ipc/shm.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/ipc/shm.c
   4  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   5  *       Many improvements/fixes by Bruno Haible.
   6  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   7  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   8  *
   9  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  10  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  11  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  12  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  13  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  14  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  15  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  16  *
  17  * support for audit of ipc object properties and permission changes
  18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19  *
  20  * namespaces support
  21  * OpenVZ, SWsoft Inc.
  22  * Pavel Emelianov <xemul@openvz.org>
  23  *
  24  * Better ipc lock (kern_ipc_perm.lock) handling
  25  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
  26  */
  27
  28 #include <linux/slab.h>
  29 #include <linux/mm.h>
  30 #include <linux/hugetlb.h>
  31 #include <linux/shm.h>
  32 #include <linux/init.h>
  33 #include <linux/file.h>
  34 #include <linux/mman.h>
  35 #include <linux/shmem_fs.h>
  36 #include <linux/security.h>
  37 #include <linux/syscalls.h>
  38 #include <linux/audit.h>
  39 #include <linux/capability.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/seq_file.h>
  42 #include <linux/rwsem.h>
  43 #include <linux/nsproxy.h>
  44 #include <linux/mount.h>
  45 #include <linux/ipc_namespace.h>
  46
  47 #include <linux/uaccess.h>
  48
  49 #include "util.h"
  50
  51 struct shmid_kernel /* private to the kernel */
  52 {
  53         struct kern_ipc_perm    shm_perm;
  54         struct file             *shm_file;
  55         unsigned long           shm_nattch;
  56         unsigned long           shm_segsz;
  57         time64_t                shm_atim;
  58         time64_t                shm_dtim;
  59         time64_t                shm_ctim;
  60         struct pid              *shm_cprid;
  61         struct pid              *shm_lprid;
  62         struct user_struct      *mlock_user;
  63
  64         /* The task created the shm object.  NULL if the task is dead. */
  65         struct task_struct      *shm_creator;
  66         struct list_head        shm_clist;      /* list by creator */
  67 } __randomize_layout;
  68
  69 /* shm_mode upper byte flags */
  70 #define SHM_DEST        01000   /* segment will be destroyed on last detach */
  71 #define SHM_LOCKED      02000   /* segment will not be swapped */
  72
  73 struct shm_file_data {
  74         int id;
  75         struct ipc_namespace *ns;
  76         struct file *file;
  77         const struct vm_operations_struct *vm_ops;
  78 };
  79
  80 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  81
  82 static const struct file_operations shm_file_operations;
  83 static const struct vm_operations_struct shm_vm_ops;
  84
  85 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  86
  87 #define shm_unlock(shp)                 \
  88         ipc_unlock(&(shp)->shm_perm)
  89
  90 static int newseg(struct ipc_namespace *, struct ipc_params *);
  91 static void shm_open(struct vm_area_struct *vma);
  92 static void shm_close(struct vm_area_struct *vma);
  93 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
  94 #ifdef CONFIG_PROC_FS
  95 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  96 #endif
  97
  98 int shm_init_ns(struct ipc_namespace *ns)
  99 {
 100         ns->shm_ctlmax = SHMMAX;
 101         ns->shm_ctlall = SHMALL;
 102         ns->shm_ctlmni = SHMMNI;
 103         ns->shm_rmid_forced = 0;
 104         ns->shm_tot = 0;
 105         return ipc_init_ids(&shm_ids(ns));
 106 }
 107
 108 /*
 109  * Called with shm_ids.rwsem (writer) and the shp structure locked.
 110  * Only shm_ids.rwsem remains locked on exit.
 111  */
 112 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 113 {
 114         struct shmid_kernel *shp;
 115
 116         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 117
 118         if (shp->shm_nattch) {
 119                 shp->shm_perm.mode |= SHM_DEST;
 120                 /* Do not find it any more */
 121                 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
 122                 shm_unlock(shp);
 123         } else
 124                 shm_destroy(ns, shp);
 125 }
 126
 127 #ifdef CONFIG_IPC_NS
 128 void shm_exit_ns(struct ipc_namespace *ns)
 129 {
 130         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 131         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 132         rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
 133 }
 134 #endif
 135
 136 static int __init ipc_ns_init(void)
 137 {
 138         const int err = shm_init_ns(&init_ipc_ns);
 139         WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
 140         return err;
 141 }
 142
 143 pure_initcall(ipc_ns_init);
 144
 145 void __init shm_init(void)
 146 {
 147         ipc_init_proc_interface("sysvipc/shm",
 148 #if BITS_PER_LONG <= 32
 149                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 150 #else
 151                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 152 #endif
 153                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
 154 }
 155
 156 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
 157 {
 158         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
 159
 160         if (IS_ERR(ipcp))
 161                 return ERR_CAST(ipcp);
 162
 163         return container_of(ipcp, struct shmid_kernel, shm_perm);
 164 }
 165
 166 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
 167 {
 168         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
 169
 170         if (IS_ERR(ipcp))
 171                 return ERR_CAST(ipcp);
 172
 173         return container_of(ipcp, struct shmid_kernel, shm_perm);
 174 }
 175
 176 /*
 177  * shm_lock_(check_) routines are called in the paths where the rwsem
 178  * is not necessarily held.
 179  */
 180 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 181 {
 182         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 183
 184         /*
 185          * Callers of shm_lock() must validate the status of the returned ipc
 186          * object pointer (as returned by ipc_lock()), and error out as
 187          * appropriate.
 188          */
 189         if (IS_ERR(ipcp))
 190                 return (void *)ipcp;
 191         return container_of(ipcp, struct shmid_kernel, shm_perm);
 192 }
 193
 194 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 195 {
 196         rcu_read_lock();
 197         ipc_lock_object(&ipcp->shm_perm);
 198 }
 199
 200 static void shm_rcu_free(struct rcu_head *head)
 201 {
 202         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
 203                                                         rcu);
 204         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
 205                                                         shm_perm);
 206         security_shm_free(&shp->shm_perm);
 207         kvfree(shp);
 208 }
 209
 210 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 211 {
 212         list_del(&s->shm_clist);
 213         ipc_rmid(&shm_ids(ns), &s->shm_perm);
 214 }
 215
 216
 217 static int __shm_open(struct vm_area_struct *vma)
 218 {
 219         struct file *file = vma->vm_file;
 220         struct shm_file_data *sfd = shm_file_data(file);
 221         struct shmid_kernel *shp;
 222
 223         shp = shm_lock(sfd->ns, sfd->id);
 224
 225         if (IS_ERR(shp))
 226                 return PTR_ERR(shp);
 227
 228         shp->shm_atim = ktime_get_real_seconds();
 229         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 230         shp->shm_nattch++;
 231         shm_unlock(shp);
 232         return 0;
 233 }
 234
 235 /* This is called by fork, once for every shm attach. */
 236 static void shm_open(struct vm_area_struct *vma)
 237 {
 238         int err = __shm_open(vma);
 239         /*
 240          * We raced in the idr lookup or with shm_destroy().
 241          * Either way, the ID is busted.
 242          */
 243         WARN_ON_ONCE(err);
 244 }
 245
 246 /*
 247  * shm_destroy - free the struct shmid_kernel
 248  *
 249  * @ns: namespace
 250  * @shp: struct to free
 251  *
 252  * It has to be called with shp and shm_ids.rwsem (writer) locked,
 253  * but returns with shp unlocked and freed.
 254  */
 255 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 256 {
 257         struct file *shm_file;
 258
 259         shm_file = shp->shm_file;
 260         shp->shm_file = NULL;
 261         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 262         shm_rmid(ns, shp);
 263         shm_unlock(shp);
 264         if (!is_file_hugepages(shm_file))
 265                 shmem_lock(shm_file, 0, shp->mlock_user);
 266         else if (shp->mlock_user)
 267                 user_shm_unlock(i_size_read(file_inode(shm_file)),
 268                                 shp->mlock_user);
 269         fput(shm_file);
 270         ipc_update_pid(&shp->shm_cprid, NULL);
 271         ipc_update_pid(&shp->shm_lprid, NULL);
 272         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
 273 }
 274
 275 /*
 276  * shm_may_destroy - identifies whether shm segment should be destroyed now
 277  *
 278  * Returns true if and only if there are no active users of the segment and
 279  * one of the following is true:
 280  *
 281  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 282  *
 283  * 2) sysctl kernel.shm_rmid_forced is set to 1.
 284  */
 285 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 286 {
 287         return (shp->shm_nattch == 0) &&
 288                (ns->shm_rmid_forced ||
 289                 (shp->shm_perm.mode & SHM_DEST));
 290 }
 291
 292 /*
 293  * remove the attach descriptor vma.
 294  * free memory for segment if it is marked destroyed.
 295  * The descriptor has already been removed from the current->mm->mmap list
 296  * and will later be kfree()d.
 297  */
 298 static void shm_close(struct vm_area_struct *vma)
 299 {
 300         struct file *file = vma->vm_file;
 301         struct shm_file_data *sfd = shm_file_data(file);
 302         struct shmid_kernel *shp;
 303         struct ipc_namespace *ns = sfd->ns;
 304
 305         down_write(&shm_ids(ns).rwsem);
 306         /* remove from the list of attaches of the shm segment */
 307         shp = shm_lock(ns, sfd->id);
 308
 309         /*
 310          * We raced in the idr lookup or with shm_destroy().
 311          * Either way, the ID is busted.
 312          */
 313         if (WARN_ON_ONCE(IS_ERR(shp)))
 314                 goto done; /* no-op */
 315
 316         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
 317         shp->shm_dtim = ktime_get_real_seconds();
 318         shp->shm_nattch--;
 319         if (shm_may_destroy(ns, shp))
 320                 shm_destroy(ns, shp);
 321         else
 322                 shm_unlock(shp);
 323 done:
 324         up_write(&shm_ids(ns).rwsem);
 325 }
 326
 327 /* Called with ns->shm_ids(ns).rwsem locked */
 328 static int shm_try_destroy_orphaned(int id, void *p, void *data)
 329 {
 330         struct ipc_namespace *ns = data;
 331         struct kern_ipc_perm *ipcp = p;
 332         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 333
 334         /*
 335          * We want to destroy segments without users and with already
 336          * exit'ed originating process.
 337          *
 338          * As shp->* are changed under rwsem, it's safe to skip shp locking.
 339          */
 340         if (shp->shm_creator != NULL)
 341                 return 0;
 342
 343         if (shm_may_destroy(ns, shp)) {
 344                 shm_lock_by_ptr(shp);
 345                 shm_destroy(ns, shp);
 346         }
 347         return 0;
 348 }
 349
 350 void shm_destroy_orphaned(struct ipc_namespace *ns)
 351 {
 352         down_write(&shm_ids(ns).rwsem);
 353         if (shm_ids(ns).in_use)
 354                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 355         up_write(&shm_ids(ns).rwsem);
 356 }
 357
 358 /* Locking assumes this will only be called with task == current */
 359 void exit_shm(struct task_struct *task)
 360 {
 361         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 362         struct shmid_kernel *shp, *n;
 363
 364         if (list_empty(&task->sysvshm.shm_clist))
 365                 return;
 366
 367         /*
 368          * If kernel.shm_rmid_forced is not set then only keep track of
 369          * which shmids are orphaned, so that a later set of the sysctl
 370          * can clean them up.
 371          */
 372         if (!ns->shm_rmid_forced) {
 373                 down_read(&shm_ids(ns).rwsem);
 374                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
 375                         shp->shm_creator = NULL;
 376                 /*
 377                  * Only under read lock but we are only called on current
 378                  * so no entry on the list will be shared.
 379                  */
 380                 list_del(&task->sysvshm.shm_clist);
 381                 up_read(&shm_ids(ns).rwsem);
 382                 return;
 383         }
 384
 385         /*
 386          * Destroy all already created segments, that were not yet mapped,
 387          * and mark any mapped as orphan to cover the sysctl toggling.
 388          * Destroy is skipped if shm_may_destroy() returns false.
 389          */
 390         down_write(&shm_ids(ns).rwsem);
 391         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
 392                 shp->shm_creator = NULL;
 393
 394                 if (shm_may_destroy(ns, shp)) {
 395                         shm_lock_by_ptr(shp);
 396                         shm_destroy(ns, shp);
 397                 }
 398         }
 399
 400         /* Remove the list head from any segments still attached. */
 401         list_del(&task->sysvshm.shm_clist);
 402         up_write(&shm_ids(ns).rwsem);
 403 }
 404
 405 static int shm_fault(struct vm_fault *vmf)
 406 {
 407         struct file *file = vmf->vma->vm_file;
 408         struct shm_file_data *sfd = shm_file_data(file);
 409
 410         return sfd->vm_ops->fault(vmf);
 411 }
 412
 413 #ifdef CONFIG_NUMA
 414 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 415 {
 416         struct file *file = vma->vm_file;
 417         struct shm_file_data *sfd = shm_file_data(file);
 418         int err = 0;
 419
 420         if (sfd->vm_ops->set_policy)
 421                 err = sfd->vm_ops->set_policy(vma, new);
 422         return err;
 423 }
 424
 425 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 426                                         unsigned long addr)
 427 {
 428         struct file *file = vma->vm_file;
 429         struct shm_file_data *sfd = shm_file_data(file);
 430         struct mempolicy *pol = NULL;
 431
 432         if (sfd->vm_ops->get_policy)
 433                 pol = sfd->vm_ops->get_policy(vma, addr);
 434         else if (vma->vm_policy)
 435                 pol = vma->vm_policy;
 436
 437         return pol;
 438 }
 439 #endif
 440
 441 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
 442 {
 443         struct shm_file_data *sfd = shm_file_data(file);
 444         int ret;
 445
 446         /*
 447          * In case of remap_file_pages() emulation, the file can represent
 448          * removed IPC ID: propogate shm_lock() error to caller.
 449          */
 450         ret = __shm_open(vma);
 451         if (ret)
 452                 return ret;
 453
 454         ret = call_mmap(sfd->file, vma);
 455         if (ret) {
 456                 shm_close(vma);
 457                 return ret;
 458         }
 459         sfd->vm_ops = vma->vm_ops;
 460 #ifdef CONFIG_MMU
 461         WARN_ON(!sfd->vm_ops->fault);
 462 #endif
 463         vma->vm_ops = &shm_vm_ops;
 464         return 0;
 465 }
 466
 467 static int shm_release(struct inode *ino, struct file *file)
 468 {
 469         struct shm_file_data *sfd = shm_file_data(file);
 470
 471         put_ipc_ns(sfd->ns);
 472         shm_file_data(file) = NULL;
 473         kfree(sfd);
 474         return 0;
 475 }
 476
 477 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 478 {
 479         struct shm_file_data *sfd = shm_file_data(file);
 480
 481         if (!sfd->file->f_op->fsync)
 482                 return -EINVAL;
 483         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 484 }
 485
 486 static long shm_fallocate(struct file *file, int mode, loff_t offset,
 487                           loff_t len)
 488 {
 489         struct shm_file_data *sfd = shm_file_data(file);
 490
 491         if (!sfd->file->f_op->fallocate)
 492                 return -EOPNOTSUPP;
 493         return sfd->file->f_op->fallocate(file, mode, offset, len);
 494 }
 495
 496 static unsigned long shm_get_unmapped_area(struct file *file,
 497         unsigned long addr, unsigned long len, unsigned long pgoff,
 498         unsigned long flags)
 499 {
 500         struct shm_file_data *sfd = shm_file_data(file);
 501
 502         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 503                                                 pgoff, flags);
 504 }
 505
 506 static const struct file_operations shm_file_operations = {
 507         .mmap           = shm_mmap,
 508         .fsync          = shm_fsync,
 509         .release        = shm_release,
 510         .get_unmapped_area      = shm_get_unmapped_area,
 511         .llseek         = noop_llseek,
 512         .fallocate      = shm_fallocate,
 513 };
 514
 515 /*
 516  * shm_file_operations_huge is now identical to shm_file_operations,
 517  * but we keep it distinct for the sake of is_file_shm_hugepages().
 518  */
 519 static const struct file_operations shm_file_operations_huge = {
 520         .mmap           = shm_mmap,
 521         .fsync          = shm_fsync,
 522         .release        = shm_release,
 523         .get_unmapped_area      = shm_get_unmapped_area,
 524         .llseek         = noop_llseek,
 525         .fallocate      = shm_fallocate,
 526 };
 527
 528 bool is_file_shm_hugepages(struct file *file)
 529 {
 530         return file->f_op == &shm_file_operations_huge;
 531 }
 532
 533 static const struct vm_operations_struct shm_vm_ops = {
 534         .open   = shm_open,     /* callback for a new vm-area open */
 535         .close  = shm_close,    /* callback for when the vm-area is released */
 536         .fault  = shm_fault,
 537 #if defined(CONFIG_NUMA)
 538         .set_policy = shm_set_policy,
 539         .get_policy = shm_get_policy,
 540 #endif
 541 };
 542
 543 /**
 544  * newseg - Create a new shared memory segment
 545  * @ns: namespace
 546  * @params: ptr to the structure that contains key, size and shmflg
 547  *
 548  * Called with shm_ids.rwsem held as a writer.
 549  */
 550 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 551 {
 552         key_t key = params->key;
 553         int shmflg = params->flg;
 554         size_t size = params->u.size;
 555         int error;
 556         struct shmid_kernel *shp;
 557         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 558         struct file *file;
 559         char name[13];
 560         vm_flags_t acctflag = 0;
 561
 562         if (size < SHMMIN || size > ns->shm_ctlmax)
 563                 return -EINVAL;
 564
 565         if (numpages << PAGE_SHIFT < size)
 566                 return -ENOSPC;
 567
 568         if (ns->shm_tot + numpages < ns->shm_tot ||
 569                         ns->shm_tot + numpages > ns->shm_ctlall)
 570                 return -ENOSPC;
 571
 572         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
 573         if (unlikely(!shp))
 574                 return -ENOMEM;
 575
 576         shp->shm_perm.key = key;
 577         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 578         shp->mlock_user = NULL;
 579
 580         shp->shm_perm.security = NULL;
 581         error = security_shm_alloc(&shp->shm_perm);
 582         if (error) {
 583                 kvfree(shp);
 584                 return error;
 585         }
 586
 587         sprintf(name, "SYSV%08x", key);
 588         if (shmflg & SHM_HUGETLB) {
 589                 struct hstate *hs;
 590                 size_t hugesize;
 591
 592                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 593                 if (!hs) {
 594                         error = -EINVAL;
 595                         goto no_file;
 596                 }
 597                 hugesize = ALIGN(size, huge_page_size(hs));
 598
 599                 /* hugetlb_file_setup applies strict accounting */
 600                 if (shmflg & SHM_NORESERVE)
 601                         acctflag = VM_NORESERVE;
 602                 file = hugetlb_file_setup(name, hugesize, acctflag,
 603                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
 604                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 605         } else {
 606                 /*
 607                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 608                  * if it's asked for.
 609                  */
 610                 if  ((shmflg & SHM_NORESERVE) &&
 611                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 612                         acctflag = VM_NORESERVE;
 613                 file = shmem_kernel_file_setup(name, size, acctflag);
 614         }
 615         error = PTR_ERR(file);
 616         if (IS_ERR(file))
 617                 goto no_file;
 618
 619         shp->shm_cprid = get_pid(task_tgid(current));
 620         shp->shm_lprid = NULL;
 621         shp->shm_atim = shp->shm_dtim = 0;
 622         shp->shm_ctim = ktime_get_real_seconds();
 623         shp->shm_segsz = size;
 624         shp->shm_nattch = 0;
 625         shp->shm_file = file;
 626         shp->shm_creator = current;
 627
 628         /* ipc_addid() locks shp upon success. */
 629         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 630         if (error < 0)
 631                 goto no_id;
 632
 633         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
 634
 635         /*
 636          * shmid gets reported as "inode#" in /proc/pid/maps.
 637          * proc-ps tools use this. Changing this will break them.
 638          */
 639         file_inode(file)->i_ino = shp->shm_perm.id;
 640
 641         ns->shm_tot += numpages;
 642         error = shp->shm_perm.id;
 643
 644         ipc_unlock_object(&shp->shm_perm);
 645         rcu_read_unlock();
 646         return error;
 647
 648 no_id:
 649         if (is_file_hugepages(file) && shp->mlock_user)
 650                 user_shm_unlock(size, shp->mlock_user);
 651         fput(file);
 652 no_file:
 653         ipc_update_pid(&shp->shm_cprid, NULL);
 654         ipc_update_pid(&shp->shm_lprid, NULL);
 655         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
 656         return error;
 657 }
 658
 659 /*
 660  * Called with shm_ids.rwsem and ipcp locked.
 661  */
 662 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 663                                 struct ipc_params *params)
 664 {
 665         struct shmid_kernel *shp;
 666
 667         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 668         if (shp->shm_segsz < params->u.size)
 669                 return -EINVAL;
 670
 671         return 0;
 672 }
 673
 674 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 675 {
 676         struct ipc_namespace *ns;
 677         static const struct ipc_ops shm_ops = {
 678                 .getnew = newseg,
 679                 .associate = security_shm_associate,
 680                 .more_checks = shm_more_checks,
 681         };
 682         struct ipc_params shm_params;
 683
 684         ns = current->nsproxy->ipc_ns;
 685
 686         shm_params.key = key;
 687         shm_params.flg = shmflg;
 688         shm_params.u.size = size;
 689
 690         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 691 }
 692
 693 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 694 {
 695         switch (version) {
 696         case IPC_64:
 697                 return copy_to_user(buf, in, sizeof(*in));
 698         case IPC_OLD:
 699             {
 700                 struct shmid_ds out;
 701
 702                 memset(&out, 0, sizeof(out));
 703                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 704                 out.shm_segsz   = in->shm_segsz;
 705                 out.shm_atime   = in->shm_atime;
 706                 out.shm_dtime   = in->shm_dtime;
 707                 out.shm_ctime   = in->shm_ctime;
 708                 out.shm_cpid    = in->shm_cpid;
 709                 out.shm_lpid    = in->shm_lpid;
 710                 out.shm_nattch  = in->shm_nattch;
 711
 712                 return copy_to_user(buf, &out, sizeof(out));
 713             }
 714         default:
 715                 return -EINVAL;
 716         }
 717 }
 718
 719 static inline unsigned long
 720 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 721 {
 722         switch (version) {
 723         case IPC_64:
 724                 if (copy_from_user(out, buf, sizeof(*out)))
 725                         return -EFAULT;
 726                 return 0;
 727         case IPC_OLD:
 728             {
 729                 struct shmid_ds tbuf_old;
 730
 731                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 732                         return -EFAULT;
 733
 734                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 735                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 736                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 737
 738                 return 0;
 739             }
 740         default:
 741                 return -EINVAL;
 742         }
 743 }
 744
 745 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 746 {
 747         switch (version) {
 748         case IPC_64:
 749                 return copy_to_user(buf, in, sizeof(*in));
 750         case IPC_OLD:
 751             {
 752                 struct shminfo out;
 753
 754                 if (in->shmmax > INT_MAX)
 755                         out.shmmax = INT_MAX;
 756                 else
 757                         out.shmmax = (int)in->shmmax;
 758
 759                 out.shmmin      = in->shmmin;
 760                 out.shmmni      = in->shmmni;
 761                 out.shmseg      = in->shmseg;
 762                 out.shmall      = in->shmall;
 763
 764                 return copy_to_user(buf, &out, sizeof(out));
 765             }
 766         default:
 767                 return -EINVAL;
 768         }
 769 }
 770
 771 /*
 772  * Calculate and add used RSS and swap pages of a shm.
 773  * Called with shm_ids.rwsem held as a reader
 774  */
 775 static void shm_add_rss_swap(struct shmid_kernel *shp,
 776         unsigned long *rss_add, unsigned long *swp_add)
 777 {
 778         struct inode *inode;
 779
 780         inode = file_inode(shp->shm_file);
 781
 782         if (is_file_hugepages(shp->shm_file)) {
 783                 struct address_space *mapping = inode->i_mapping;
 784                 struct hstate *h = hstate_file(shp->shm_file);
 785                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 786         } else {
 787 #ifdef CONFIG_SHMEM
 788                 struct shmem_inode_info *info = SHMEM_I(inode);
 789
 790                 spin_lock_irq(&info->lock);
 791                 *rss_add += inode->i_mapping->nrpages;
 792                 *swp_add += info->swapped;
 793                 spin_unlock_irq(&info->lock);
 794 #else
 795                 *rss_add += inode->i_mapping->nrpages;
 796 #endif
 797         }
 798 }
 799
 800 /*
 801  * Called with shm_ids.rwsem held as a reader
 802  */
 803 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 804                 unsigned long *swp)
 805 {
 806         int next_id;
 807         int total, in_use;
 808
 809         *rss = 0;
 810         *swp = 0;
 811
 812         in_use = shm_ids(ns).in_use;
 813
 814         for (total = 0, next_id = 0; total < in_use; next_id++) {
 815                 struct kern_ipc_perm *ipc;
 816                 struct shmid_kernel *shp;
 817
 818                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 819                 if (ipc == NULL)
 820                         continue;
 821                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
 822
 823                 shm_add_rss_swap(shp, rss, swp);
 824
 825                 total++;
 826         }
 827 }
 828
 829 /*
 830  * This function handles some shmctl commands which require the rwsem
 831  * to be held in write mode.
 832  * NOTE: no locks must be held, the rwsem is taken inside this function.
 833  */
 834 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 835                        struct shmid64_ds *shmid64)
 836 {
 837         struct kern_ipc_perm *ipcp;
 838         struct shmid_kernel *shp;
 839         int err;
 840
 841         down_write(&shm_ids(ns).rwsem);
 842         rcu_read_lock();
 843
 844         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
 845                                       &shmid64->shm_perm, 0);
 846         if (IS_ERR(ipcp)) {
 847                 err = PTR_ERR(ipcp);
 848                 goto out_unlock1;
 849         }
 850
 851         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 852
 853         err = security_shm_shmctl(&shp->shm_perm, cmd);
 854         if (err)
 855                 goto out_unlock1;
 856
 857         switch (cmd) {
 858         case IPC_RMID:
 859                 ipc_lock_object(&shp->shm_perm);
 860                 /* do_shm_rmid unlocks the ipc object and rcu */
 861                 do_shm_rmid(ns, ipcp);
 862                 goto out_up;
 863         case IPC_SET:
 864                 ipc_lock_object(&shp->shm_perm);
 865                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
 866                 if (err)
 867                         goto out_unlock0;
 868                 shp->shm_ctim = ktime_get_real_seconds();
 869                 break;
 870         default:
 871                 err = -EINVAL;
 872                 goto out_unlock1;
 873         }
 874
 875 out_unlock0:
 876         ipc_unlock_object(&shp->shm_perm);
 877 out_unlock1:
 878         rcu_read_unlock();
 879 out_up:
 880         up_write(&shm_ids(ns).rwsem);
 881         return err;
 882 }
 883
 884 static int shmctl_ipc_info(struct ipc_namespace *ns,
 885                            struct shminfo64 *shminfo)
 886 {
 887         int err = security_shm_shmctl(NULL, IPC_INFO);
 888         if (!err) {
 889                 memset(shminfo, 0, sizeof(*shminfo));
 890                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
 891                 shminfo->shmmax = ns->shm_ctlmax;
 892                 shminfo->shmall = ns->shm_ctlall;
 893                 shminfo->shmmin = SHMMIN;
 894                 down_read(&shm_ids(ns).rwsem);
 895                 err = ipc_get_maxid(&shm_ids(ns));
 896                 up_read(&shm_ids(ns).rwsem);
 897                 if (err < 0)
 898                         err = 0;
 899         }
 900         return err;
 901 }
 902
 903 static int shmctl_shm_info(struct ipc_namespace *ns,
 904                            struct shm_info *shm_info)
 905 {
 906         int err = security_shm_shmctl(NULL, SHM_INFO);
 907         if (!err) {
 908                 memset(shm_info, 0, sizeof(*shm_info));
 909                 down_read(&shm_ids(ns).rwsem);
 910                 shm_info->used_ids = shm_ids(ns).in_use;
 911                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
 912                 shm_info->shm_tot = ns->shm_tot;
 913                 shm_info->swap_attempts = 0;
 914                 shm_info->swap_successes = 0;
 915                 err = ipc_get_maxid(&shm_ids(ns));
 916                 up_read(&shm_ids(ns).rwsem);
 917                 if (err < 0)
 918                         err = 0;
 919         }
 920         return err;
 921 }
 922
 923 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
 924                         int cmd, struct shmid64_ds *tbuf)
 925 {
 926         struct shmid_kernel *shp;
 927         int id = 0;
 928         int err;
 929
 930         memset(tbuf, 0, sizeof(*tbuf));
 931
 932         rcu_read_lock();
 933         if (cmd == SHM_STAT) {
 934                 shp = shm_obtain_object(ns, shmid);
 935                 if (IS_ERR(shp)) {
 936                         err = PTR_ERR(shp);
 937                         goto out_unlock;
 938                 }
 939                 id = shp->shm_perm.id;
 940         } else {
 941                 shp = shm_obtain_object_check(ns, shmid);
 942                 if (IS_ERR(shp)) {
 943                         err = PTR_ERR(shp);
 944                         goto out_unlock;
 945                 }
 946         }
 947
 948         err = -EACCES;
 949         if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 950                 goto out_unlock;
 951
 952         err = security_shm_shmctl(&shp->shm_perm, cmd);
 953         if (err)
 954                 goto out_unlock;
 955
 956         ipc_lock_object(&shp->shm_perm);
 957
 958         if (!ipc_valid_object(&shp->shm_perm)) {
 959                 ipc_unlock_object(&shp->shm_perm);
 960                 err = -EIDRM;
 961                 goto out_unlock;
 962         }
 963
 964         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
 965         tbuf->shm_segsz = shp->shm_segsz;
 966         tbuf->shm_atime = shp->shm_atim;
 967         tbuf->shm_dtime = shp->shm_dtim;
 968         tbuf->shm_ctime = shp->shm_ctim;
 969         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
 970         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
 971         tbuf->shm_nattch = shp->shm_nattch;
 972
 973         ipc_unlock_object(&shp->shm_perm);
 974         rcu_read_unlock();
 975         return id;
 976
 977 out_unlock:
 978         rcu_read_unlock();
 979         return err;
 980 }
 981
 982 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
 983 {
 984         struct shmid_kernel *shp;
 985         struct file *shm_file;
 986         int err;
 987
 988         rcu_read_lock();
 989         shp = shm_obtain_object_check(ns, shmid);
 990         if (IS_ERR(shp)) {
 991                 err = PTR_ERR(shp);
 992                 goto out_unlock1;
 993         }
 994
 995         audit_ipc_obj(&(shp->shm_perm));
 996         err = security_shm_shmctl(&shp->shm_perm, cmd);
 997         if (err)
 998                 goto out_unlock1;
 999
1000         ipc_lock_object(&shp->shm_perm);
1001
1002         /* check if shm_destroy() is tearing down shp */
1003         if (!ipc_valid_object(&shp->shm_perm)) {
1004                 err = -EIDRM;
1005                 goto out_unlock0;
1006         }
1007
1008         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1009                 kuid_t euid = current_euid();
1010
1011                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1012                     !uid_eq(euid, shp->shm_perm.cuid)) {
1013                         err = -EPERM;
1014                         goto out_unlock0;
1015                 }
1016                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1017                         err = -EPERM;
1018                         goto out_unlock0;
1019                 }
1020         }
1021
1022         shm_file = shp->shm_file;
1023         if (is_file_hugepages(shm_file))
1024                 goto out_unlock0;
1025
1026         if (cmd == SHM_LOCK) {
1027                 struct user_struct *user = current_user();
1028
1029                 err = shmem_lock(shm_file, 1, user);
1030                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1031                         shp->shm_perm.mode |= SHM_LOCKED;
1032                         shp->mlock_user = user;
1033                 }
1034                 goto out_unlock0;
1035         }
1036
1037         /* SHM_UNLOCK */
1038         if (!(shp->shm_perm.mode & SHM_LOCKED))
1039                 goto out_unlock0;
1040         shmem_lock(shm_file, 0, shp->mlock_user);
1041         shp->shm_perm.mode &= ~SHM_LOCKED;
1042         shp->mlock_user = NULL;
1043         get_file(shm_file);
1044         ipc_unlock_object(&shp->shm_perm);
1045         rcu_read_unlock();
1046         shmem_unlock_mapping(shm_file->f_mapping);
1047
1048         fput(shm_file);
1049         return err;
1050
1051 out_unlock0:
1052         ipc_unlock_object(&shp->shm_perm);
1053 out_unlock1:
1054         rcu_read_unlock();
1055         return err;
1056 }
1057
1058 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1059 {
1060         int err, version;
1061         struct ipc_namespace *ns;
1062         struct shmid64_ds sem64;
1063
1064         if (cmd < 0 || shmid < 0)
1065                 return -EINVAL;
1066
1067         version = ipc_parse_version(&cmd);
1068         ns = current->nsproxy->ipc_ns;
1069
1070         switch (cmd) {
1071         case IPC_INFO: {
1072                 struct shminfo64 shminfo;
1073                 err = shmctl_ipc_info(ns, &shminfo);
1074                 if (err < 0)
1075                         return err;
1076                 if (copy_shminfo_to_user(buf, &shminfo, version))
1077                         err = -EFAULT;
1078                 return err;
1079         }
1080         case SHM_INFO: {
1081                 struct shm_info shm_info;
1082                 err = shmctl_shm_info(ns, &shm_info);
1083                 if (err < 0)
1084                         return err;
1085                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1086                         err = -EFAULT;
1087                 return err;
1088         }
1089         case SHM_STAT:
1090         case IPC_STAT: {
1091                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1092                 if (err < 0)
1093                         return err;
1094                 if (copy_shmid_to_user(buf, &sem64, version))
1095                         err = -EFAULT;
1096                 return err;
1097         }
1098         case IPC_SET:
1099                 if (copy_shmid_from_user(&sem64, buf, version))
1100                         return -EFAULT;
1101                 /* fallthru */
1102         case IPC_RMID:
1103                 return shmctl_down(ns, shmid, cmd, &sem64);
1104         case SHM_LOCK:
1105         case SHM_UNLOCK:
1106                 return shmctl_do_lock(ns, shmid, cmd);
1107         default:
1108                 return -EINVAL;
1109         }
1110 }
1111
1112 #ifdef CONFIG_COMPAT
1113
1114 struct compat_shmid_ds {
1115         struct compat_ipc_perm shm_perm;
1116         int shm_segsz;
1117         compat_time_t shm_atime;
1118         compat_time_t shm_dtime;
1119         compat_time_t shm_ctime;
1120         compat_ipc_pid_t shm_cpid;
1121         compat_ipc_pid_t shm_lpid;
1122         unsigned short shm_nattch;
1123         unsigned short shm_unused;
1124         compat_uptr_t shm_unused2;
1125         compat_uptr_t shm_unused3;
1126 };
1127
1128 struct compat_shminfo64 {
1129         compat_ulong_t shmmax;
1130         compat_ulong_t shmmin;
1131         compat_ulong_t shmmni;
1132         compat_ulong_t shmseg;
1133         compat_ulong_t shmall;
1134         compat_ulong_t __unused1;
1135         compat_ulong_t __unused2;
1136         compat_ulong_t __unused3;
1137         compat_ulong_t __unused4;
1138 };
1139
1140 struct compat_shm_info {
1141         compat_int_t used_ids;
1142         compat_ulong_t shm_tot, shm_rss, shm_swp;
1143         compat_ulong_t swap_attempts, swap_successes;
1144 };
1145
1146 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1147                                         int version)
1148 {
1149         if (in->shmmax > INT_MAX)
1150                 in->shmmax = INT_MAX;
1151         if (version == IPC_64) {
1152                 struct compat_shminfo64 info;
1153                 memset(&info, 0, sizeof(info));
1154                 info.shmmax = in->shmmax;
1155                 info.shmmin = in->shmmin;
1156                 info.shmmni = in->shmmni;
1157                 info.shmseg = in->shmseg;
1158                 info.shmall = in->shmall;
1159                 return copy_to_user(buf, &info, sizeof(info));
1160         } else {
1161                 struct shminfo info;
1162                 memset(&info, 0, sizeof(info));
1163                 info.shmmax = in->shmmax;
1164                 info.shmmin = in->shmmin;
1165                 info.shmmni = in->shmmni;
1166                 info.shmseg = in->shmseg;
1167                 info.shmall = in->shmall;
1168                 return copy_to_user(buf, &info, sizeof(info));
1169         }
1170 }
1171
1172 static int put_compat_shm_info(struct shm_info *ip,
1173                                 struct compat_shm_info __user *uip)
1174 {
1175         struct compat_shm_info info;
1176
1177         memset(&info, 0, sizeof(info));
1178         info.used_ids = ip->used_ids;
1179         info.shm_tot = ip->shm_tot;
1180         info.shm_rss = ip->shm_rss;
1181         info.shm_swp = ip->shm_swp;
1182         info.swap_attempts = ip->swap_attempts;
1183         info.swap_successes = ip->swap_successes;
1184         return copy_to_user(uip, &info, sizeof(info));
1185 }
1186
1187 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1188                                         int version)
1189 {
1190         if (version == IPC_64) {
1191                 struct compat_shmid64_ds v;
1192                 memset(&v, 0, sizeof(v));
1193                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1194                 v.shm_atime = in->shm_atime;
1195                 v.shm_dtime = in->shm_dtime;
1196                 v.shm_ctime = in->shm_ctime;
1197                 v.shm_segsz = in->shm_segsz;
1198                 v.shm_nattch = in->shm_nattch;
1199                 v.shm_cpid = in->shm_cpid;
1200                 v.shm_lpid = in->shm_lpid;
1201                 return copy_to_user(buf, &v, sizeof(v));
1202         } else {
1203                 struct compat_shmid_ds v;
1204                 memset(&v, 0, sizeof(v));
1205                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1206                 v.shm_perm.key = in->shm_perm.key;
1207                 v.shm_atime = in->shm_atime;
1208                 v.shm_dtime = in->shm_dtime;
1209                 v.shm_ctime = in->shm_ctime;
1210                 v.shm_segsz = in->shm_segsz;
1211                 v.shm_nattch = in->shm_nattch;
1212                 v.shm_cpid = in->shm_cpid;
1213                 v.shm_lpid = in->shm_lpid;
1214                 return copy_to_user(buf, &v, sizeof(v));
1215         }
1216 }
1217
1218 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1219                                         int version)
1220 {
1221         memset(out, 0, sizeof(*out));
1222         if (version == IPC_64) {
1223                 struct compat_shmid64_ds __user *p = buf;
1224                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1225         } else {
1226                 struct compat_shmid_ds __user *p = buf;
1227                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1228         }
1229 }
1230
1231 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1232 {
1233         struct ipc_namespace *ns;
1234         struct shmid64_ds sem64;
1235         int version = compat_ipc_parse_version(&cmd);
1236         int err;
1237
1238         ns = current->nsproxy->ipc_ns;
1239
1240         if (cmd < 0 || shmid < 0)
1241                 return -EINVAL;
1242
1243         switch (cmd) {
1244         case IPC_INFO: {
1245                 struct shminfo64 shminfo;
1246                 err = shmctl_ipc_info(ns, &shminfo);
1247                 if (err < 0)
1248                         return err;
1249                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1250                         err = -EFAULT;
1251                 return err;
1252         }
1253         case SHM_INFO: {
1254                 struct shm_info shm_info;
1255                 err = shmctl_shm_info(ns, &shm_info);
1256                 if (err < 0)
1257                         return err;
1258                 if (put_compat_shm_info(&shm_info, uptr))
1259                         err = -EFAULT;
1260                 return err;
1261         }
1262         case IPC_STAT:
1263         case SHM_STAT:
1264                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1265                 if (err < 0)
1266                         return err;
1267                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1268                         err = -EFAULT;
1269                 return err;
1270
1271         case IPC_SET:
1272                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1273                         return -EFAULT;
1274                 /* fallthru */
1275         case IPC_RMID:
1276                 return shmctl_down(ns, shmid, cmd, &sem64);
1277         case SHM_LOCK:
1278         case SHM_UNLOCK:
1279                 return shmctl_do_lock(ns, shmid, cmd);
1280                 break;
1281         default:
1282                 return -EINVAL;
1283         }
1284         return err;
1285 }
1286 #endif
1287
1288 /*
1289  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1290  *
1291  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1292  * "raddr" thing points to kernel space, and there has to be a wrapper around
1293  * this.
1294  */
1295 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1296               ulong *raddr, unsigned long shmlba)
1297 {
1298         struct shmid_kernel *shp;
1299         unsigned long addr = (unsigned long)shmaddr;
1300         unsigned long size;
1301         struct file *file;
1302         int    err;
1303         unsigned long flags = MAP_SHARED;
1304         unsigned long prot;
1305         int acc_mode;
1306         struct ipc_namespace *ns;
1307         struct shm_file_data *sfd;
1308         struct path path;
1309         fmode_t f_mode;
1310         unsigned long populate = 0;
1311
1312         err = -EINVAL;
1313         if (shmid < 0)
1314                 goto out;
1315
1316         if (addr) {
1317                 if (addr & (shmlba - 1)) {
1318                         /*
1319                          * Round down to the nearest multiple of shmlba.
1320                          * For sane do_mmap_pgoff() parameters, avoid
1321                          * round downs that trigger nil-page and MAP_FIXED.
1322                          */
1323                         if ((shmflg & SHM_RND) && addr >= shmlba)
1324                                 addr &= ~(shmlba - 1);
1325                         else
1326 #ifndef __ARCH_FORCE_SHMLBA
1327                                 if (addr & ~PAGE_MASK)
1328 #endif
1329                                         goto out;
1330                 }
1331
1332                 flags |= MAP_FIXED;
1333         } else if ((shmflg & SHM_REMAP))
1334                 goto out;
1335
1336         if (shmflg & SHM_RDONLY) {
1337                 prot = PROT_READ;
1338                 acc_mode = S_IRUGO;
1339                 f_mode = FMODE_READ;
1340         } else {
1341                 prot = PROT_READ | PROT_WRITE;
1342                 acc_mode = S_IRUGO | S_IWUGO;
1343                 f_mode = FMODE_READ | FMODE_WRITE;
1344         }
1345         if (shmflg & SHM_EXEC) {
1346                 prot |= PROT_EXEC;
1347                 acc_mode |= S_IXUGO;
1348         }
1349
1350         /*
1351          * We cannot rely on the fs check since SYSV IPC does have an
1352          * additional creator id...
1353          */
1354         ns = current->nsproxy->ipc_ns;
1355         rcu_read_lock();
1356         shp = shm_obtain_object_check(ns, shmid);
1357         if (IS_ERR(shp)) {
1358                 err = PTR_ERR(shp);
1359                 goto out_unlock;
1360         }
1361
1362         err = -EACCES;
1363         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1364                 goto out_unlock;
1365
1366         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1367         if (err)
1368                 goto out_unlock;
1369
1370         ipc_lock_object(&shp->shm_perm);
1371
1372         /* check if shm_destroy() is tearing down shp */
1373         if (!ipc_valid_object(&shp->shm_perm)) {
1374                 ipc_unlock_object(&shp->shm_perm);
1375                 err = -EIDRM;
1376                 goto out_unlock;
1377         }
1378
1379         path = shp->shm_file->f_path;
1380         path_get(&path);
1381         shp->shm_nattch++;
1382         size = i_size_read(d_inode(path.dentry));
1383         ipc_unlock_object(&shp->shm_perm);
1384         rcu_read_unlock();
1385
1386         err = -ENOMEM;
1387         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1388         if (!sfd) {
1389                 path_put(&path);
1390                 goto out_nattch;
1391         }
1392
1393         file = alloc_file(&path, f_mode,
1394                           is_file_hugepages(shp->shm_file) ?
1395                                 &shm_file_operations_huge :
1396                                 &shm_file_operations);
1397         err = PTR_ERR(file);
1398         if (IS_ERR(file)) {
1399                 kfree(sfd);
1400                 path_put(&path);
1401                 goto out_nattch;
1402         }
1403
1404         file->private_data = sfd;
1405         file->f_mapping = shp->shm_file->f_mapping;
1406         sfd->id = shp->shm_perm.id;
1407         sfd->ns = get_ipc_ns(ns);
1408         sfd->file = shp->shm_file;
1409         sfd->vm_ops = NULL;
1410
1411         err = security_mmap_file(file, prot, flags);
1412         if (err)
1413                 goto out_fput;
1414
1415         if (down_write_killable(&current->mm->mmap_sem)) {
1416                 err = -EINTR;
1417                 goto out_fput;
1418         }
1419
1420         if (addr && !(shmflg & SHM_REMAP)) {
1421                 err = -EINVAL;
1422                 if (addr + size < addr)
1423                         goto invalid;
1424
1425                 if (find_vma_intersection(current->mm, addr, addr + size))
1426                         goto invalid;
1427         }
1428
1429         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1430         *raddr = addr;
1431         err = 0;
1432         if (IS_ERR_VALUE(addr))
1433                 err = (long)addr;
1434 invalid:
1435         up_write(&current->mm->mmap_sem);
1436         if (populate)
1437                 mm_populate(addr, populate);
1438
1439 out_fput:
1440         fput(file);
1441
1442 out_nattch:
1443         down_write(&shm_ids(ns).rwsem);
1444         shp = shm_lock(ns, shmid);
1445         shp->shm_nattch--;
1446         if (shm_may_destroy(ns, shp))
1447                 shm_destroy(ns, shp);
1448         else
1449                 shm_unlock(shp);
1450         up_write(&shm_ids(ns).rwsem);
1451         return err;
1452
1453 out_unlock:
1454         rcu_read_unlock();
1455 out:
1456         return err;
1457 }
1458
1459 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1460 {
1461         unsigned long ret;
1462         long err;
1463
1464         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1465         if (err)
1466                 return err;
1467         force_successful_syscall_return();
1468         return (long)ret;
1469 }
1470
1471 #ifdef CONFIG_COMPAT
1472
1473 #ifndef COMPAT_SHMLBA
1474 #define COMPAT_SHMLBA   SHMLBA
1475 #endif
1476
1477 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1478 {
1479         unsigned long ret;
1480         long err;
1481
1482         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1483         if (err)
1484                 return err;
1485         force_successful_syscall_return();
1486         return (long)ret;
1487 }
1488 #endif
1489
1490 /*
1491  * detach and kill segment if marked destroyed.
1492  * The work is done in shm_close.
1493  */
1494 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1495 {
1496         struct mm_struct *mm = current->mm;
1497         struct vm_area_struct *vma;
1498         unsigned long addr = (unsigned long)shmaddr;
1499         int retval = -EINVAL;
1500 #ifdef CONFIG_MMU
1501         loff_t size = 0;
1502         struct file *file;
1503         struct vm_area_struct *next;
1504 #endif
1505
1506         if (addr & ~PAGE_MASK)
1507                 return retval;
1508
1509         if (down_write_killable(&mm->mmap_sem))
1510                 return -EINTR;
1511
1512         /*
1513          * This function tries to be smart and unmap shm segments that
1514          * were modified by partial mlock or munmap calls:
1515          * - It first determines the size of the shm segment that should be
1516          *   unmapped: It searches for a vma that is backed by shm and that
1517          *   started at address shmaddr. It records it's size and then unmaps
1518          *   it.
1519          * - Then it unmaps all shm vmas that started at shmaddr and that
1520          *   are within the initially determined size and that are from the
1521          *   same shm segment from which we determined the size.
1522          * Errors from do_munmap are ignored: the function only fails if
1523          * it's called with invalid parameters or if it's called to unmap
1524          * a part of a vma. Both calls in this function are for full vmas,
1525          * the parameters are directly copied from the vma itself and always
1526          * valid - therefore do_munmap cannot fail. (famous last words?)
1527          */
1528         /*
1529          * If it had been mremap()'d, the starting address would not
1530          * match the usual checks anyway. So assume all vma's are
1531          * above the starting address given.
1532          */
1533         vma = find_vma(mm, addr);
1534
1535 #ifdef CONFIG_MMU
1536         while (vma) {
1537                 next = vma->vm_next;
1538
1539                 /*
1540                  * Check if the starting address would match, i.e. it's
1541                  * a fragment created by mprotect() and/or munmap(), or it
1542                  * otherwise it starts at this address with no hassles.
1543                  */
1544                 if ((vma->vm_ops == &shm_vm_ops) &&
1545                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1546
1547                         /*
1548                          * Record the file of the shm segment being
1549                          * unmapped.  With mremap(), someone could place
1550                          * page from another segment but with equal offsets
1551                          * in the range we are unmapping.
1552                          */
1553                         file = vma->vm_file;
1554                         size = i_size_read(file_inode(vma->vm_file));
1555                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1556                         /*
1557                          * We discovered the size of the shm segment, so
1558                          * break out of here and fall through to the next
1559                          * loop that uses the size information to stop
1560                          * searching for matching vma's.
1561                          */
1562                         retval = 0;
1563                         vma = next;
1564                         break;
1565                 }
1566                 vma = next;
1567         }
1568
1569         /*
1570          * We need look no further than the maximum address a fragment
1571          * could possibly have landed at. Also cast things to loff_t to
1572          * prevent overflows and make comparisons vs. equal-width types.
1573          */
1574         size = PAGE_ALIGN(size);
1575         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1576                 next = vma->vm_next;
1577
1578                 /* finding a matching vma now does not alter retval */
1579                 if ((vma->vm_ops == &shm_vm_ops) &&
1580                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1581                     (vma->vm_file == file))
1582                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1583                 vma = next;
1584         }
1585
1586 #else   /* CONFIG_MMU */
1587         /* under NOMMU conditions, the exact address to be destroyed must be
1588          * given
1589          */
1590         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1591                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1592                 retval = 0;
1593         }
1594
1595 #endif
1596
1597         up_write(&mm->mmap_sem);
1598         return retval;
1599 }
1600
1601 #ifdef CONFIG_PROC_FS
1602 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1603 {
1604         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1605         struct user_namespace *user_ns = seq_user_ns(s);
1606         struct kern_ipc_perm *ipcp = it;
1607         struct shmid_kernel *shp;
1608         unsigned long rss = 0, swp = 0;
1609
1610         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1611         shm_add_rss_swap(shp, &rss, &swp);
1612
1613 #if BITS_PER_LONG <= 32
1614 #define SIZE_SPEC "%10lu"
1615 #else
1616 #define SIZE_SPEC "%21lu"
1617 #endif
1618
1619         seq_printf(s,
1620                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1621                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1622                    SIZE_SPEC " " SIZE_SPEC "\n",
1623                    shp->shm_perm.key,
1624                    shp->shm_perm.id,
1625                    shp->shm_perm.mode,
1626                    shp->shm_segsz,
1627                    pid_nr_ns(shp->shm_cprid, pid_ns),
1628                    pid_nr_ns(shp->shm_lprid, pid_ns),
1629                    shp->shm_nattch,
1630                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1631                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1632                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1633                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1634                    shp->shm_atim,
1635                    shp->shm_dtim,
1636                    shp->shm_ctim,
1637                    rss * PAGE_SIZE,
1638                    swp * PAGE_SIZE);
1639
1640         return 0;
1641 }
1642 #endif