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/smp_lock.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/slab.h>
  27 #include <linux/capability.h>
  28 #include <linux/highuid.h>
  29 #include <linux/security.h>
  30 #include <linux/rcupdate.h>
  31 #include <linux/workqueue.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/proc_fs.h>
  34 #include <linux/audit.h>
  35 #include <linux/nsproxy.h>
  36
  37 #include <asm/unistd.h>
  38
  39 #include "util.h"
  40
  41 struct ipc_proc_iface {
  42         const char *path;
  43         const char *header;
  44         int ids;
  45         int (*show)(struct seq_file *, void *);
  46 };
  47
  48 struct ipc_namespace init_ipc_ns = {
  49         .kref = {
  50                 .refcount       = ATOMIC_INIT(2),
  51         },
  52 };
  53
  54 #ifdef CONFIG_IPC_NS
  55 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
  56 {
  57         int err;
  58         struct ipc_namespace *ns;
  59
  60         err = -ENOMEM;
  61         ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
  62         if (ns == NULL)
  63                 goto err_mem;
  64
  65         err = sem_init_ns(ns);
  66         if (err)
  67                 goto err_sem;
  68         err = msg_init_ns(ns);
  69         if (err)
  70                 goto err_msg;
  71         err = shm_init_ns(ns);
  72         if (err)
  73                 goto err_shm;
  74
  75         kref_init(&ns->kref);
  76         return ns;
  77
  78 err_shm:
  79         msg_exit_ns(ns);
  80 err_msg:
  81         sem_exit_ns(ns);
  82 err_sem:
  83         kfree(ns);
  84 err_mem:
  85         return ERR_PTR(err);
  86 }
  87
  88 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
  89 {
  90         struct ipc_namespace *new_ns;
  91
  92         BUG_ON(!ns);
  93         get_ipc_ns(ns);
  94
  95         if (!(flags & CLONE_NEWIPC))
  96                 return ns;
  97
  98         new_ns = clone_ipc_ns(ns);
  99
 100         put_ipc_ns(ns);
 101         return new_ns;
 102 }
 103
 104 void free_ipc_ns(struct kref *kref)
 105 {
 106         struct ipc_namespace *ns;
 107
 108         ns = container_of(kref, struct ipc_namespace, kref);
 109         sem_exit_ns(ns);
 110         msg_exit_ns(ns);
 111         shm_exit_ns(ns);
 112         kfree(ns);
 113 }
 114 #else
 115 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
 116 {
 117         if (flags & CLONE_NEWIPC)
 118                 return ERR_PTR(-EINVAL);
 119         return ns;
 120 }
 121 #endif
 122
 123 /**
 124  *      ipc_init        -       initialise IPC subsystem
 125  *
 126  *      The various system5 IPC resources (semaphores, messages and shared
 127  *      memory) are initialised
 128  */
 129
 130 static int __init ipc_init(void)
 131 {
 132         sem_init();
 133         msg_init();
 134         shm_init();
 135         return 0;
 136 }
 137 __initcall(ipc_init);
 138
 139 /**
 140  *      ipc_init_ids            -       initialise IPC identifiers
 141  *      @ids: Identifier set
 142  *      @size: Number of identifiers
 143  *
 144  *      Given a size for the ipc identifier range (limited below IPCMNI)
 145  *      set up the sequence range to use then allocate and initialise the
 146  *      array itself.
 147  */
 148
 149 void __ipc_init ipc_init_ids(struct ipc_ids* ids, int size)
 150 {
 151         int i;
 152
 153         mutex_init(&ids->mutex);
 154
 155         if(size > IPCMNI)
 156                 size = IPCMNI;
 157         ids->in_use = 0;
 158         ids->max_id = -1;
 159         ids->seq = 0;
 160         {
 161                 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
 162                 if(seq_limit > USHRT_MAX)
 163                         ids->seq_max = USHRT_MAX;
 164                  else
 165                         ids->seq_max = seq_limit;
 166         }
 167
 168         ids->entries = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*size +
 169                                      sizeof(struct ipc_id_ary));
 170
 171         if(ids->entries == NULL) {
 172                 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
 173                 size = 0;
 174                 ids->entries = &ids->nullentry;
 175         }
 176         ids->entries->size = size;
 177         for(i=0;i<size;i++)
 178                 ids->entries->p[i] = NULL;
 179 }
 180
 181 #ifdef CONFIG_PROC_FS
 182 static const struct file_operations sysvipc_proc_fops;
 183 /**
 184  *      ipc_init_proc_interface -  Create a proc interface for sysipc types using a seq_file interface.
 185  *      @path: Path in procfs
 186  *      @header: Banner to be printed at the beginning of the file.
 187  *      @ids: ipc id table to iterate.
 188  *      @show: show routine.
 189  */
 190 void __init ipc_init_proc_interface(const char *path, const char *header,
 191                 int ids, int (*show)(struct seq_file *, void *))
 192 {
 193         struct proc_dir_entry *pde;
 194         struct ipc_proc_iface *iface;
 195
 196         iface = kmalloc(sizeof(*iface), GFP_KERNEL);
 197         if (!iface)
 198                 return;
 199         iface->path     = path;
 200         iface->header   = header;
 201         iface->ids      = ids;
 202         iface->show     = show;
 203
 204         pde = create_proc_entry(path,
 205                                 S_IRUGO,        /* world readable */
 206                                 NULL            /* parent dir */);
 207         if (pde) {
 208                 pde->data = iface;
 209                 pde->proc_fops = &sysvipc_proc_fops;
 210         } else {
 211                 kfree(iface);
 212         }
 213 }
 214 #endif
 215
 216 /**
 217  *      ipc_findkey     -       find a key in an ipc identifier set
 218  *      @ids: Identifier set
 219  *      @key: The key to find
 220  *
 221  *      Requires ipc_ids.mutex locked.
 222  *      Returns the identifier if found or -1 if not.
 223  */
 224
 225 int ipc_findkey(struct ipc_ids* ids, key_t key)
 226 {
 227         int id;
 228         struct kern_ipc_perm* p;
 229         int max_id = ids->max_id;
 230
 231         /*
 232          * rcu_dereference() is not needed here
 233          * since ipc_ids.mutex is held
 234          */
 235         for (id = 0; id <= max_id; id++) {
 236                 p = ids->entries->p[id];
 237                 if(p==NULL)
 238                         continue;
 239                 if (key == p->key)
 240                         return id;
 241         }
 242         return -1;
 243 }
 244
 245 /*
 246  * Requires ipc_ids.mutex locked
 247  */
 248 static int grow_ary(struct ipc_ids* ids, int newsize)
 249 {
 250         struct ipc_id_ary* new;
 251         struct ipc_id_ary* old;
 252         int i;
 253         int size = ids->entries->size;
 254
 255         if(newsize > IPCMNI)
 256                 newsize = IPCMNI;
 257         if(newsize <= size)
 258                 return newsize;
 259
 260         new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
 261                             sizeof(struct ipc_id_ary));
 262         if(new == NULL)
 263                 return size;
 264         new->size = newsize;
 265         memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
 266         for(i=size;i<newsize;i++) {
 267                 new->p[i] = NULL;
 268         }
 269         old = ids->entries;
 270
 271         /*
 272          * Use rcu_assign_pointer() to make sure the memcpyed contents
 273          * of the new array are visible before the new array becomes visible.
 274          */
 275         rcu_assign_pointer(ids->entries, new);
 276
 277         __ipc_fini_ids(ids, old);
 278         return newsize;
 279 }
 280
 281 /**
 282  *      ipc_addid       -       add an IPC identifier
 283  *      @ids: IPC identifier set
 284  *      @new: new IPC permission set
 285  *      @size: new size limit for the id array
 286  *
 287  *      Add an entry 'new' to the IPC arrays. The permissions object is
 288  *      initialised and the first free entry is set up and the id assigned
 289  *      is returned. The list is returned in a locked state on success.
 290  *      On failure the list is not locked and -1 is returned.
 291  *
 292  *      Called with ipc_ids.mutex held.
 293  */
 294
 295 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
 296 {
 297         int id;
 298
 299         size = grow_ary(ids,size);
 300
 301         /*
 302          * rcu_dereference()() is not needed here since
 303          * ipc_ids.mutex is held
 304          */
 305         for (id = 0; id < size; id++) {
 306                 if(ids->entries->p[id] == NULL)
 307                         goto found;
 308         }
 309         return -1;
 310 found:
 311         ids->in_use++;
 312         if (id > ids->max_id)
 313                 ids->max_id = id;
 314
 315         new->cuid = new->uid = current->euid;
 316         new->gid = new->cgid = current->egid;
 317
 318         new->seq = ids->seq++;
 319         if(ids->seq > ids->seq_max)
 320                 ids->seq = 0;
 321
 322         spin_lock_init(&new->lock);
 323         new->deleted = 0;
 324         rcu_read_lock();
 325         spin_lock(&new->lock);
 326         ids->entries->p[id] = new;
 327         return id;
 328 }
 329
 330 /**
 331  *      ipc_rmid        -       remove an IPC identifier
 332  *      @ids: identifier set
 333  *      @id: Identifier to remove
 334  *
 335  *      The identifier must be valid, and in use. The kernel will panic if
 336  *      fed an invalid identifier. The entry is removed and internal
 337  *      variables recomputed. The object associated with the identifier
 338  *      is returned.
 339  *      ipc_ids.mutex and the spinlock for this ID is hold before this function
 340  *      is called, and remain locked on the exit.
 341  */
 342
 343 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
 344 {
 345         struct kern_ipc_perm* p;
 346         int lid = id % SEQ_MULTIPLIER;
 347         BUG_ON(lid >= ids->entries->size);
 348
 349         /*
 350          * do not need a rcu_dereference()() here to force ordering
 351          * on Alpha, since the ipc_ids.mutex is held.
 352          */
 353         p = ids->entries->p[lid];
 354         ids->entries->p[lid] = NULL;
 355         BUG_ON(p==NULL);
 356         ids->in_use--;
 357
 358         if (lid == ids->max_id) {
 359                 do {
 360                         lid--;
 361                         if(lid == -1)
 362                                 break;
 363                 } while (ids->entries->p[lid] == NULL);
 364                 ids->max_id = lid;
 365         }
 366         p->deleted = 1;
 367         return p;
 368 }
 369
 370 /**
 371  *      ipc_alloc       -       allocate ipc space
 372  *      @size: size desired
 373  *
 374  *      Allocate memory from the appropriate pools and return a pointer to it.
 375  *      NULL is returned if the allocation fails
 376  */
 377
 378 void* ipc_alloc(int size)
 379 {
 380         void* out;
 381         if(size > PAGE_SIZE)
 382                 out = vmalloc(size);
 383         else
 384                 out = kmalloc(size, GFP_KERNEL);
 385         return out;
 386 }
 387
 388 /**
 389  *      ipc_free        -       free ipc space
 390  *      @ptr: pointer returned by ipc_alloc
 391  *      @size: size of block
 392  *
 393  *      Free a block created with ipc_alloc(). The caller must know the size
 394  *      used in the allocation call.
 395  */
 396
 397 void ipc_free(void* ptr, int size)
 398 {
 399         if(size > PAGE_SIZE)
 400                 vfree(ptr);
 401         else
 402                 kfree(ptr);
 403 }
 404
 405 /*
 406  * rcu allocations:
 407  * There are three headers that are prepended to the actual allocation:
 408  * - during use: ipc_rcu_hdr.
 409  * - during the rcu grace period: ipc_rcu_grace.
 410  * - [only if vmalloc]: ipc_rcu_sched.
 411  * Their lifetime doesn't overlap, thus the headers share the same memory.
 412  * Unlike a normal union, they are right-aligned, thus some container_of
 413  * forward/backward casting is necessary:
 414  */
 415 struct ipc_rcu_hdr
 416 {
 417         int refcount;
 418         int is_vmalloc;
 419         void *data[0];
 420 };
 421
 422
 423 struct ipc_rcu_grace
 424 {
 425         struct rcu_head rcu;
 426         /* "void *" makes sure alignment of following data is sane. */
 427         void *data[0];
 428 };
 429
 430 struct ipc_rcu_sched
 431 {
 432         struct work_struct work;
 433         /* "void *" makes sure alignment of following data is sane. */
 434         void *data[0];
 435 };
 436
 437 #define HDRLEN_KMALLOC          (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
 438                                         sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
 439 #define HDRLEN_VMALLOC          (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
 440                                         sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
 441
 442 static inline int rcu_use_vmalloc(int size)
 443 {
 444         /* Too big for a single page? */
 445         if (HDRLEN_KMALLOC + size > PAGE_SIZE)
 446                 return 1;
 447         return 0;
 448 }
 449
 450 /**
 451  *      ipc_rcu_alloc   -       allocate ipc and rcu space
 452  *      @size: size desired
 453  *
 454  *      Allocate memory for the rcu header structure +  the object.
 455  *      Returns the pointer to the object.
 456  *      NULL is returned if the allocation fails.
 457  */
 458
 459 void* ipc_rcu_alloc(int size)
 460 {
 461         void* out;
 462         /*
 463          * We prepend the allocation with the rcu struct, and
 464          * workqueue if necessary (for vmalloc).
 465          */
 466         if (rcu_use_vmalloc(size)) {
 467                 out = vmalloc(HDRLEN_VMALLOC + size);
 468                 if (out) {
 469                         out += HDRLEN_VMALLOC;
 470                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
 471                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 472                 }
 473         } else {
 474                 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
 475                 if (out) {
 476                         out += HDRLEN_KMALLOC;
 477                         container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
 478                         container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
 479                 }
 480         }
 481
 482         return out;
 483 }
 484
 485 void ipc_rcu_getref(void *ptr)
 486 {
 487         container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
 488 }
 489
 490 static void ipc_do_vfree(struct work_struct *work)
 491 {
 492         vfree(container_of(work, struct ipc_rcu_sched, work));
 493 }
 494
 495 /**
 496  * ipc_schedule_free - free ipc + rcu space
 497  * @head: RCU callback structure for queued work
 498  *
 499  * Since RCU callback function is called in bh,
 500  * we need to defer the vfree to schedule_work().
 501  */
 502 static void ipc_schedule_free(struct rcu_head *head)
 503 {
 504         struct ipc_rcu_grace *grace =
 505                 container_of(head, struct ipc_rcu_grace, rcu);
 506         struct ipc_rcu_sched *sched =
 507                         container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
 508
 509         INIT_WORK(&sched->work, ipc_do_vfree);
 510         schedule_work(&sched->work);
 511 }
 512
 513 /**
 514  * ipc_immediate_free - free ipc + rcu space
 515  * @head: RCU callback structure that contains pointer to be freed
 516  *
 517  * Free from the RCU callback context.
 518  */
 519 static void ipc_immediate_free(struct rcu_head *head)
 520 {
 521         struct ipc_rcu_grace *free =
 522                 container_of(head, struct ipc_rcu_grace, rcu);
 523         kfree(free);
 524 }
 525
 526 void ipc_rcu_putref(void *ptr)
 527 {
 528         if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
 529                 return;
 530
 531         if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
 532                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 533                                 ipc_schedule_free);
 534         } else {
 535                 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
 536                                 ipc_immediate_free);
 537         }
 538 }
 539
 540 /**
 541  *      ipcperms        -       check IPC permissions
 542  *      @ipcp: IPC permission set
 543  *      @flag: desired permission set.
 544  *
 545  *      Check user, group, other permissions for access
 546  *      to ipc resources. return 0 if allowed
 547  */
 548
 549 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
 550 {       /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
 551         int requested_mode, granted_mode, err;
 552
 553         if (unlikely((err = audit_ipc_obj(ipcp))))
 554                 return err;
 555         requested_mode = (flag >> 6) | (flag >> 3) | flag;
 556         granted_mode = ipcp->mode;
 557         if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
 558                 granted_mode >>= 6;
 559         else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
 560                 granted_mode >>= 3;
 561         /* is there some bit set in requested_mode but not in granted_mode? */
 562         if ((requested_mode & ~granted_mode & 0007) &&
 563             !capable(CAP_IPC_OWNER))
 564                 return -1;
 565
 566         return security_ipc_permission(ipcp, flag);
 567 }
 568
 569 /*
 570  * Functions to convert between the kern_ipc_perm structure and the
 571  * old/new ipc_perm structures
 572  */
 573
 574 /**
 575  *      kernel_to_ipc64_perm    -       convert kernel ipc permissions to user
 576  *      @in: kernel permissions
 577  *      @out: new style IPC permissions
 578  *
 579  *      Turn the kernel object @in into a set of permissions descriptions
 580  *      for returning to userspace (@out).
 581  */
 582
 583
 584 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
 585 {
 586         out->key        = in->key;
 587         out->uid        = in->uid;
 588         out->gid        = in->gid;
 589         out->cuid       = in->cuid;
 590         out->cgid       = in->cgid;
 591         out->mode       = in->mode;
 592         out->seq        = in->seq;
 593 }
 594
 595 /**
 596  *      ipc64_perm_to_ipc_perm  -       convert old ipc permissions to new
 597  *      @in: new style IPC permissions
 598  *      @out: old style IPC permissions
 599  *
 600  *      Turn the new style permissions object @in into a compatibility
 601  *      object and store it into the @out pointer.
 602  */
 603
 604 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
 605 {
 606         out->key        = in->key;
 607         SET_UID(out->uid, in->uid);
 608         SET_GID(out->gid, in->gid);
 609         SET_UID(out->cuid, in->cuid);
 610         SET_GID(out->cgid, in->cgid);
 611         out->mode       = in->mode;
 612         out->seq        = in->seq;
 613 }
 614
 615 /*
 616  * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
 617  * is called with shm_ids.mutex locked.  Since grow_ary() is also called with
 618  * shm_ids.mutex down(for Shared Memory), there is no need to add read
 619  * barriers here to gurantee the writes in grow_ary() are seen in order
 620  * here (for Alpha).
 621  *
 622  * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
 623  * if in the future ipc_get() is used by other places without ipc_ids.mutex
 624  * down, then ipc_get() needs read memery barriers as ipc_lock() does.
 625  */
 626 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
 627 {
 628         struct kern_ipc_perm* out;
 629         int lid = id % SEQ_MULTIPLIER;
 630         if(lid >= ids->entries->size)
 631                 return NULL;
 632         out = ids->entries->p[lid];
 633         return out;
 634 }
 635
 636 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
 637 {
 638         struct kern_ipc_perm* out;
 639         int lid = id % SEQ_MULTIPLIER;
 640         struct ipc_id_ary* entries;
 641
 642         rcu_read_lock();
 643         entries = rcu_dereference(ids->entries);
 644         if(lid >= entries->size) {
 645                 rcu_read_unlock();
 646                 return NULL;
 647         }
 648         out = entries->p[lid];
 649         if(out == NULL) {
 650                 rcu_read_unlock();
 651                 return NULL;
 652         }
 653         spin_lock(&out->lock);
 654
 655         /* ipc_rmid() may have already freed the ID while ipc_lock
 656          * was spinning: here verify that the structure is still valid
 657          */
 658         if (out->deleted) {
 659                 spin_unlock(&out->lock);
 660                 rcu_read_unlock();
 661                 return NULL;
 662         }
 663         return out;
 664 }
 665
 666 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
 667 {
 668         rcu_read_lock();
 669         spin_lock(&perm->lock);
 670 }
 671
 672 void ipc_unlock(struct kern_ipc_perm* perm)
 673 {
 674         spin_unlock(&perm->lock);
 675         rcu_read_unlock();
 676 }
 677
 678 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
 679 {
 680         return SEQ_MULTIPLIER*seq + id;
 681 }
 682
 683 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
 684 {
 685         if(uid/SEQ_MULTIPLIER != ipcp->seq)
 686                 return 1;
 687         return 0;
 688 }
 689
 690 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
 691
 692
 693 /**
 694  *      ipc_parse_version       -       IPC call version
 695  *      @cmd: pointer to command
 696  *
 697  *      Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
 698  *      The @cmd value is turned from an encoding command and version into
 699  *      just the command code.
 700  */
 701
 702 int ipc_parse_version (int *cmd)
 703 {
 704         if (*cmd & IPC_64) {
 705                 *cmd ^= IPC_64;
 706                 return IPC_64;
 707         } else {
 708                 return IPC_OLD;
 709         }
 710 }
 711
 712 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
 713
 714 #ifdef CONFIG_PROC_FS
 715 struct ipc_proc_iter {
 716         struct ipc_namespace *ns;
 717         struct ipc_proc_iface *iface;
 718 };
 719
 720 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
 721 {
 722         struct ipc_proc_iter *iter = s->private;
 723         struct ipc_proc_iface *iface = iter->iface;
 724         struct kern_ipc_perm *ipc = it;
 725         loff_t p;
 726         struct ipc_ids *ids;
 727
 728         ids = iter->ns->ids[iface->ids];
 729
 730         /* If we had an ipc id locked before, unlock it */
 731         if (ipc && ipc != SEQ_START_TOKEN)
 732                 ipc_unlock(ipc);
 733
 734         /*
 735          * p = *pos - 1 (because id 0 starts at position 1)
 736          *          + 1 (because we increment the position by one)
 737          */
 738         for (p = *pos; p <= ids->max_id; p++) {
 739                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 740                         *pos = p + 1;
 741                         return ipc;
 742                 }
 743         }
 744
 745         /* Out of range - return NULL to terminate iteration */
 746         return NULL;
 747 }
 748
 749 /*
 750  * File positions: pos 0 -> header, pos n -> ipc id + 1.
 751  * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
 752  */
 753 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 754 {
 755         struct ipc_proc_iter *iter = s->private;
 756         struct ipc_proc_iface *iface = iter->iface;
 757         struct kern_ipc_perm *ipc;
 758         loff_t p;
 759         struct ipc_ids *ids;
 760
 761         ids = iter->ns->ids[iface->ids];
 762
 763         /*
 764          * Take the lock - this will be released by the corresponding
 765          * call to stop().
 766          */
 767         mutex_lock(&ids->mutex);
 768
 769         /* pos < 0 is invalid */
 770         if (*pos < 0)
 771                 return NULL;
 772
 773         /* pos == 0 means header */
 774         if (*pos == 0)
 775                 return SEQ_START_TOKEN;
 776
 777         /* Find the (pos-1)th ipc */
 778         for (p = *pos - 1; p <= ids->max_id; p++) {
 779                 if ((ipc = ipc_lock(ids, p)) != NULL) {
 780                         *pos = p + 1;
 781                         return ipc;
 782                 }
 783         }
 784         return NULL;
 785 }
 786
 787 static void sysvipc_proc_stop(struct seq_file *s, void *it)
 788 {
 789         struct kern_ipc_perm *ipc = it;
 790         struct ipc_proc_iter *iter = s->private;
 791         struct ipc_proc_iface *iface = iter->iface;
 792         struct ipc_ids *ids;
 793
 794         /* If we had a locked segment, release it */
 795         if (ipc && ipc != SEQ_START_TOKEN)
 796                 ipc_unlock(ipc);
 797
 798         ids = iter->ns->ids[iface->ids];
 799         /* Release the lock we took in start() */
 800         mutex_unlock(&ids->mutex);
 801 }
 802
 803 static int sysvipc_proc_show(struct seq_file *s, void *it)
 804 {
 805         struct ipc_proc_iter *iter = s->private;
 806         struct ipc_proc_iface *iface = iter->iface;
 807
 808         if (it == SEQ_START_TOKEN)
 809                 return seq_puts(s, iface->header);
 810
 811         return iface->show(s, it);
 812 }
 813
 814 static struct seq_operations sysvipc_proc_seqops = {
 815         .start = sysvipc_proc_start,
 816         .stop  = sysvipc_proc_stop,
 817         .next  = sysvipc_proc_next,
 818         .show  = sysvipc_proc_show,
 819 };
 820
 821 static int sysvipc_proc_open(struct inode *inode, struct file *file)
 822 {
 823         int ret;
 824         struct seq_file *seq;
 825         struct ipc_proc_iter *iter;
 826
 827         ret = -ENOMEM;
 828         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 829         if (!iter)
 830                 goto out;
 831
 832         ret = seq_open(file, &sysvipc_proc_seqops);
 833         if (ret)
 834                 goto out_kfree;
 835
 836         seq = file->private_data;
 837         seq->private = iter;
 838
 839         iter->iface = PDE(inode)->data;
 840         iter->ns    = get_ipc_ns(current->nsproxy->ipc_ns);
 841 out:
 842         return ret;
 843 out_kfree:
 844         kfree(iter);
 845         goto out;
 846 }
 847
 848 static int sysvipc_proc_release(struct inode *inode, struct file *file)
 849 {
 850         struct seq_file *seq = file->private_data;
 851         struct ipc_proc_iter *iter = seq->private;
 852         put_ipc_ns(iter->ns);
 853         return seq_release_private(inode, file);
 854 }
 855
 856 static const struct file_operations sysvipc_proc_fops = {
 857         .open    = sysvipc_proc_open,
 858         .read    = seq_read,
 859         .llseek  = seq_lseek,
 860         .release = sysvipc_proc_release,
 861 };
 862 #endif /* CONFIG_PROC_FS */