1 // SPDX-License-Identifier: GPL-2.0
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.
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>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Better ipc lock (kern_ipc_perm.lock) handling
25 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
28 #include <linux/slab.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>
47 #include <linux/uaccess.h>
51 struct shmid_kernel /* private to the kernel */
53 struct kern_ipc_perm shm_perm;
54 struct file *shm_file;
55 unsigned long shm_nattch;
56 unsigned long shm_segsz;
60 struct pid *shm_cprid;
61 struct pid *shm_lprid;
62 struct user_struct *mlock_user;
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 */
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 */
73 struct shm_file_data {
75 struct ipc_namespace *ns;
77 const struct vm_operations_struct *vm_ops;
80 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
82 static const struct file_operations shm_file_operations;
83 static const struct vm_operations_struct shm_vm_ops;
85 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
87 #define shm_unlock(shp) \
88 ipc_unlock(&(shp)->shm_perm)
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);
95 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
98 int shm_init_ns(struct ipc_namespace *ns)
100 ns->shm_ctlmax = SHMMAX;
101 ns->shm_ctlall = SHMALL;
102 ns->shm_ctlmni = SHMMNI;
103 ns->shm_rmid_forced = 0;
105 return ipc_init_ids(&shm_ids(ns));
109 * Called with shm_ids.rwsem (writer) and the shp structure locked.
110 * Only shm_ids.rwsem remains locked on exit.
112 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
114 struct shmid_kernel *shp;
116 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
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);
124 shm_destroy(ns, shp);
128 void shm_exit_ns(struct ipc_namespace *ns)
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);
136 static int __init ipc_ns_init(void)
138 const int err = shm_init_ns(&init_ipc_ns);
139 WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
143 pure_initcall(ipc_ns_init);
145 void __init shm_init(void)
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",
151 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
153 IPC_SHM_IDS, sysvipc_shm_proc_show);
156 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
158 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
161 return ERR_CAST(ipcp);
163 return container_of(ipcp, struct shmid_kernel, shm_perm);
166 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
168 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
171 return ERR_CAST(ipcp);
173 return container_of(ipcp, struct shmid_kernel, shm_perm);
177 * shm_lock_(check_) routines are called in the paths where the rwsem
178 * is not necessarily held.
180 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
182 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
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
191 return container_of(ipcp, struct shmid_kernel, shm_perm);
194 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
197 ipc_lock_object(&ipcp->shm_perm);
200 static void shm_rcu_free(struct rcu_head *head)
202 struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
204 struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
206 security_shm_free(&shp->shm_perm);
210 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
212 list_del(&s->shm_clist);
213 ipc_rmid(&shm_ids(ns), &s->shm_perm);
217 static int __shm_open(struct vm_area_struct *vma)
219 struct file *file = vma->vm_file;
220 struct shm_file_data *sfd = shm_file_data(file);
221 struct shmid_kernel *shp;
223 shp = shm_lock(sfd->ns, sfd->id);
228 if (shp->shm_file != sfd->file) {
234 shp->shm_atim = ktime_get_real_seconds();
235 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
241 /* This is called by fork, once for every shm attach. */
242 static void shm_open(struct vm_area_struct *vma)
244 int err = __shm_open(vma);
246 * We raced in the idr lookup or with shm_destroy().
247 * Either way, the ID is busted.
253 * shm_destroy - free the struct shmid_kernel
256 * @shp: struct to free
258 * It has to be called with shp and shm_ids.rwsem (writer) locked,
259 * but returns with shp unlocked and freed.
261 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
263 struct file *shm_file;
265 shm_file = shp->shm_file;
266 shp->shm_file = NULL;
267 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
270 if (!is_file_hugepages(shm_file))
271 shmem_lock(shm_file, 0, shp->mlock_user);
272 else if (shp->mlock_user)
273 user_shm_unlock(i_size_read(file_inode(shm_file)),
276 ipc_update_pid(&shp->shm_cprid, NULL);
277 ipc_update_pid(&shp->shm_lprid, NULL);
278 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
282 * shm_may_destroy - identifies whether shm segment should be destroyed now
284 * Returns true if and only if there are no active users of the segment and
285 * one of the following is true:
287 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
289 * 2) sysctl kernel.shm_rmid_forced is set to 1.
291 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
293 return (shp->shm_nattch == 0) &&
294 (ns->shm_rmid_forced ||
295 (shp->shm_perm.mode & SHM_DEST));
299 * remove the attach descriptor vma.
300 * free memory for segment if it is marked destroyed.
301 * The descriptor has already been removed from the current->mm->mmap list
302 * and will later be kfree()d.
304 static void shm_close(struct vm_area_struct *vma)
306 struct file *file = vma->vm_file;
307 struct shm_file_data *sfd = shm_file_data(file);
308 struct shmid_kernel *shp;
309 struct ipc_namespace *ns = sfd->ns;
311 down_write(&shm_ids(ns).rwsem);
312 /* remove from the list of attaches of the shm segment */
313 shp = shm_lock(ns, sfd->id);
316 * We raced in the idr lookup or with shm_destroy().
317 * Either way, the ID is busted.
319 if (WARN_ON_ONCE(IS_ERR(shp)))
320 goto done; /* no-op */
322 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
323 shp->shm_dtim = ktime_get_real_seconds();
325 if (shm_may_destroy(ns, shp))
326 shm_destroy(ns, shp);
330 up_write(&shm_ids(ns).rwsem);
333 /* Called with ns->shm_ids(ns).rwsem locked */
334 static int shm_try_destroy_orphaned(int id, void *p, void *data)
336 struct ipc_namespace *ns = data;
337 struct kern_ipc_perm *ipcp = p;
338 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
341 * We want to destroy segments without users and with already
342 * exit'ed originating process.
344 * As shp->* are changed under rwsem, it's safe to skip shp locking.
346 if (shp->shm_creator != NULL)
349 if (shm_may_destroy(ns, shp)) {
350 shm_lock_by_ptr(shp);
351 shm_destroy(ns, shp);
356 void shm_destroy_orphaned(struct ipc_namespace *ns)
358 down_write(&shm_ids(ns).rwsem);
359 if (shm_ids(ns).in_use)
360 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
361 up_write(&shm_ids(ns).rwsem);
364 /* Locking assumes this will only be called with task == current */
365 void exit_shm(struct task_struct *task)
367 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
368 struct shmid_kernel *shp, *n;
370 if (list_empty(&task->sysvshm.shm_clist))
374 * If kernel.shm_rmid_forced is not set then only keep track of
375 * which shmids are orphaned, so that a later set of the sysctl
378 if (!ns->shm_rmid_forced) {
379 down_read(&shm_ids(ns).rwsem);
380 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
381 shp->shm_creator = NULL;
383 * Only under read lock but we are only called on current
384 * so no entry on the list will be shared.
386 list_del(&task->sysvshm.shm_clist);
387 up_read(&shm_ids(ns).rwsem);
392 * Destroy all already created segments, that were not yet mapped,
393 * and mark any mapped as orphan to cover the sysctl toggling.
394 * Destroy is skipped if shm_may_destroy() returns false.
396 down_write(&shm_ids(ns).rwsem);
397 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
398 shp->shm_creator = NULL;
400 if (shm_may_destroy(ns, shp)) {
401 shm_lock_by_ptr(shp);
402 shm_destroy(ns, shp);
406 /* Remove the list head from any segments still attached. */
407 list_del(&task->sysvshm.shm_clist);
408 up_write(&shm_ids(ns).rwsem);
411 static vm_fault_t shm_fault(struct vm_fault *vmf)
413 struct file *file = vmf->vma->vm_file;
414 struct shm_file_data *sfd = shm_file_data(file);
416 return sfd->vm_ops->fault(vmf);
419 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
421 struct file *file = vma->vm_file;
422 struct shm_file_data *sfd = shm_file_data(file);
424 if (sfd->vm_ops->split)
425 return sfd->vm_ops->split(vma, addr);
430 static unsigned long shm_pagesize(struct vm_area_struct *vma)
432 struct file *file = vma->vm_file;
433 struct shm_file_data *sfd = shm_file_data(file);
435 if (sfd->vm_ops->pagesize)
436 return sfd->vm_ops->pagesize(vma);
442 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
444 struct file *file = vma->vm_file;
445 struct shm_file_data *sfd = shm_file_data(file);
448 if (sfd->vm_ops->set_policy)
449 err = sfd->vm_ops->set_policy(vma, new);
453 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
456 struct file *file = vma->vm_file;
457 struct shm_file_data *sfd = shm_file_data(file);
458 struct mempolicy *pol = NULL;
460 if (sfd->vm_ops->get_policy)
461 pol = sfd->vm_ops->get_policy(vma, addr);
462 else if (vma->vm_policy)
463 pol = vma->vm_policy;
469 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
471 struct shm_file_data *sfd = shm_file_data(file);
475 * In case of remap_file_pages() emulation, the file can represent an
476 * IPC ID that was removed, and possibly even reused by another shm
477 * segment already. Propagate this case as an error to caller.
479 ret = __shm_open(vma);
483 ret = call_mmap(sfd->file, vma);
488 sfd->vm_ops = vma->vm_ops;
490 WARN_ON(!sfd->vm_ops->fault);
492 vma->vm_ops = &shm_vm_ops;
496 static int shm_release(struct inode *ino, struct file *file)
498 struct shm_file_data *sfd = shm_file_data(file);
502 shm_file_data(file) = NULL;
507 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
509 struct shm_file_data *sfd = shm_file_data(file);
511 if (!sfd->file->f_op->fsync)
513 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
516 static long shm_fallocate(struct file *file, int mode, loff_t offset,
519 struct shm_file_data *sfd = shm_file_data(file);
521 if (!sfd->file->f_op->fallocate)
523 return sfd->file->f_op->fallocate(file, mode, offset, len);
526 static unsigned long shm_get_unmapped_area(struct file *file,
527 unsigned long addr, unsigned long len, unsigned long pgoff,
530 struct shm_file_data *sfd = shm_file_data(file);
532 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
536 static const struct file_operations shm_file_operations = {
539 .release = shm_release,
540 .get_unmapped_area = shm_get_unmapped_area,
541 .llseek = noop_llseek,
542 .fallocate = shm_fallocate,
546 * shm_file_operations_huge is now identical to shm_file_operations,
547 * but we keep it distinct for the sake of is_file_shm_hugepages().
549 static const struct file_operations shm_file_operations_huge = {
552 .release = shm_release,
553 .get_unmapped_area = shm_get_unmapped_area,
554 .llseek = noop_llseek,
555 .fallocate = shm_fallocate,
558 bool is_file_shm_hugepages(struct file *file)
560 return file->f_op == &shm_file_operations_huge;
563 static const struct vm_operations_struct shm_vm_ops = {
564 .open = shm_open, /* callback for a new vm-area open */
565 .close = shm_close, /* callback for when the vm-area is released */
568 .pagesize = shm_pagesize,
569 #if defined(CONFIG_NUMA)
570 .set_policy = shm_set_policy,
571 .get_policy = shm_get_policy,
576 * newseg - Create a new shared memory segment
578 * @params: ptr to the structure that contains key, size and shmflg
580 * Called with shm_ids.rwsem held as a writer.
582 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
584 key_t key = params->key;
585 int shmflg = params->flg;
586 size_t size = params->u.size;
588 struct shmid_kernel *shp;
589 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
592 vm_flags_t acctflag = 0;
594 if (size < SHMMIN || size > ns->shm_ctlmax)
597 if (numpages << PAGE_SHIFT < size)
600 if (ns->shm_tot + numpages < ns->shm_tot ||
601 ns->shm_tot + numpages > ns->shm_ctlall)
604 shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
608 shp->shm_perm.key = key;
609 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
610 shp->mlock_user = NULL;
612 shp->shm_perm.security = NULL;
613 error = security_shm_alloc(&shp->shm_perm);
619 sprintf(name, "SYSV%08x", key);
620 if (shmflg & SHM_HUGETLB) {
624 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
629 hugesize = ALIGN(size, huge_page_size(hs));
631 /* hugetlb_file_setup applies strict accounting */
632 if (shmflg & SHM_NORESERVE)
633 acctflag = VM_NORESERVE;
634 file = hugetlb_file_setup(name, hugesize, acctflag,
635 &shp->mlock_user, HUGETLB_SHMFS_INODE,
636 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
639 * Do not allow no accounting for OVERCOMMIT_NEVER, even
642 if ((shmflg & SHM_NORESERVE) &&
643 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
644 acctflag = VM_NORESERVE;
645 file = shmem_kernel_file_setup(name, size, acctflag);
647 error = PTR_ERR(file);
651 shp->shm_cprid = get_pid(task_tgid(current));
652 shp->shm_lprid = NULL;
653 shp->shm_atim = shp->shm_dtim = 0;
654 shp->shm_ctim = ktime_get_real_seconds();
655 shp->shm_segsz = size;
657 shp->shm_file = file;
658 shp->shm_creator = current;
660 /* ipc_addid() locks shp upon success. */
661 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
665 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
668 * shmid gets reported as "inode#" in /proc/pid/maps.
669 * proc-ps tools use this. Changing this will break them.
671 file_inode(file)->i_ino = shp->shm_perm.id;
673 ns->shm_tot += numpages;
674 error = shp->shm_perm.id;
676 ipc_unlock_object(&shp->shm_perm);
681 ipc_update_pid(&shp->shm_cprid, NULL);
682 ipc_update_pid(&shp->shm_lprid, NULL);
683 if (is_file_hugepages(file) && shp->mlock_user)
684 user_shm_unlock(size, shp->mlock_user);
687 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
692 * Called with shm_ids.rwsem and ipcp locked.
694 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
695 struct ipc_params *params)
697 struct shmid_kernel *shp;
699 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
700 if (shp->shm_segsz < params->u.size)
706 long ksys_shmget(key_t key, size_t size, int shmflg)
708 struct ipc_namespace *ns;
709 static const struct ipc_ops shm_ops = {
711 .associate = security_shm_associate,
712 .more_checks = shm_more_checks,
714 struct ipc_params shm_params;
716 ns = current->nsproxy->ipc_ns;
718 shm_params.key = key;
719 shm_params.flg = shmflg;
720 shm_params.u.size = size;
722 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
725 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
727 return ksys_shmget(key, size, shmflg);
730 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
734 return copy_to_user(buf, in, sizeof(*in));
739 memset(&out, 0, sizeof(out));
740 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
741 out.shm_segsz = in->shm_segsz;
742 out.shm_atime = in->shm_atime;
743 out.shm_dtime = in->shm_dtime;
744 out.shm_ctime = in->shm_ctime;
745 out.shm_cpid = in->shm_cpid;
746 out.shm_lpid = in->shm_lpid;
747 out.shm_nattch = in->shm_nattch;
749 return copy_to_user(buf, &out, sizeof(out));
756 static inline unsigned long
757 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
761 if (copy_from_user(out, buf, sizeof(*out)))
766 struct shmid_ds tbuf_old;
768 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
771 out->shm_perm.uid = tbuf_old.shm_perm.uid;
772 out->shm_perm.gid = tbuf_old.shm_perm.gid;
773 out->shm_perm.mode = tbuf_old.shm_perm.mode;
782 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
786 return copy_to_user(buf, in, sizeof(*in));
791 if (in->shmmax > INT_MAX)
792 out.shmmax = INT_MAX;
794 out.shmmax = (int)in->shmmax;
796 out.shmmin = in->shmmin;
797 out.shmmni = in->shmmni;
798 out.shmseg = in->shmseg;
799 out.shmall = in->shmall;
801 return copy_to_user(buf, &out, sizeof(out));
809 * Calculate and add used RSS and swap pages of a shm.
810 * Called with shm_ids.rwsem held as a reader
812 static void shm_add_rss_swap(struct shmid_kernel *shp,
813 unsigned long *rss_add, unsigned long *swp_add)
817 inode = file_inode(shp->shm_file);
819 if (is_file_hugepages(shp->shm_file)) {
820 struct address_space *mapping = inode->i_mapping;
821 struct hstate *h = hstate_file(shp->shm_file);
822 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
825 struct shmem_inode_info *info = SHMEM_I(inode);
827 spin_lock_irq(&info->lock);
828 *rss_add += inode->i_mapping->nrpages;
829 *swp_add += info->swapped;
830 spin_unlock_irq(&info->lock);
832 *rss_add += inode->i_mapping->nrpages;
838 * Called with shm_ids.rwsem held as a reader
840 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
849 in_use = shm_ids(ns).in_use;
851 for (total = 0, next_id = 0; total < in_use; next_id++) {
852 struct kern_ipc_perm *ipc;
853 struct shmid_kernel *shp;
855 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
858 shp = container_of(ipc, struct shmid_kernel, shm_perm);
860 shm_add_rss_swap(shp, rss, swp);
867 * This function handles some shmctl commands which require the rwsem
868 * to be held in write mode.
869 * NOTE: no locks must be held, the rwsem is taken inside this function.
871 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
872 struct shmid64_ds *shmid64)
874 struct kern_ipc_perm *ipcp;
875 struct shmid_kernel *shp;
878 down_write(&shm_ids(ns).rwsem);
881 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
882 &shmid64->shm_perm, 0);
888 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
890 err = security_shm_shmctl(&shp->shm_perm, cmd);
896 ipc_lock_object(&shp->shm_perm);
897 /* do_shm_rmid unlocks the ipc object and rcu */
898 do_shm_rmid(ns, ipcp);
901 ipc_lock_object(&shp->shm_perm);
902 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
905 shp->shm_ctim = ktime_get_real_seconds();
913 ipc_unlock_object(&shp->shm_perm);
917 up_write(&shm_ids(ns).rwsem);
921 static int shmctl_ipc_info(struct ipc_namespace *ns,
922 struct shminfo64 *shminfo)
924 int err = security_shm_shmctl(NULL, IPC_INFO);
926 memset(shminfo, 0, sizeof(*shminfo));
927 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
928 shminfo->shmmax = ns->shm_ctlmax;
929 shminfo->shmall = ns->shm_ctlall;
930 shminfo->shmmin = SHMMIN;
931 down_read(&shm_ids(ns).rwsem);
932 err = ipc_get_maxid(&shm_ids(ns));
933 up_read(&shm_ids(ns).rwsem);
940 static int shmctl_shm_info(struct ipc_namespace *ns,
941 struct shm_info *shm_info)
943 int err = security_shm_shmctl(NULL, SHM_INFO);
945 memset(shm_info, 0, sizeof(*shm_info));
946 down_read(&shm_ids(ns).rwsem);
947 shm_info->used_ids = shm_ids(ns).in_use;
948 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
949 shm_info->shm_tot = ns->shm_tot;
950 shm_info->swap_attempts = 0;
951 shm_info->swap_successes = 0;
952 err = ipc_get_maxid(&shm_ids(ns));
953 up_read(&shm_ids(ns).rwsem);
960 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
961 int cmd, struct shmid64_ds *tbuf)
963 struct shmid_kernel *shp;
967 memset(tbuf, 0, sizeof(*tbuf));
970 if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
971 shp = shm_obtain_object(ns, shmid);
976 id = shp->shm_perm.id;
977 } else { /* IPC_STAT */
978 shp = shm_obtain_object_check(ns, shmid);
986 * Semantically SHM_STAT_ANY ought to be identical to
987 * that functionality provided by the /proc/sysvipc/
988 * interface. As such, only audit these calls and
989 * do not do traditional S_IRUGO permission checks on
992 if (cmd == SHM_STAT_ANY)
993 audit_ipc_obj(&shp->shm_perm);
996 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1000 err = security_shm_shmctl(&shp->shm_perm, cmd);
1004 ipc_lock_object(&shp->shm_perm);
1006 if (!ipc_valid_object(&shp->shm_perm)) {
1007 ipc_unlock_object(&shp->shm_perm);
1012 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1013 tbuf->shm_segsz = shp->shm_segsz;
1014 tbuf->shm_atime = shp->shm_atim;
1015 tbuf->shm_dtime = shp->shm_dtim;
1016 tbuf->shm_ctime = shp->shm_ctim;
1017 #ifndef CONFIG_64BIT
1018 tbuf->shm_atime_high = shp->shm_atim >> 32;
1019 tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1020 tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1022 tbuf->shm_cpid = pid_vnr(shp->shm_cprid);
1023 tbuf->shm_lpid = pid_vnr(shp->shm_lprid);
1024 tbuf->shm_nattch = shp->shm_nattch;
1026 ipc_unlock_object(&shp->shm_perm);
1035 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1037 struct shmid_kernel *shp;
1038 struct file *shm_file;
1042 shp = shm_obtain_object_check(ns, shmid);
1048 audit_ipc_obj(&(shp->shm_perm));
1049 err = security_shm_shmctl(&shp->shm_perm, cmd);
1053 ipc_lock_object(&shp->shm_perm);
1055 /* check if shm_destroy() is tearing down shp */
1056 if (!ipc_valid_object(&shp->shm_perm)) {
1061 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1062 kuid_t euid = current_euid();
1064 if (!uid_eq(euid, shp->shm_perm.uid) &&
1065 !uid_eq(euid, shp->shm_perm.cuid)) {
1069 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1075 shm_file = shp->shm_file;
1076 if (is_file_hugepages(shm_file))
1079 if (cmd == SHM_LOCK) {
1080 struct user_struct *user = current_user();
1082 err = shmem_lock(shm_file, 1, user);
1083 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1084 shp->shm_perm.mode |= SHM_LOCKED;
1085 shp->mlock_user = user;
1091 if (!(shp->shm_perm.mode & SHM_LOCKED))
1093 shmem_lock(shm_file, 0, shp->mlock_user);
1094 shp->shm_perm.mode &= ~SHM_LOCKED;
1095 shp->mlock_user = NULL;
1097 ipc_unlock_object(&shp->shm_perm);
1099 shmem_unlock_mapping(shm_file->f_mapping);
1105 ipc_unlock_object(&shp->shm_perm);
1111 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1114 struct ipc_namespace *ns;
1115 struct shmid64_ds sem64;
1117 if (cmd < 0 || shmid < 0)
1120 version = ipc_parse_version(&cmd);
1121 ns = current->nsproxy->ipc_ns;
1125 struct shminfo64 shminfo;
1126 err = shmctl_ipc_info(ns, &shminfo);
1129 if (copy_shminfo_to_user(buf, &shminfo, version))
1134 struct shm_info shm_info;
1135 err = shmctl_shm_info(ns, &shm_info);
1138 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1145 err = shmctl_stat(ns, shmid, cmd, &sem64);
1148 if (copy_shmid_to_user(buf, &sem64, version))
1153 if (copy_shmid_from_user(&sem64, buf, version))
1157 return shmctl_down(ns, shmid, cmd, &sem64);
1160 return shmctl_do_lock(ns, shmid, cmd);
1166 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1168 return ksys_shmctl(shmid, cmd, buf);
1171 #ifdef CONFIG_COMPAT
1173 struct compat_shmid_ds {
1174 struct compat_ipc_perm shm_perm;
1176 compat_time_t shm_atime;
1177 compat_time_t shm_dtime;
1178 compat_time_t shm_ctime;
1179 compat_ipc_pid_t shm_cpid;
1180 compat_ipc_pid_t shm_lpid;
1181 unsigned short shm_nattch;
1182 unsigned short shm_unused;
1183 compat_uptr_t shm_unused2;
1184 compat_uptr_t shm_unused3;
1187 struct compat_shminfo64 {
1188 compat_ulong_t shmmax;
1189 compat_ulong_t shmmin;
1190 compat_ulong_t shmmni;
1191 compat_ulong_t shmseg;
1192 compat_ulong_t shmall;
1193 compat_ulong_t __unused1;
1194 compat_ulong_t __unused2;
1195 compat_ulong_t __unused3;
1196 compat_ulong_t __unused4;
1199 struct compat_shm_info {
1200 compat_int_t used_ids;
1201 compat_ulong_t shm_tot, shm_rss, shm_swp;
1202 compat_ulong_t swap_attempts, swap_successes;
1205 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1208 if (in->shmmax > INT_MAX)
1209 in->shmmax = INT_MAX;
1210 if (version == IPC_64) {
1211 struct compat_shminfo64 info;
1212 memset(&info, 0, sizeof(info));
1213 info.shmmax = in->shmmax;
1214 info.shmmin = in->shmmin;
1215 info.shmmni = in->shmmni;
1216 info.shmseg = in->shmseg;
1217 info.shmall = in->shmall;
1218 return copy_to_user(buf, &info, sizeof(info));
1220 struct shminfo info;
1221 memset(&info, 0, sizeof(info));
1222 info.shmmax = in->shmmax;
1223 info.shmmin = in->shmmin;
1224 info.shmmni = in->shmmni;
1225 info.shmseg = in->shmseg;
1226 info.shmall = in->shmall;
1227 return copy_to_user(buf, &info, sizeof(info));
1231 static int put_compat_shm_info(struct shm_info *ip,
1232 struct compat_shm_info __user *uip)
1234 struct compat_shm_info info;
1236 memset(&info, 0, sizeof(info));
1237 info.used_ids = ip->used_ids;
1238 info.shm_tot = ip->shm_tot;
1239 info.shm_rss = ip->shm_rss;
1240 info.shm_swp = ip->shm_swp;
1241 info.swap_attempts = ip->swap_attempts;
1242 info.swap_successes = ip->swap_successes;
1243 return copy_to_user(uip, &info, sizeof(info));
1246 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1249 if (version == IPC_64) {
1250 struct compat_shmid64_ds v;
1251 memset(&v, 0, sizeof(v));
1252 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1253 v.shm_atime = lower_32_bits(in->shm_atime);
1254 v.shm_atime_high = upper_32_bits(in->shm_atime);
1255 v.shm_dtime = lower_32_bits(in->shm_dtime);
1256 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1257 v.shm_ctime = lower_32_bits(in->shm_ctime);
1258 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1259 v.shm_segsz = in->shm_segsz;
1260 v.shm_nattch = in->shm_nattch;
1261 v.shm_cpid = in->shm_cpid;
1262 v.shm_lpid = in->shm_lpid;
1263 return copy_to_user(buf, &v, sizeof(v));
1265 struct compat_shmid_ds v;
1266 memset(&v, 0, sizeof(v));
1267 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1268 v.shm_perm.key = in->shm_perm.key;
1269 v.shm_atime = in->shm_atime;
1270 v.shm_dtime = in->shm_dtime;
1271 v.shm_ctime = in->shm_ctime;
1272 v.shm_segsz = in->shm_segsz;
1273 v.shm_nattch = in->shm_nattch;
1274 v.shm_cpid = in->shm_cpid;
1275 v.shm_lpid = in->shm_lpid;
1276 return copy_to_user(buf, &v, sizeof(v));
1280 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1283 memset(out, 0, sizeof(*out));
1284 if (version == IPC_64) {
1285 struct compat_shmid64_ds __user *p = buf;
1286 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1288 struct compat_shmid_ds __user *p = buf;
1289 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1293 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1295 struct ipc_namespace *ns;
1296 struct shmid64_ds sem64;
1297 int version = compat_ipc_parse_version(&cmd);
1300 ns = current->nsproxy->ipc_ns;
1302 if (cmd < 0 || shmid < 0)
1307 struct shminfo64 shminfo;
1308 err = shmctl_ipc_info(ns, &shminfo);
1311 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1316 struct shm_info shm_info;
1317 err = shmctl_shm_info(ns, &shm_info);
1320 if (put_compat_shm_info(&shm_info, uptr))
1327 err = shmctl_stat(ns, shmid, cmd, &sem64);
1330 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1335 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1339 return shmctl_down(ns, shmid, cmd, &sem64);
1342 return shmctl_do_lock(ns, shmid, cmd);
1350 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1352 return compat_ksys_shmctl(shmid, cmd, uptr);
1357 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1359 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1360 * "raddr" thing points to kernel space, and there has to be a wrapper around
1363 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1364 ulong *raddr, unsigned long shmlba)
1366 struct shmid_kernel *shp;
1367 unsigned long addr = (unsigned long)shmaddr;
1371 unsigned long flags = MAP_SHARED;
1374 struct ipc_namespace *ns;
1375 struct shm_file_data *sfd;
1378 unsigned long populate = 0;
1385 if (addr & (shmlba - 1)) {
1386 if (shmflg & SHM_RND) {
1387 addr &= ~(shmlba - 1); /* round down */
1390 * Ensure that the round-down is non-nil
1391 * when remapping. This can happen for
1392 * cases when addr < shmlba.
1394 if (!addr && (shmflg & SHM_REMAP))
1397 #ifndef __ARCH_FORCE_SHMLBA
1398 if (addr & ~PAGE_MASK)
1404 } else if ((shmflg & SHM_REMAP))
1407 if (shmflg & SHM_RDONLY) {
1410 f_mode = FMODE_READ;
1412 prot = PROT_READ | PROT_WRITE;
1413 acc_mode = S_IRUGO | S_IWUGO;
1414 f_mode = FMODE_READ | FMODE_WRITE;
1416 if (shmflg & SHM_EXEC) {
1418 acc_mode |= S_IXUGO;
1422 * We cannot rely on the fs check since SYSV IPC does have an
1423 * additional creator id...
1425 ns = current->nsproxy->ipc_ns;
1427 shp = shm_obtain_object_check(ns, shmid);
1434 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1437 err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1441 ipc_lock_object(&shp->shm_perm);
1443 /* check if shm_destroy() is tearing down shp */
1444 if (!ipc_valid_object(&shp->shm_perm)) {
1445 ipc_unlock_object(&shp->shm_perm);
1450 path = shp->shm_file->f_path;
1453 size = i_size_read(d_inode(path.dentry));
1454 ipc_unlock_object(&shp->shm_perm);
1458 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1464 file = alloc_file(&path, f_mode,
1465 is_file_hugepages(shp->shm_file) ?
1466 &shm_file_operations_huge :
1467 &shm_file_operations);
1468 err = PTR_ERR(file);
1475 file->private_data = sfd;
1476 file->f_mapping = shp->shm_file->f_mapping;
1477 sfd->id = shp->shm_perm.id;
1478 sfd->ns = get_ipc_ns(ns);
1480 * We need to take a reference to the real shm file to prevent the
1481 * pointer from becoming stale in cases where the lifetime of the outer
1482 * file extends beyond that of the shm segment. It's not usually
1483 * possible, but it can happen during remap_file_pages() emulation as
1484 * that unmaps the memory, then does ->mmap() via file reference only.
1485 * We'll deny the ->mmap() if the shm segment was since removed, but to
1486 * detect shm ID reuse we need to compare the file pointers.
1488 sfd->file = get_file(shp->shm_file);
1491 err = security_mmap_file(file, prot, flags);
1495 if (down_write_killable(¤t->mm->mmap_sem)) {
1500 if (addr && !(shmflg & SHM_REMAP)) {
1502 if (addr + size < addr)
1505 if (find_vma_intersection(current->mm, addr, addr + size))
1509 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1512 if (IS_ERR_VALUE(addr))
1515 up_write(¤t->mm->mmap_sem);
1517 mm_populate(addr, populate);
1523 down_write(&shm_ids(ns).rwsem);
1524 shp = shm_lock(ns, shmid);
1526 if (shm_may_destroy(ns, shp))
1527 shm_destroy(ns, shp);
1530 up_write(&shm_ids(ns).rwsem);
1539 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1544 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1547 force_successful_syscall_return();
1551 #ifdef CONFIG_COMPAT
1553 #ifndef COMPAT_SHMLBA
1554 #define COMPAT_SHMLBA SHMLBA
1557 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1562 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1565 force_successful_syscall_return();
1571 * detach and kill segment if marked destroyed.
1572 * The work is done in shm_close.
1574 long ksys_shmdt(char __user *shmaddr)
1576 struct mm_struct *mm = current->mm;
1577 struct vm_area_struct *vma;
1578 unsigned long addr = (unsigned long)shmaddr;
1579 int retval = -EINVAL;
1583 struct vm_area_struct *next;
1586 if (addr & ~PAGE_MASK)
1589 if (down_write_killable(&mm->mmap_sem))
1593 * This function tries to be smart and unmap shm segments that
1594 * were modified by partial mlock or munmap calls:
1595 * - It first determines the size of the shm segment that should be
1596 * unmapped: It searches for a vma that is backed by shm and that
1597 * started at address shmaddr. It records it's size and then unmaps
1599 * - Then it unmaps all shm vmas that started at shmaddr and that
1600 * are within the initially determined size and that are from the
1601 * same shm segment from which we determined the size.
1602 * Errors from do_munmap are ignored: the function only fails if
1603 * it's called with invalid parameters or if it's called to unmap
1604 * a part of a vma. Both calls in this function are for full vmas,
1605 * the parameters are directly copied from the vma itself and always
1606 * valid - therefore do_munmap cannot fail. (famous last words?)
1609 * If it had been mremap()'d, the starting address would not
1610 * match the usual checks anyway. So assume all vma's are
1611 * above the starting address given.
1613 vma = find_vma(mm, addr);
1617 next = vma->vm_next;
1620 * Check if the starting address would match, i.e. it's
1621 * a fragment created by mprotect() and/or munmap(), or it
1622 * otherwise it starts at this address with no hassles.
1624 if ((vma->vm_ops == &shm_vm_ops) &&
1625 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1628 * Record the file of the shm segment being
1629 * unmapped. With mremap(), someone could place
1630 * page from another segment but with equal offsets
1631 * in the range we are unmapping.
1633 file = vma->vm_file;
1634 size = i_size_read(file_inode(vma->vm_file));
1635 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1637 * We discovered the size of the shm segment, so
1638 * break out of here and fall through to the next
1639 * loop that uses the size information to stop
1640 * searching for matching vma's.
1650 * We need look no further than the maximum address a fragment
1651 * could possibly have landed at. Also cast things to loff_t to
1652 * prevent overflows and make comparisons vs. equal-width types.
1654 size = PAGE_ALIGN(size);
1655 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1656 next = vma->vm_next;
1658 /* finding a matching vma now does not alter retval */
1659 if ((vma->vm_ops == &shm_vm_ops) &&
1660 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1661 (vma->vm_file == file))
1662 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1666 #else /* CONFIG_MMU */
1667 /* under NOMMU conditions, the exact address to be destroyed must be
1670 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1671 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1677 up_write(&mm->mmap_sem);
1681 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1683 return ksys_shmdt(shmaddr);
1686 #ifdef CONFIG_PROC_FS
1687 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1689 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1690 struct user_namespace *user_ns = seq_user_ns(s);
1691 struct kern_ipc_perm *ipcp = it;
1692 struct shmid_kernel *shp;
1693 unsigned long rss = 0, swp = 0;
1695 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1696 shm_add_rss_swap(shp, &rss, &swp);
1698 #if BITS_PER_LONG <= 32
1699 #define SIZE_SPEC "%10lu"
1701 #define SIZE_SPEC "%21lu"
1705 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1706 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1707 SIZE_SPEC " " SIZE_SPEC "\n",
1712 pid_nr_ns(shp->shm_cprid, pid_ns),
1713 pid_nr_ns(shp->shm_lprid, pid_ns),
1715 from_kuid_munged(user_ns, shp->shm_perm.uid),
1716 from_kgid_munged(user_ns, shp->shm_perm.gid),
1717 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1718 from_kgid_munged(user_ns, shp->shm_perm.cgid),