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