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