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 #ifdef CONFIG_IPC_NS
  54 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
  55 {
  56         int err;
  57         struct ipc_namespace *ns;
  58
  59         err = -ENOMEM;
  60         ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
  61         if (ns == NULL)
  62                 goto err_mem;
  63
  64         err = sem_init_ns(ns);
  65         if (err)
  66                 goto err_sem;
  67         err = msg_init_ns(ns);
  68         if (err)
  69                 goto err_msg;
  70         err = shm_init_ns(ns);
  71         if (err)
  72                 goto err_shm;
  73
  74         kref_init(&ns->kref);
  75         return ns;
  76
  77 err_shm:
  78         msg_exit_ns(ns);
  79 err_msg:
  80         sem_exit_ns(ns);
  81 err_sem:
  82         kfree(ns);
  83 err_mem:
  84         return ERR_PTR(err);
  85 }
  86
  87 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
  88 {
  89         struct ipc_namespace *new_ns;
  90
  91         BUG_ON(!ns);
  92         get_ipc_ns(ns);
  93
  94         if (!(flags & CLONE_NEWIPC))
  95                 return ns;
  96
  97         new_ns = clone_ipc_ns(ns);
  98
  99         put_ipc_ns(ns);
 100         return new_ns;
 101 }
 102
 103 void free_ipc_ns(struct kref *kref)
 104 {
 105         struct ipc_namespace *ns;
 106
 107         ns = container_of(kref, struct ipc_namespace, kref);
 108         sem_exit_ns(ns);
 109         msg_exit_ns(ns);
 110         shm_exit_ns(ns);
 111         kfree(ns);
 112 }
 113 #else
 114 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
 115 {
 116         if (flags & CLONE_NEWIPC)
 117                 return ERR_PTR(-EINVAL);
 118         return ns;
 119 }
 120 #endif
 121
 122 /**
 123  *      ipc_init        -       initialise IPC subsystem
 124  *
 125  *      The various system5 IPC resources (semaphores, messages and shared
 126  *      memory) are initialised
 127  */
 128
 129 static int __init ipc_init(void)
 130 {
 131         sem_init();
 132         msg_init();
 133         shm_init();
 134         return 0;
 135 }
 136 __initcall(ipc_init);
 137
 138 /**
 139  *      ipc_init_ids            -       initialise IPC identifiers
 140  *      @ids: Identifier set
 141  *      @size: Number of identifiers
 142  *
 143  *      Given a size for the ipc identifier range (limited below IPCMNI)
 144  *      set up the sequence range to use then allocate and initialise the
 145  *      array itself.
 146  */
 147
 148 void __ipc_init ipc_init_ids(struct ipc_ids* ids, int size)
 149 {
 150         int i;
 151
 152         mutex_init(&ids->mutex);
 153
 154         if(size > IPCMNI)
 155                 size = IPCMNI;
 156         ids->in_use = 0;
 157         ids->max_id = -1;
 158         ids->seq = 0;
 159         {
 160                 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
 161                 if(seq_limit > USHRT_MAX)
 162                         ids->seq_max = USHRT_MAX;
 163                  else
 164                         ids->seq_max = seq_limit;
 165         }
 166
 167         ids->entries = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*size +
 168                                      sizeof(struct ipc_id_ary));
 169
 170         if(ids->entries == NULL) {
 171                 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
 172                 size = 0;
 173                 ids->entries = &ids->nullentry;
 174         }
 175         ids->entries->size = size;
 176         for(i=0;i<size;i++)
 177                 ids->entries->p[i] = NULL;
 178 }
 179
 180 #ifdef CONFIG_PROC_FS
 181 static const struct file_operations sysvipc_proc_fops;
 182 /**
 183  *      ipc_init_proc_interface -  Create a proc interface for sysipc types using a seq_file interface.
 184  *      @path: Path in procfs
 185  *      @header: Banner to be printed at the beginning of the file.
 186  *      @ids: ipc id table to iterate.
 187  *      @show: show routine.
 188  */
 189 void __init ipc_init_proc_interface(const char *path, const char *header,
 190                 int ids, int (*show)(struct seq_file *, void *))
 191 {
 192         struct proc_dir_entry *pde;
 193         struct ipc_proc_iface *iface;
 194
 195         iface = kmalloc(sizeof(*iface), GFP_KERNEL);
 196         if (!iface)
 197                 return;
 198         iface->path     = path;
 199         iface->header   = header;
 200         iface->ids      = ids;
 201         iface->show     = show;
 202
 203         pde = create_proc_entry(path,
 204                                 S_IRUGO,        /* world readable */
 205                                 NULL            /* parent dir */);
 206         if (pde) {
 207                 pde->data = iface;
 208                 pde->proc_fops = &sysvipc_proc_fops;
 209         } else {
 210                 kfree(iface);
 211         }
 212 }
 213 #endif
 214
 215 /**
 216  *      ipc_findkey     -       find a key in an ipc identifier set
 217  *      @ids: Identifier set
 218  *      @key: The key to find
 219  *
 220  *      Requires ipc_ids.mutex locked.
 221  *      Returns the identifier if found or -1 if not.
 222  */
 223
 224 int ipc_findkey(struct ipc_ids* ids, key_t key)
 225 {
 226         int id;
 227         struct kern_ipc_perm* p;
 228         int max_id = ids->max_id;
 229
 230         /*
 231          * rcu_dereference() is not needed here
 232          * since ipc_ids.mutex is held
 233          */
 234         for (id = 0; id <= max_id; id++) {
 235                 p = ids->entries->p[id];
 236                 if(p==NULL)
 237                         continue;
 238                 if (key == p->key)
 239                         return id;
 240         }
 241         return -1;
 242 }
 243
 244 /*
 245  * Requires ipc_ids.mutex locked
 246  */
 247 static int grow_ary(struct ipc_ids* ids, int newsize)
 248 {
 249         struct ipc_id_ary* new;
 250         struct ipc_id_ary* old;
 251         int i;
 252         int size = ids->entries->size;
 253
 254         if(newsize > IPCMNI)
 255                 newsize = IPCMNI;
 256         if(newsize <= size)
 257                 return newsize;
 258
 259         new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
 260                             sizeof(struct ipc_id_ary));
 261         if(new == NULL)
 262                 return size;
 263         new->size = newsize;
 264         memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
 265         for(i=size;i<newsize;i++) {
 266                 new->p[i] = NULL;
 267         }
 268         old = ids->entries;
 269
 270         /*
 271          * Use rcu_assign_pointer() to make sure the memcpyed contents
 272          * of the new array are visible before the new array becomes visible.
 273          */
 274         rcu_assign_pointer(ids->entries, new);
 275
 276         __ipc_fini_ids(ids, old);
 277         return newsize;
 278 }
 279
 280 /**
 281  *      ipc_addid       -       add an IPC identifier
 282  *      @ids: IPC identifier set
 283  *      @new: new IPC permission set
 284  *      @size: new size limit for the id array
 285  *
 286  *      Add an entry 'new' to the IPC arrays. The permissions object is
 287  *      initialised and the first free entry is set up and the id assigned
 288  *      is returned. The list is returned in a locked state on success.
 289  *      On failure the list is not locked and -1 is returned.
 290  *
 291  *      Called with ipc_ids.mutex held.
 292  */
 293
 294 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
 295 {
 296         int id;
 297
 298         size = grow_ary(ids,size);
 299
 300         /*
 301          * rcu_dereference()() is not needed here since
 302          * ipc_ids.mutex is held
 303          */
 304         for (id = 0; id < size; id++) {
 305                 if(ids->entries->p[id] == NULL)
 306                         goto found;
 307         }
 308         return -1;
 309 found:
 310         ids->in_use++;
 311         if (id > ids->max_id)
 312                 ids->max_id = id;
 313
 314         new->cuid = new->uid = current->euid;
 315         new->gid = new->cgid = current->egid;
 316
 317         new->seq = ids->seq++;
 318         if(ids->seq > ids->seq_max)
 319                 ids->seq = 0;
 320
 321         spin_lock_init(&new->lock);
 322         new->deleted = 0;
 323         rcu_read_lock();
 324         spin_lock(&new->lock);
 325         ids->entries->p[id] = new;
 326         return id;
 327 }
 328
 329 /**
 330  *      ipc_rmid        -       remove an IPC identifier
 331  *      @ids: identifier set
 332  *      @id: Identifier to remove
 333  *
 334  *      The identifier must be valid, and in use. The kernel will panic if
 335  *      fed an invalid identifier. The entry is removed and internal
 336  *      variables recomputed. The object associated with the identifier
 337  *      is returned.
 338  *      ipc_ids.mutex and the spinlock for this ID is hold before this function
 339  *      is called, and remain locked on the exit.
 340  */
 341
 342 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
 343 {
 344         struct kern_ipc_perm* p;
 345         int lid = id % SEQ_MULTIPLIER;
 346         BUG_ON(lid >= ids->entries->size);
 347
 348         /*
 349          * do not need a rcu_dereference()() here to force ordering
 350          * on Alpha, since the ipc_ids.mutex is held.
 351          */
 352         p = ids->entries->p[lid];
 353         ids->entries->p[lid] = NULL;
 354         BUG_ON(p==NULL);
 355         ids->in_use--;
 356
 357         if (lid == ids->max_id) {
 358                 do {
 359                         lid--;
 360                         if(lid == -1)
 361                                 break;
 362                 } while (ids->entries->p[lid] == NULL);
 363                 ids->max_id = lid;
 364         }
 365         p->deleted = 1;
 366         return p;
 367 }
 368
 369 /**
 370  *      ipc_alloc       -       allocate ipc space
 371  *      @size: size desired
 372  *
 373  *      Allocate memory from the appropriate pools and return a pointer to it.
 374  *      NULL is returned if the allocation fails
 375  */
 376
 377 void* ipc_alloc(int size)
 378 {
 379         void* out;
 380         if(size > PAGE_SIZE)
 381                 out = vmalloc(size);
 382         else
 383                 out = kmalloc(size, GFP_KERNEL);
 384         return out;
 385 }
 386
 387 /**
 388  *      ipc_free        -       free ipc space
 389  *      @ptr: pointer returned by ipc_alloc
 390  *      @size: size of block
 391  *
 392  *      Free a block created with ipc_alloc(). The caller must know the size
 393  *      used in the allocation call.
 394  */
 395
 396 void ipc_free(void* ptr, int size)
 397 {
 398         if(size > PAGE_SIZE)
 399                 vfree(ptr);
 400         else
 401                 kfree(ptr);
 402 }
 403
 404 /*
 405  * rcu allocations:
 406  * There are three headers that are prepended to the actual allocation:
 407  * - during use: ipc_rcu_hdr.
 408  * - during the rcu grace period: ipc_rcu_grace.
 409  * - [only if vmalloc]: ipc_rcu_sched.
 410  * Their lifetime doesn't overlap, thus the headers share the same memory.
 411  * Unlike a normal union, they are right-aligned, thus some container_of
 412  * forward/backward casting is necessary:
 413  */
 414 struct ipc_rcu_hdr
 415 {
 416         int refcount;
 417         int is_vmalloc;
 418         void *data[0];
 419 };
 420
 421
 422 struct ipc_rcu_grace
 423 {
 424         struct rcu_head rcu;
 425         /* "void *" makes sure alignment of following data is sane. */
 426         void *data[0];
 427 };
 428
 429 struct ipc_rcu_sched
 430 {
 431         struct work_struct work;
 432         /* "void *" makes sure alignment of following data is sane. */
 433         void *data[0];
 434 };
 435
 436 #define HDRLEN_KMALLOC          (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
 437                                         sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
 438 #define HDRLEN_VMALLOC          (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
 439                                         sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
 440
 441 static inline int rcu_use_vmalloc(int size)
 442 {
 443         /* Too big for a single page? */
 444         if (HDRLEN_KMALLOC + size > PAGE_SIZE)
 445                 return 1;
 446         return 0;
 447 }
 448
 449 /**
 450  *      ipc_rcu_alloc   -       allocate ipc and rcu space
 451  *      @size: size desired
 452  *
 453  *      Allocate memory for the rcu header structure +  the object.
 454  *      Returns the pointer to the object.
 455  *      NULL is returned if the allocation fails.
 456  */
 457
 458 void* ipc_rcu_alloc(int size)
 459 {
 460         void* out;
 461         /*
 462          * We prepend the allocation with the rcu struct, and
 463          * workqueue if necessary (for vmalloc).
 464          */
 465         if (rcu_use_vmalloc(size)) {
 466                 out = vmalloc(HDRLEN_VMALLOC + size);
 467                 if (out) {
 468                         out += HDRLEN_VMALLOC;
 469                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
 470                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 471                 }
 472         } else {
 473                 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
 474                 if (out) {
 475                         out += HDRLEN_KMALLOC;
 476                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
 477                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 478                 }
 479         }
 480
 481         return out;
 482 }
 483
 484 void ipc_rcu_getref(void *ptr)
 485 {
 486         container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
 487 }
 488
 489 static void ipc_do_vfree(struct work_struct *work)
 490 {
 491         vfree(container_of(work, struct ipc_rcu_sched, work));
 492 }
 493
 494 /**
 495  * ipc_schedule_free - free ipc + rcu space
 496  * @head: RCU callback structure for queued work
 497  *
 498  * Since RCU callback function is called in bh,
 499  * we need to defer the vfree to schedule_work().
 500  */
 501 static void ipc_schedule_free(struct rcu_head *head)
 502 {
 503         struct ipc_rcu_grace *grace =
 504                 container_of(head, struct ipc_rcu_grace, rcu);
 505         struct ipc_rcu_sched *sched =
 506                         container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
 507
 508         INIT_WORK(&sched->work, ipc_do_vfree);
 509         schedule_work(&sched->work);
 510 }
 511
 512 /**
 513  * ipc_immediate_free - free ipc + rcu space
 514  * @head: RCU callback structure that contains pointer to be freed
 515  *
 516  * Free from the RCU callback context.
 517  */
 518 static void ipc_immediate_free(struct rcu_head *head)
 519 {
 520         struct ipc_rcu_grace *free =
 521                 container_of(head, struct ipc_rcu_grace, rcu);
 522         kfree(free);
 523 }
 524
 525 void ipc_rcu_putref(void *ptr)
 526 {
 527         if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
 528                 return;
 529
 530         if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
 531                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 532                                 ipc_schedule_free);
 533         } else {
 534                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 535                                 ipc_immediate_free);
 536         }
 537 }
 538
 539 /**
 540  *      ipcperms        -       check IPC permissions
 541  *      @ipcp: IPC permission set
 542  *      @flag: desired permission set.
 543  *
 544  *      Check user, group, other permissions for access
 545  *      to ipc resources. return 0 if allowed
 546  */
 547
 548 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
 549 {       /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
 550         int requested_mode, granted_mode, err;
 551
 552         if (unlikely((err = audit_ipc_obj(ipcp))))
 553                 return err;
 554         requested_mode = (flag >> 6) | (flag >> 3) | flag;
 555         granted_mode = ipcp->mode;
 556         if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
 557                 granted_mode >>= 6;
 558         else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
 559                 granted_mode >>= 3;
 560         /* is there some bit set in requested_mode but not in granted_mode? */
 561         if ((requested_mode & ~granted_mode & 0007) &&
 562             !capable(CAP_IPC_OWNER))
 563                 return -1;
 564
 565         return security_ipc_permission(ipcp, flag);
 566 }
 567
 568 /*
 569  * Functions to convert between the kern_ipc_perm structure and the
 570  * old/new ipc_perm structures
 571  */
 572
 573 /**
 574  *      kernel_to_ipc64_perm    -       convert kernel ipc permissions to user
 575  *      @in: kernel permissions
 576  *      @out: new style IPC permissions
 577  *
 578  *      Turn the kernel object @in into a set of permissions descriptions
 579  *      for returning to userspace (@out).
 580  */
 581
 582
 583 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
 584 {
 585         out->key        = in->key;
 586         out->uid        = in->uid;
 587         out->gid        = in->gid;
 588         out->cuid       = in->cuid;
 589         out->cgid       = in->cgid;
 590         out->mode       = in->mode;
 591         out->seq        = in->seq;
 592 }
 593
 594 /**
 595  *      ipc64_perm_to_ipc_perm  -       convert old ipc permissions to new
 596  *      @in: new style IPC permissions
 597  *      @out: old style IPC permissions
 598  *
 599  *      Turn the new style permissions object @in into a compatibility
 600  *      object and store it into the @out pointer.
 601  */
 602
 603 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
 604 {
 605         out->key        = in->key;
 606         SET_UID(out->uid, in->uid);
 607         SET_GID(out->gid, in->gid);
 608         SET_UID(out->cuid, in->cuid);
 609         SET_GID(out->cgid, in->cgid);
 610         out->mode       = in->mode;
 611         out->seq        = in->seq;
 612 }
 613
 614 /*
 615  * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
 616  * is called with shm_ids.mutex locked.  Since grow_ary() is also called with
 617  * shm_ids.mutex down(for Shared Memory), there is no need to add read
 618  * barriers here to gurantee the writes in grow_ary() are seen in order
 619  * here (for Alpha).
 620  *
 621  * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
 622  * if in the future ipc_get() is used by other places without ipc_ids.mutex
 623  * down, then ipc_get() needs read memery barriers as ipc_lock() does.
 624  */
 625 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
 626 {
 627         struct kern_ipc_perm* out;
 628         int lid = id % SEQ_MULTIPLIER;
 629         if(lid >= ids->entries->size)
 630                 return NULL;
 631         out = ids->entries->p[lid];
 632         return out;
 633 }
 634
 635 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
 636 {
 637         struct kern_ipc_perm* out;
 638         int lid = id % SEQ_MULTIPLIER;
 639         struct ipc_id_ary* entries;
 640
 641         rcu_read_lock();
 642         entries = rcu_dereference(ids->entries);
 643         if(lid >= entries->size) {
 644                 rcu_read_unlock();
 645                 return NULL;
 646         }
 647         out = entries->p[lid];
 648         if(out == NULL) {
 649                 rcu_read_unlock();
 650                 return NULL;
 651         }
 652         spin_lock(&out->lock);
 653
 654         /* ipc_rmid() may have already freed the ID while ipc_lock
 655          * was spinning: here verify that the structure is still valid
 656          */
 657         if (out->deleted) {
 658                 spin_unlock(&out->lock);
 659                 rcu_read_unlock();
 660                 return NULL;
 661         }
 662         return out;
 663 }
 664
 665 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
 666 {
 667         rcu_read_lock();
 668         spin_lock(&perm->lock);
 669 }
 670
 671 void ipc_unlock(struct kern_ipc_perm* perm)
 672 {
 673         spin_unlock(&perm->lock);
 674         rcu_read_unlock();
 675 }
 676
 677 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
 678 {
 679         return SEQ_MULTIPLIER*seq + id;
 680 }
 681
 682 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
 683 {
 684         if(uid/SEQ_MULTIPLIER != ipcp->seq)
 685                 return 1;
 686         return 0;
 687 }
 688
 689 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
 690
 691
 692 /**
 693  *      ipc_parse_version       -       IPC call version
 694  *      @cmd: pointer to command
 695  *
 696  *      Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
 697  *      The @cmd value is turned from an encoding command and version into
 698  *      just the command code.
 699  */
 700
 701 int ipc_parse_version (int *cmd)
 702 {
 703         if (*cmd & IPC_64) {
 704                 *cmd ^= IPC_64;
 705                 return IPC_64;
 706         } else {
 707                 return IPC_OLD;
 708         }
 709 }
 710
 711 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
 712
 713 #ifdef CONFIG_PROC_FS
 714 struct ipc_proc_iter {
 715         struct ipc_namespace *ns;
 716         struct ipc_proc_iface *iface;
 717 };
 718
 719 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
 720 {
 721         struct ipc_proc_iter *iter = s->private;
 722         struct ipc_proc_iface *iface = iter->iface;
 723         struct kern_ipc_perm *ipc = it;
 724         loff_t p;
 725         struct ipc_ids *ids;
 726
 727         ids = iter->ns->ids[iface->ids];
 728
 729         /* If we had an ipc id locked before, unlock it */
 730         if (ipc && ipc != SEQ_START_TOKEN)
 731                 ipc_unlock(ipc);
 732
 733         /*
 734          * p = *pos - 1 (because id 0 starts at position 1)
 735          *          + 1 (because we increment the position by one)
 736          */
 737         for (p = *pos; p <= ids->max_id; p++) {
 738                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 739                         *pos = p + 1;
 740                         return ipc;
 741                 }
 742         }
 743
 744         /* Out of range - return NULL to terminate iteration */
 745         return NULL;
 746 }
 747
 748 /*
 749  * File positions: pos 0 -> header, pos n -> ipc id + 1.
 750  * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
 751  */
 752 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 753 {
 754         struct ipc_proc_iter *iter = s->private;
 755         struct ipc_proc_iface *iface = iter->iface;
 756         struct kern_ipc_perm *ipc;
 757         loff_t p;
 758         struct ipc_ids *ids;
 759
 760         ids = iter->ns->ids[iface->ids];
 761
 762         /*
 763          * Take the lock - this will be released by the corresponding
 764          * call to stop().
 765          */
 766         mutex_lock(&ids->mutex);
 767
 768         /* pos < 0 is invalid */
 769         if (*pos < 0)
 770                 return NULL;
 771
 772         /* pos == 0 means header */
 773         if (*pos == 0)
 774                 return SEQ_START_TOKEN;
 775
 776         /* Find the (pos-1)th ipc */
 777         for (p = *pos - 1; p <= ids->max_id; p++) {
 778                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 779                         *pos = p + 1;
 780                         return ipc;
 781                 }
 782         }
 783         return NULL;
 784 }
 785
 786 static void sysvipc_proc_stop(struct seq_file *s, void *it)
 787 {
 788         struct kern_ipc_perm *ipc = it;
 789         struct ipc_proc_iter *iter = s->private;
 790         struct ipc_proc_iface *iface = iter->iface;
 791         struct ipc_ids *ids;
 792
 793         /* If we had a locked segment, release it */
 794         if (ipc && ipc != SEQ_START_TOKEN)
 795                 ipc_unlock(ipc);
 796
 797         ids = iter->ns->ids[iface->ids];
 798         /* Release the lock we took in start() */
 799         mutex_unlock(&ids->mutex);
 800 }
 801
 802 static int sysvipc_proc_show(struct seq_file *s, void *it)
 803 {
 804         struct ipc_proc_iter *iter = s->private;
 805         struct ipc_proc_iface *iface = iter->iface;
 806
 807         if (it == SEQ_START_TOKEN)
 808                 return seq_puts(s, iface->header);
 809
 810         return iface->show(s, it);
 811 }
 812
 813 static struct seq_operations sysvipc_proc_seqops = {
 814         .start = sysvipc_proc_start,
 815         .stop  = sysvipc_proc_stop,
 816         .next  = sysvipc_proc_next,
 817         .show  = sysvipc_proc_show,
 818 };
 819
 820 static int sysvipc_proc_open(struct inode *inode, struct file *file)
 821 {
 822         int ret;
 823         struct seq_file *seq;
 824         struct ipc_proc_iter *iter;
 825
 826         ret = -ENOMEM;
 827         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 828         if (!iter)
 829                 goto out;
 830
 831         ret = seq_open(file, &sysvipc_proc_seqops);
 832         if (ret)
 833                 goto out_kfree;
 834
 835         seq = file->private_data;
 836         seq->private = iter;
 837
 838         iter->iface = PDE(inode)->data;
 839         iter->ns    = get_ipc_ns(current->nsproxy->ipc_ns);
 840 out:
 841         return ret;
 842 out_kfree:
 843         kfree(iter);
 844         goto out;
 845 }
 846
 847 static int sysvipc_proc_release(struct inode *inode, struct file *file)
 848 {
 849         struct seq_file *seq = file->private_data;
 850         struct ipc_proc_iter *iter = seq->private;
 851         put_ipc_ns(iter->ns);
 852         return seq_release_private(inode, file);
 853 }
 854
 855 static const struct file_operations sysvipc_proc_fops = {
 856         .open    = sysvipc_proc_open,
 857         .read    = seq_read,
 858         .llseek  = seq_lseek,
 859         .release = sysvipc_proc_release,
 860 };
 861 #endif /* CONFIG_PROC_FS */