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 #include <linux/ipc_namespace.h>
  42
  43 #include <asm/uaccess.h>
  44
  45 #include "util.h"
  46
  47 struct shm_file_data {
  48         int id;
  49         struct ipc_namespace *ns;
  50         struct file *file;
  51         const struct vm_operations_struct *vm_ops;
  52 };
  53
  54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  55
  56 static const struct file_operations shm_file_operations;
  57 static const struct vm_operations_struct shm_vm_ops;
  58
  59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  60
  61 #define shm_unlock(shp)                 \
  62         ipc_unlock(&(shp)->shm_perm)
  63
  64 static int newseg(struct ipc_namespace *, struct ipc_params *);
  65 static void shm_open(struct vm_area_struct *vma);
  66 static void shm_close(struct vm_area_struct *vma);
  67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  68 #ifdef CONFIG_PROC_FS
  69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  70 #endif
  71
  72 void shm_init_ns(struct ipc_namespace *ns)
  73 {
  74         ns->shm_ctlmax = SHMMAX;
  75         ns->shm_ctlall = SHMALL;
  76         ns->shm_ctlmni = SHMMNI;
  77         ns->shm_rmid_forced = 0;
  78         ns->shm_tot = 0;
  79         ipc_init_ids(&shm_ids(ns));
  80 }
  81
  82 /*
  83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  84  * Only shm_ids.rw_mutex remains locked on exit.
  85  */
  86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  87 {
  88         struct shmid_kernel *shp;
  89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  90
  91         if (shp->shm_nattch){
  92                 shp->shm_perm.mode |= SHM_DEST;
  93                 /* Do not find it any more */
  94                 shp->shm_perm.key = IPC_PRIVATE;
  95                 shm_unlock(shp);
  96         } else
  97                 shm_destroy(ns, shp);
  98 }
  99
 100 #ifdef CONFIG_IPC_NS
 101 void shm_exit_ns(struct ipc_namespace *ns)
 102 {
 103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 104         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 105 }
 106 #endif
 107
 108 static int __init ipc_ns_init(void)
 109 {
 110         shm_init_ns(&init_ipc_ns);
 111         return 0;
 112 }
 113
 114 pure_initcall(ipc_ns_init);
 115
 116 void __init shm_init (void)
 117 {
 118         ipc_init_proc_interface("sysvipc/shm",
 119 #if BITS_PER_LONG <= 32
 120                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 121 #else
 122                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 123 #endif
 124                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 125 }
 126
 127 /*
 128  * shm_lock_(check_) routines are called in the paths where the rw_mutex
 129  * is not necessarily held.
 130  */
 131 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 132 {
 133         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 134
 135         if (IS_ERR(ipcp))
 136                 return (struct shmid_kernel *)ipcp;
 137
 138         return container_of(ipcp, struct shmid_kernel, shm_perm);
 139 }
 140
 141 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 142 {
 143         rcu_read_lock();
 144         spin_lock(&ipcp->shm_perm.lock);
 145 }
 146
 147 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 148                                                 int id)
 149 {
 150         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 151
 152         if (IS_ERR(ipcp))
 153                 return (struct shmid_kernel *)ipcp;
 154
 155         return container_of(ipcp, struct shmid_kernel, shm_perm);
 156 }
 157
 158 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 159 {
 160         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 161 }
 162
 163
 164 /* This is called by fork, once for every shm attach. */
 165 static void shm_open(struct vm_area_struct *vma)
 166 {
 167         struct file *file = vma->vm_file;
 168         struct shm_file_data *sfd = shm_file_data(file);
 169         struct shmid_kernel *shp;
 170
 171         shp = shm_lock(sfd->ns, sfd->id);
 172         BUG_ON(IS_ERR(shp));
 173         shp->shm_atim = get_seconds();
 174         shp->shm_lprid = task_tgid_vnr(current);
 175         shp->shm_nattch++;
 176         shm_unlock(shp);
 177 }
 178
 179 /*
 180  * shm_destroy - free the struct shmid_kernel
 181  *
 182  * @ns: namespace
 183  * @shp: struct to free
 184  *
 185  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 186  * but returns with shp unlocked and freed.
 187  */
 188 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 189 {
 190         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 191         shm_rmid(ns, shp);
 192         shm_unlock(shp);
 193         if (!is_file_hugepages(shp->shm_file))
 194                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 195         else if (shp->mlock_user)
 196                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 197                                                 shp->mlock_user);
 198         fput (shp->shm_file);
 199         security_shm_free(shp);
 200         ipc_rcu_putref(shp);
 201 }
 202
 203 /*
 204  * shm_may_destroy - identifies whether shm segment should be destroyed now
 205  *
 206  * Returns true if and only if there are no active users of the segment and
 207  * one of the following is true:
 208  *
 209  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 210  *
 211  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 212  */
 213 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 214 {
 215         return (shp->shm_nattch == 0) &&
 216                (ns->shm_rmid_forced ||
 217                 (shp->shm_perm.mode & SHM_DEST));
 218 }
 219
 220 /*
 221  * remove the attach descriptor vma.
 222  * free memory for segment if it is marked destroyed.
 223  * The descriptor has already been removed from the current->mm->mmap list
 224  * and will later be kfree()d.
 225  */
 226 static void shm_close(struct vm_area_struct *vma)
 227 {
 228         struct file * file = vma->vm_file;
 229         struct shm_file_data *sfd = shm_file_data(file);
 230         struct shmid_kernel *shp;
 231         struct ipc_namespace *ns = sfd->ns;
 232
 233         down_write(&shm_ids(ns).rw_mutex);
 234         /* remove from the list of attaches of the shm segment */
 235         shp = shm_lock(ns, sfd->id);
 236         BUG_ON(IS_ERR(shp));
 237         shp->shm_lprid = task_tgid_vnr(current);
 238         shp->shm_dtim = get_seconds();
 239         shp->shm_nattch--;
 240         if (shm_may_destroy(ns, shp))
 241                 shm_destroy(ns, shp);
 242         else
 243                 shm_unlock(shp);
 244         up_write(&shm_ids(ns).rw_mutex);
 245 }
 246
 247 /* Called with ns->shm_ids(ns).rw_mutex locked */
 248 static int shm_try_destroy_current(int id, void *p, void *data)
 249 {
 250         struct ipc_namespace *ns = data;
 251         struct kern_ipc_perm *ipcp = p;
 252         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 253
 254         if (shp->shm_creator != current)
 255                 return 0;
 256
 257         /*
 258          * Mark it as orphaned to destroy the segment when
 259          * kernel.shm_rmid_forced is changed.
 260          * It is noop if the following shm_may_destroy() returns true.
 261          */
 262         shp->shm_creator = NULL;
 263
 264         /*
 265          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
 266          * is not set, it shouldn't be deleted here.
 267          */
 268         if (!ns->shm_rmid_forced)
 269                 return 0;
 270
 271         if (shm_may_destroy(ns, shp)) {
 272                 shm_lock_by_ptr(shp);
 273                 shm_destroy(ns, shp);
 274         }
 275         return 0;
 276 }
 277
 278 /* Called with ns->shm_ids(ns).rw_mutex locked */
 279 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 280 {
 281         struct ipc_namespace *ns = data;
 282         struct kern_ipc_perm *ipcp = p;
 283         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 284
 285         /*
 286          * We want to destroy segments without users and with already
 287          * exit'ed originating process.
 288          *
 289          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
 290          */
 291         if (shp->shm_creator != NULL)
 292                 return 0;
 293
 294         if (shm_may_destroy(ns, shp)) {
 295                 shm_lock_by_ptr(shp);
 296                 shm_destroy(ns, shp);
 297         }
 298         return 0;
 299 }
 300
 301 void shm_destroy_orphaned(struct ipc_namespace *ns)
 302 {
 303         down_write(&shm_ids(ns).rw_mutex);
 304         if (shm_ids(ns).in_use)
 305                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 306         up_write(&shm_ids(ns).rw_mutex);
 307 }
 308
 309
 310 void exit_shm(struct task_struct *task)
 311 {
 312         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 313
 314         if (shm_ids(ns).in_use == 0)
 315                 return;
 316
 317         /* Destroy all already created segments, but not mapped yet */
 318         down_write(&shm_ids(ns).rw_mutex);
 319         if (shm_ids(ns).in_use)
 320                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
 321         up_write(&shm_ids(ns).rw_mutex);
 322 }
 323
 324 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 325 {
 326         struct file *file = vma->vm_file;
 327         struct shm_file_data *sfd = shm_file_data(file);
 328
 329         return sfd->vm_ops->fault(vma, vmf);
 330 }
 331
 332 #ifdef CONFIG_NUMA
 333 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 334 {
 335         struct file *file = vma->vm_file;
 336         struct shm_file_data *sfd = shm_file_data(file);
 337         int err = 0;
 338         if (sfd->vm_ops->set_policy)
 339                 err = sfd->vm_ops->set_policy(vma, new);
 340         return err;
 341 }
 342
 343 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 344                                         unsigned long addr)
 345 {
 346         struct file *file = vma->vm_file;
 347         struct shm_file_data *sfd = shm_file_data(file);
 348         struct mempolicy *pol = NULL;
 349
 350         if (sfd->vm_ops->get_policy)
 351                 pol = sfd->vm_ops->get_policy(vma, addr);
 352         else if (vma->vm_policy)
 353                 pol = vma->vm_policy;
 354
 355         return pol;
 356 }
 357 #endif
 358
 359 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 360 {
 361         struct shm_file_data *sfd = shm_file_data(file);
 362         int ret;
 363
 364         ret = sfd->file->f_op->mmap(sfd->file, vma);
 365         if (ret != 0)
 366                 return ret;
 367         sfd->vm_ops = vma->vm_ops;
 368 #ifdef CONFIG_MMU
 369         BUG_ON(!sfd->vm_ops->fault);
 370 #endif
 371         vma->vm_ops = &shm_vm_ops;
 372         shm_open(vma);
 373
 374         return ret;
 375 }
 376
 377 static int shm_release(struct inode *ino, struct file *file)
 378 {
 379         struct shm_file_data *sfd = shm_file_data(file);
 380
 381         put_ipc_ns(sfd->ns);
 382         shm_file_data(file) = NULL;
 383         kfree(sfd);
 384         return 0;
 385 }
 386
 387 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 388 {
 389         struct shm_file_data *sfd = shm_file_data(file);
 390
 391         if (!sfd->file->f_op->fsync)
 392                 return -EINVAL;
 393         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 394 }
 395
 396 static unsigned long shm_get_unmapped_area(struct file *file,
 397         unsigned long addr, unsigned long len, unsigned long pgoff,
 398         unsigned long flags)
 399 {
 400         struct shm_file_data *sfd = shm_file_data(file);
 401         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 402                                                 pgoff, flags);
 403 }
 404
 405 static const struct file_operations shm_file_operations = {
 406         .mmap           = shm_mmap,
 407         .fsync          = shm_fsync,
 408         .release        = shm_release,
 409 #ifndef CONFIG_MMU
 410         .get_unmapped_area      = shm_get_unmapped_area,
 411 #endif
 412         .llseek         = noop_llseek,
 413 };
 414
 415 static const struct file_operations shm_file_operations_huge = {
 416         .mmap           = shm_mmap,
 417         .fsync          = shm_fsync,
 418         .release        = shm_release,
 419         .get_unmapped_area      = shm_get_unmapped_area,
 420         .llseek         = noop_llseek,
 421 };
 422
 423 int is_file_shm_hugepages(struct file *file)
 424 {
 425         return file->f_op == &shm_file_operations_huge;
 426 }
 427
 428 static const struct vm_operations_struct shm_vm_ops = {
 429         .open   = shm_open,     /* callback for a new vm-area open */
 430         .close  = shm_close,    /* callback for when the vm-area is released */
 431         .fault  = shm_fault,
 432 #if defined(CONFIG_NUMA)
 433         .set_policy = shm_set_policy,
 434         .get_policy = shm_get_policy,
 435 #endif
 436 };
 437
 438 /**
 439  * newseg - Create a new shared memory segment
 440  * @ns: namespace
 441  * @params: ptr to the structure that contains key, size and shmflg
 442  *
 443  * Called with shm_ids.rw_mutex held as a writer.
 444  */
 445
 446 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 447 {
 448         key_t key = params->key;
 449         int shmflg = params->flg;
 450         size_t size = params->u.size;
 451         int error;
 452         struct shmid_kernel *shp;
 453         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 454         struct file * file;
 455         char name[13];
 456         int id;
 457         vm_flags_t acctflag = 0;
 458
 459         if (size < SHMMIN || size > ns->shm_ctlmax)
 460                 return -EINVAL;
 461
 462         if (ns->shm_tot + numpages > ns->shm_ctlall)
 463                 return -ENOSPC;
 464
 465         shp = ipc_rcu_alloc(sizeof(*shp));
 466         if (!shp)
 467                 return -ENOMEM;
 468
 469         shp->shm_perm.key = key;
 470         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 471         shp->mlock_user = NULL;
 472
 473         shp->shm_perm.security = NULL;
 474         error = security_shm_alloc(shp);
 475         if (error) {
 476                 ipc_rcu_putref(shp);
 477                 return error;
 478         }
 479
 480         sprintf (name, "SYSV%08x", key);
 481         if (shmflg & SHM_HUGETLB) {
 482                 /* hugetlb_file_setup applies strict accounting */
 483                 if (shmflg & SHM_NORESERVE)
 484                         acctflag = VM_NORESERVE;
 485                 file = hugetlb_file_setup(name, 0, size, acctflag,
 486                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
 487         } else {
 488                 /*
 489                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 490                  * if it's asked for.
 491                  */
 492                 if  ((shmflg & SHM_NORESERVE) &&
 493                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 494                         acctflag = VM_NORESERVE;
 495                 file = shmem_file_setup(name, size, acctflag);
 496         }
 497         error = PTR_ERR(file);
 498         if (IS_ERR(file))
 499                 goto no_file;
 500
 501         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 502         if (id < 0) {
 503                 error = id;
 504                 goto no_id;
 505         }
 506
 507         shp->shm_cprid = task_tgid_vnr(current);
 508         shp->shm_lprid = 0;
 509         shp->shm_atim = shp->shm_dtim = 0;
 510         shp->shm_ctim = get_seconds();
 511         shp->shm_segsz = size;
 512         shp->shm_nattch = 0;
 513         shp->shm_file = file;
 514         shp->shm_creator = current;
 515         /*
 516          * shmid gets reported as "inode#" in /proc/pid/maps.
 517          * proc-ps tools use this. Changing this will break them.
 518          */
 519         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 520
 521         ns->shm_tot += numpages;
 522         error = shp->shm_perm.id;
 523         shm_unlock(shp);
 524         return error;
 525
 526 no_id:
 527         if (is_file_hugepages(file) && shp->mlock_user)
 528                 user_shm_unlock(size, shp->mlock_user);
 529         fput(file);
 530 no_file:
 531         security_shm_free(shp);
 532         ipc_rcu_putref(shp);
 533         return error;
 534 }
 535
 536 /*
 537  * Called with shm_ids.rw_mutex and ipcp locked.
 538  */
 539 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 540 {
 541         struct shmid_kernel *shp;
 542
 543         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 544         return security_shm_associate(shp, shmflg);
 545 }
 546
 547 /*
 548  * Called with shm_ids.rw_mutex and ipcp locked.
 549  */
 550 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 551                                 struct ipc_params *params)
 552 {
 553         struct shmid_kernel *shp;
 554
 555         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 556         if (shp->shm_segsz < params->u.size)
 557                 return -EINVAL;
 558
 559         return 0;
 560 }
 561
 562 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 563 {
 564         struct ipc_namespace *ns;
 565         struct ipc_ops shm_ops;
 566         struct ipc_params shm_params;
 567
 568         ns = current->nsproxy->ipc_ns;
 569
 570         shm_ops.getnew = newseg;
 571         shm_ops.associate = shm_security;
 572         shm_ops.more_checks = shm_more_checks;
 573
 574         shm_params.key = key;
 575         shm_params.flg = shmflg;
 576         shm_params.u.size = size;
 577
 578         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 579 }
 580
 581 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 582 {
 583         switch(version) {
 584         case IPC_64:
 585                 return copy_to_user(buf, in, sizeof(*in));
 586         case IPC_OLD:
 587             {
 588                 struct shmid_ds out;
 589
 590                 memset(&out, 0, sizeof(out));
 591                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 592                 out.shm_segsz   = in->shm_segsz;
 593                 out.shm_atime   = in->shm_atime;
 594                 out.shm_dtime   = in->shm_dtime;
 595                 out.shm_ctime   = in->shm_ctime;
 596                 out.shm_cpid    = in->shm_cpid;
 597                 out.shm_lpid    = in->shm_lpid;
 598                 out.shm_nattch  = in->shm_nattch;
 599
 600                 return copy_to_user(buf, &out, sizeof(out));
 601             }
 602         default:
 603                 return -EINVAL;
 604         }
 605 }
 606
 607 static inline unsigned long
 608 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 609 {
 610         switch(version) {
 611         case IPC_64:
 612                 if (copy_from_user(out, buf, sizeof(*out)))
 613                         return -EFAULT;
 614                 return 0;
 615         case IPC_OLD:
 616             {
 617                 struct shmid_ds tbuf_old;
 618
 619                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 620                         return -EFAULT;
 621
 622                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 623                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 624                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 625
 626                 return 0;
 627             }
 628         default:
 629                 return -EINVAL;
 630         }
 631 }
 632
 633 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 634 {
 635         switch(version) {
 636         case IPC_64:
 637                 return copy_to_user(buf, in, sizeof(*in));
 638         case IPC_OLD:
 639             {
 640                 struct shminfo out;
 641
 642                 if(in->shmmax > INT_MAX)
 643                         out.shmmax = INT_MAX;
 644                 else
 645                         out.shmmax = (int)in->shmmax;
 646
 647                 out.shmmin      = in->shmmin;
 648                 out.shmmni      = in->shmmni;
 649                 out.shmseg      = in->shmseg;
 650                 out.shmall      = in->shmall;
 651
 652                 return copy_to_user(buf, &out, sizeof(out));
 653             }
 654         default:
 655                 return -EINVAL;
 656         }
 657 }
 658
 659 /*
 660  * Calculate and add used RSS and swap pages of a shm.
 661  * Called with shm_ids.rw_mutex held as a reader
 662  */
 663 static void shm_add_rss_swap(struct shmid_kernel *shp,
 664         unsigned long *rss_add, unsigned long *swp_add)
 665 {
 666         struct inode *inode;
 667
 668         inode = shp->shm_file->f_path.dentry->d_inode;
 669
 670         if (is_file_hugepages(shp->shm_file)) {
 671                 struct address_space *mapping = inode->i_mapping;
 672                 struct hstate *h = hstate_file(shp->shm_file);
 673                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 674         } else {
 675 #ifdef CONFIG_SHMEM
 676                 struct shmem_inode_info *info = SHMEM_I(inode);
 677                 spin_lock(&info->lock);
 678                 *rss_add += inode->i_mapping->nrpages;
 679                 *swp_add += info->swapped;
 680                 spin_unlock(&info->lock);
 681 #else
 682                 *rss_add += inode->i_mapping->nrpages;
 683 #endif
 684         }
 685 }
 686
 687 /*
 688  * Called with shm_ids.rw_mutex held as a reader
 689  */
 690 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 691                 unsigned long *swp)
 692 {
 693         int next_id;
 694         int total, in_use;
 695
 696         *rss = 0;
 697         *swp = 0;
 698
 699         in_use = shm_ids(ns).in_use;
 700
 701         for (total = 0, next_id = 0; total < in_use; next_id++) {
 702                 struct kern_ipc_perm *ipc;
 703                 struct shmid_kernel *shp;
 704
 705                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 706                 if (ipc == NULL)
 707                         continue;
 708                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 709
 710                 shm_add_rss_swap(shp, rss, swp);
 711
 712                 total++;
 713         }
 714 }
 715
 716 /*
 717  * This function handles some shmctl commands which require the rw_mutex
 718  * to be held in write mode.
 719  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 720  */
 721 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 722                        struct shmid_ds __user *buf, int version)
 723 {
 724         struct kern_ipc_perm *ipcp;
 725         struct shmid64_ds shmid64;
 726         struct shmid_kernel *shp;
 727         int err;
 728
 729         if (cmd == IPC_SET) {
 730                 if (copy_shmid_from_user(&shmid64, buf, version))
 731                         return -EFAULT;
 732         }
 733
 734         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
 735                                &shmid64.shm_perm, 0);
 736         if (IS_ERR(ipcp))
 737                 return PTR_ERR(ipcp);
 738
 739         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 740
 741         err = security_shm_shmctl(shp, cmd);
 742         if (err)
 743                 goto out_unlock;
 744         switch (cmd) {
 745         case IPC_RMID:
 746                 do_shm_rmid(ns, ipcp);
 747                 goto out_up;
 748         case IPC_SET:
 749                 ipc_update_perm(&shmid64.shm_perm, ipcp);
 750                 shp->shm_ctim = get_seconds();
 751                 break;
 752         default:
 753                 err = -EINVAL;
 754         }
 755 out_unlock:
 756         shm_unlock(shp);
 757 out_up:
 758         up_write(&shm_ids(ns).rw_mutex);
 759         return err;
 760 }
 761
 762 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 763 {
 764         struct shmid_kernel *shp;
 765         int err, version;
 766         struct ipc_namespace *ns;
 767
 768         if (cmd < 0 || shmid < 0) {
 769                 err = -EINVAL;
 770                 goto out;
 771         }
 772
 773         version = ipc_parse_version(&cmd);
 774         ns = current->nsproxy->ipc_ns;
 775
 776         switch (cmd) { /* replace with proc interface ? */
 777         case IPC_INFO:
 778         {
 779                 struct shminfo64 shminfo;
 780
 781                 err = security_shm_shmctl(NULL, cmd);
 782                 if (err)
 783                         return err;
 784
 785                 memset(&shminfo, 0, sizeof(shminfo));
 786                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 787                 shminfo.shmmax = ns->shm_ctlmax;
 788                 shminfo.shmall = ns->shm_ctlall;
 789
 790                 shminfo.shmmin = SHMMIN;
 791                 if(copy_shminfo_to_user (buf, &shminfo, version))
 792                         return -EFAULT;
 793
 794                 down_read(&shm_ids(ns).rw_mutex);
 795                 err = ipc_get_maxid(&shm_ids(ns));
 796                 up_read(&shm_ids(ns).rw_mutex);
 797
 798                 if(err<0)
 799                         err = 0;
 800                 goto out;
 801         }
 802         case SHM_INFO:
 803         {
 804                 struct shm_info shm_info;
 805
 806                 err = security_shm_shmctl(NULL, cmd);
 807                 if (err)
 808                         return err;
 809
 810                 memset(&shm_info, 0, sizeof(shm_info));
 811                 down_read(&shm_ids(ns).rw_mutex);
 812                 shm_info.used_ids = shm_ids(ns).in_use;
 813                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 814                 shm_info.shm_tot = ns->shm_tot;
 815                 shm_info.swap_attempts = 0;
 816                 shm_info.swap_successes = 0;
 817                 err = ipc_get_maxid(&shm_ids(ns));
 818                 up_read(&shm_ids(ns).rw_mutex);
 819                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 820                         err = -EFAULT;
 821                         goto out;
 822                 }
 823
 824                 err = err < 0 ? 0 : err;
 825                 goto out;
 826         }
 827         case SHM_STAT:
 828         case IPC_STAT:
 829         {
 830                 struct shmid64_ds tbuf;
 831                 int result;
 832
 833                 if (cmd == SHM_STAT) {
 834                         shp = shm_lock(ns, shmid);
 835                         if (IS_ERR(shp)) {
 836                                 err = PTR_ERR(shp);
 837                                 goto out;
 838                         }
 839                         result = shp->shm_perm.id;
 840                 } else {
 841                         shp = shm_lock_check(ns, shmid);
 842                         if (IS_ERR(shp)) {
 843                                 err = PTR_ERR(shp);
 844                                 goto out;
 845                         }
 846                         result = 0;
 847                 }
 848                 err = -EACCES;
 849                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 850                         goto out_unlock;
 851                 err = security_shm_shmctl(shp, cmd);
 852                 if (err)
 853                         goto out_unlock;
 854                 memset(&tbuf, 0, sizeof(tbuf));
 855                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 856                 tbuf.shm_segsz  = shp->shm_segsz;
 857                 tbuf.shm_atime  = shp->shm_atim;
 858                 tbuf.shm_dtime  = shp->shm_dtim;
 859                 tbuf.shm_ctime  = shp->shm_ctim;
 860                 tbuf.shm_cpid   = shp->shm_cprid;
 861                 tbuf.shm_lpid   = shp->shm_lprid;
 862                 tbuf.shm_nattch = shp->shm_nattch;
 863                 shm_unlock(shp);
 864                 if(copy_shmid_to_user (buf, &tbuf, version))
 865                         err = -EFAULT;
 866                 else
 867                         err = result;
 868                 goto out;
 869         }
 870         case SHM_LOCK:
 871         case SHM_UNLOCK:
 872         {
 873                 struct file *shm_file;
 874
 875                 shp = shm_lock_check(ns, shmid);
 876                 if (IS_ERR(shp)) {
 877                         err = PTR_ERR(shp);
 878                         goto out;
 879                 }
 880
 881                 audit_ipc_obj(&(shp->shm_perm));
 882
 883                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
 884                         uid_t euid = current_euid();
 885                         err = -EPERM;
 886                         if (euid != shp->shm_perm.uid &&
 887                             euid != shp->shm_perm.cuid)
 888                                 goto out_unlock;
 889                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
 890                                 goto out_unlock;
 891                 }
 892
 893                 err = security_shm_shmctl(shp, cmd);
 894                 if (err)
 895                         goto out_unlock;
 896
 897                 shm_file = shp->shm_file;
 898                 if (is_file_hugepages(shm_file))
 899                         goto out_unlock;
 900
 901                 if (cmd == SHM_LOCK) {
 902                         struct user_struct *user = current_user();
 903                         err = shmem_lock(shm_file, 1, user);
 904                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
 905                                 shp->shm_perm.mode |= SHM_LOCKED;
 906                                 shp->mlock_user = user;
 907                         }
 908                         goto out_unlock;
 909                 }
 910
 911                 /* SHM_UNLOCK */
 912                 if (!(shp->shm_perm.mode & SHM_LOCKED))
 913                         goto out_unlock;
 914                 shmem_lock(shm_file, 0, shp->mlock_user);
 915                 shp->shm_perm.mode &= ~SHM_LOCKED;
 916                 shp->mlock_user = NULL;
 917                 get_file(shm_file);
 918                 shm_unlock(shp);
 919                 shmem_unlock_mapping(shm_file->f_mapping);
 920                 fput(shm_file);
 921                 goto out;
 922         }
 923         case IPC_RMID:
 924         case IPC_SET:
 925                 err = shmctl_down(ns, shmid, cmd, buf, version);
 926                 return err;
 927         default:
 928                 return -EINVAL;
 929         }
 930
 931 out_unlock:
 932         shm_unlock(shp);
 933 out:
 934         return err;
 935 }
 936
 937 /*
 938  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 939  *
 940  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 941  * "raddr" thing points to kernel space, and there has to be a wrapper around
 942  * this.
 943  */
 944 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 945 {
 946         struct shmid_kernel *shp;
 947         unsigned long addr;
 948         unsigned long size;
 949         struct file * file;
 950         int    err;
 951         unsigned long flags;
 952         unsigned long prot;
 953         int acc_mode;
 954         unsigned long user_addr;
 955         struct ipc_namespace *ns;
 956         struct shm_file_data *sfd;
 957         struct path path;
 958         fmode_t f_mode;
 959
 960         err = -EINVAL;
 961         if (shmid < 0)
 962                 goto out;
 963         else if ((addr = (ulong)shmaddr)) {
 964                 if (addr & (SHMLBA-1)) {
 965                         if (shmflg & SHM_RND)
 966                                 addr &= ~(SHMLBA-1);       /* round down */
 967                         else
 968 #ifndef __ARCH_FORCE_SHMLBA
 969                                 if (addr & ~PAGE_MASK)
 970 #endif
 971                                         goto out;
 972                 }
 973                 flags = MAP_SHARED | MAP_FIXED;
 974         } else {
 975                 if ((shmflg & SHM_REMAP))
 976                         goto out;
 977
 978                 flags = MAP_SHARED;
 979         }
 980
 981         if (shmflg & SHM_RDONLY) {
 982                 prot = PROT_READ;
 983                 acc_mode = S_IRUGO;
 984                 f_mode = FMODE_READ;
 985         } else {
 986                 prot = PROT_READ | PROT_WRITE;
 987                 acc_mode = S_IRUGO | S_IWUGO;
 988                 f_mode = FMODE_READ | FMODE_WRITE;
 989         }
 990         if (shmflg & SHM_EXEC) {
 991                 prot |= PROT_EXEC;
 992                 acc_mode |= S_IXUGO;
 993         }
 994
 995         /*
 996          * We cannot rely on the fs check since SYSV IPC does have an
 997          * additional creator id...
 998          */
 999         ns = current->nsproxy->ipc_ns;
1000         shp = shm_lock_check(ns, shmid);
1001         if (IS_ERR(shp)) {
1002                 err = PTR_ERR(shp);
1003                 goto out;
1004         }
1005
1006         err = -EACCES;
1007         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1008                 goto out_unlock;
1009
1010         err = security_shm_shmat(shp, shmaddr, shmflg);
1011         if (err)
1012                 goto out_unlock;
1013
1014         path = shp->shm_file->f_path;
1015         path_get(&path);
1016         shp->shm_nattch++;
1017         size = i_size_read(path.dentry->d_inode);
1018         shm_unlock(shp);
1019
1020         err = -ENOMEM;
1021         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1022         if (!sfd)
1023                 goto out_put_dentry;
1024
1025         file = alloc_file(&path, f_mode,
1026                           is_file_hugepages(shp->shm_file) ?
1027                                 &shm_file_operations_huge :
1028                                 &shm_file_operations);
1029         if (!file)
1030                 goto out_free;
1031
1032         file->private_data = sfd;
1033         file->f_mapping = shp->shm_file->f_mapping;
1034         sfd->id = shp->shm_perm.id;
1035         sfd->ns = get_ipc_ns(ns);
1036         sfd->file = shp->shm_file;
1037         sfd->vm_ops = NULL;
1038
1039         err = security_mmap_file(file, prot, flags);
1040         if (err)
1041                 goto out_fput;
1042
1043         down_write(&current->mm->mmap_sem);
1044         if (addr && !(shmflg & SHM_REMAP)) {
1045                 err = -EINVAL;
1046                 if (find_vma_intersection(current->mm, addr, addr + size))
1047                         goto invalid;
1048                 /*
1049                  * If shm segment goes below stack, make sure there is some
1050                  * space left for the stack to grow (at least 4 pages).
1051                  */
1052                 if (addr < current->mm->start_stack &&
1053                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1054                         goto invalid;
1055         }
1056
1057         user_addr = do_mmap_pgoff(file, addr, size, prot, flags, 0);
1058         *raddr = user_addr;
1059         err = 0;
1060         if (IS_ERR_VALUE(user_addr))
1061                 err = (long)user_addr;
1062 invalid:
1063         up_write(&current->mm->mmap_sem);
1064
1065 out_fput:
1066         fput(file);
1067
1068 out_nattch:
1069         down_write(&shm_ids(ns).rw_mutex);
1070         shp = shm_lock(ns, shmid);
1071         BUG_ON(IS_ERR(shp));
1072         shp->shm_nattch--;
1073         if (shm_may_destroy(ns, shp))
1074                 shm_destroy(ns, shp);
1075         else
1076                 shm_unlock(shp);
1077         up_write(&shm_ids(ns).rw_mutex);
1078
1079 out:
1080         return err;
1081
1082 out_unlock:
1083         shm_unlock(shp);
1084         goto out;
1085
1086 out_free:
1087         kfree(sfd);
1088 out_put_dentry:
1089         path_put(&path);
1090         goto out_nattch;
1091 }
1092
1093 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1094 {
1095         unsigned long ret;
1096         long err;
1097
1098         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1099         if (err)
1100                 return err;
1101         force_successful_syscall_return();
1102         return (long)ret;
1103 }
1104
1105 /*
1106  * detach and kill segment if marked destroyed.
1107  * The work is done in shm_close.
1108  */
1109 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1110 {
1111         struct mm_struct *mm = current->mm;
1112         struct vm_area_struct *vma;
1113         unsigned long addr = (unsigned long)shmaddr;
1114         int retval = -EINVAL;
1115 #ifdef CONFIG_MMU
1116         loff_t size = 0;
1117         struct vm_area_struct *next;
1118 #endif
1119
1120         if (addr & ~PAGE_MASK)
1121                 return retval;
1122
1123         down_write(&mm->mmap_sem);
1124
1125         /*
1126          * This function tries to be smart and unmap shm segments that
1127          * were modified by partial mlock or munmap calls:
1128          * - It first determines the size of the shm segment that should be
1129          *   unmapped: It searches for a vma that is backed by shm and that
1130          *   started at address shmaddr. It records it's size and then unmaps
1131          *   it.
1132          * - Then it unmaps all shm vmas that started at shmaddr and that
1133          *   are within the initially determined size.
1134          * Errors from do_munmap are ignored: the function only fails if
1135          * it's called with invalid parameters or if it's called to unmap
1136          * a part of a vma. Both calls in this function are for full vmas,
1137          * the parameters are directly copied from the vma itself and always
1138          * valid - therefore do_munmap cannot fail. (famous last words?)
1139          */
1140         /*
1141          * If it had been mremap()'d, the starting address would not
1142          * match the usual checks anyway. So assume all vma's are
1143          * above the starting address given.
1144          */
1145         vma = find_vma(mm, addr);
1146
1147 #ifdef CONFIG_MMU
1148         while (vma) {
1149                 next = vma->vm_next;
1150
1151                 /*
1152                  * Check if the starting address would match, i.e. it's
1153                  * a fragment created by mprotect() and/or munmap(), or it
1154                  * otherwise it starts at this address with no hassles.
1155                  */
1156                 if ((vma->vm_ops == &shm_vm_ops) &&
1157                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1158
1159
1160                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1161                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1162                         /*
1163                          * We discovered the size of the shm segment, so
1164                          * break out of here and fall through to the next
1165                          * loop that uses the size information to stop
1166                          * searching for matching vma's.
1167                          */
1168                         retval = 0;
1169                         vma = next;
1170                         break;
1171                 }
1172                 vma = next;
1173         }
1174
1175         /*
1176          * We need look no further than the maximum address a fragment
1177          * could possibly have landed at. Also cast things to loff_t to
1178          * prevent overflows and make comparisons vs. equal-width types.
1179          */
1180         size = PAGE_ALIGN(size);
1181         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1182                 next = vma->vm_next;
1183
1184                 /* finding a matching vma now does not alter retval */
1185                 if ((vma->vm_ops == &shm_vm_ops) &&
1186                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1187
1188                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1189                 vma = next;
1190         }
1191
1192 #else /* CONFIG_MMU */
1193         /* under NOMMU conditions, the exact address to be destroyed must be
1194          * given */
1195         retval = -EINVAL;
1196         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1197                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1198                 retval = 0;
1199         }
1200
1201 #endif
1202
1203         up_write(&mm->mmap_sem);
1204         return retval;
1205 }
1206
1207 #ifdef CONFIG_PROC_FS
1208 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1209 {
1210         struct shmid_kernel *shp = it;
1211         unsigned long rss = 0, swp = 0;
1212
1213         shm_add_rss_swap(shp, &rss, &swp);
1214
1215 #if BITS_PER_LONG <= 32
1216 #define SIZE_SPEC "%10lu"
1217 #else
1218 #define SIZE_SPEC "%21lu"
1219 #endif
1220
1221         return seq_printf(s,
1222                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1223                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1224                           SIZE_SPEC " " SIZE_SPEC "\n",
1225                           shp->shm_perm.key,
1226                           shp->shm_perm.id,
1227                           shp->shm_perm.mode,
1228                           shp->shm_segsz,
1229                           shp->shm_cprid,
1230                           shp->shm_lprid,
1231                           shp->shm_nattch,
1232                           shp->shm_perm.uid,
1233                           shp->shm_perm.gid,
1234                           shp->shm_perm.cuid,
1235                           shp->shm_perm.cgid,
1236                           shp->shm_atim,
1237                           shp->shm_dtim,
1238                           shp->shm_ctim,
1239                           rss * PAGE_SIZE,
1240                           swp * PAGE_SIZE);
1241 }
1242 #endif