvipc_proc_stop,
 807         .next  = sysvipc_proc_next,
 808         .show  = sysvipc_proc_show,
 809 };
 810
 811 static int sysvipc_proc_open(struct inode *inode, struct file *file)
 812 {
 813         int ret;
 814         struct seq_file *seq;
 815         struct ipc_proc_iter *iter;
 816
 817         ret = -ENOMEM;
 818         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 819         if (!iter)
 820                 goto out;
 821
 822         ret = seq_open(file, &sysvipc_proc_seqops);
 823         if (ret)
 824                 goto out_kfree;
 825
 826         seq = file->private_data;
 827         seq->private = iter;
 828
 829         iter->iface = PDE(inode)->data;
 830         iter->ns    = get_ipc_ns(current->nsproxy->ipc_ns);
 831 out:
 832         return ret;
 833 out_kfree:
 834         kfree(iter);
 835         goto out;
 836 }
 837
 838 static int sysvipc_proc_release(struct inode *inode, struct file *file)
 839 {
 840         struct seq_file *seq = file->private_data;
 841         struct ipc_proc_iter *iter = seq->private;
 842         put_ipc_ns(iter->ns);
 843         return seq_release_private(inode, file);
 844 }
 845
 846 static const struct file_operations sysvipc_proc_fops = {
 847         .open    = sysvipc_proc_open,
 848         .read    = seq_read,
 849         .llseek  = seq_lseek,
 850         .release = sysvipc_proc_release,
 851 };
 852 #endif /* CONFIG_PROC_FS */