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 shm_file_data {
53 struct ipc_namespace *ns;
55 const struct vm_operations_struct *vm_ops;
58 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
60 static const struct file_operations shm_file_operations;
61 static const struct vm_operations_struct shm_vm_ops;
63 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
65 #define shm_unlock(shp) \
66 ipc_unlock(&(shp)->shm_perm)
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);
73 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
76 int shm_init_ns(struct ipc_namespace *ns)
78 ns->shm_ctlmax = SHMMAX;
79 ns->shm_ctlall = SHMALL;
80 ns->shm_ctlmni = SHMMNI;
81 ns->shm_rmid_forced = 0;
83 return ipc_init_ids(&shm_ids(ns));
87 * Called with shm_ids.rwsem (writer) and the shp structure locked.
88 * Only shm_ids.rwsem remains locked on exit.
90 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
92 struct shmid_kernel *shp;
94 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
95 WARN_ON(ns != shp->ns);
97 if (shp->shm_nattch) {
98 shp->shm_perm.mode |= SHM_DEST;
99 /* Do not find it any more */
100 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
103 shm_destroy(ns, shp);
107 void shm_exit_ns(struct ipc_namespace *ns)
109 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
110 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
111 rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
115 static int __init ipc_ns_init(void)
117 const int err = shm_init_ns(&init_ipc_ns);
118 WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
122 pure_initcall(ipc_ns_init);
124 void __init shm_init(void)
126 ipc_init_proc_interface("sysvipc/shm",
127 #if BITS_PER_LONG <= 32
128 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
130 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
132 IPC_SHM_IDS, sysvipc_shm_proc_show);
135 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
137 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
140 return ERR_CAST(ipcp);
142 return container_of(ipcp, struct shmid_kernel, shm_perm);
145 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
147 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
150 return ERR_CAST(ipcp);
152 return container_of(ipcp, struct shmid_kernel, shm_perm);
156 * shm_lock_(check_) routines are called in the paths where the rwsem
157 * is not necessarily held.
159 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
161 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
164 * Callers of shm_lock() must validate the status of the returned ipc
165 * object pointer (as returned by ipc_lock()), and error out as
170 return container_of(ipcp, struct shmid_kernel, shm_perm);
173 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
176 ipc_lock_object(&ipcp->shm_perm);
179 static void shm_rcu_free(struct rcu_head *head)
181 struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
183 struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
185 security_shm_free(shp);
190 * It has to be called with shp locked.
191 * It must be called before ipc_rmid()
193 static inline void shm_clist_rm(struct shmid_kernel *shp)
195 struct task_struct *creator;
197 /* ensure that shm_creator does not disappear */
201 * A concurrent exit_shm may do a list_del_init() as well.
202 * Just do nothing if exit_shm already did the work
204 if (!list_empty(&shp->shm_clist)) {
206 * shp->shm_creator is guaranteed to be valid *only*
207 * if shp->shm_clist is not empty.
209 creator = shp->shm_creator;
213 * list_del_init() is a nop if the entry was already removed
216 list_del_init(&shp->shm_clist);
217 task_unlock(creator);
222 static inline void shm_rmid(struct shmid_kernel *s)
225 ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
229 static int __shm_open(struct vm_area_struct *vma)
231 struct file *file = vma->vm_file;
232 struct shm_file_data *sfd = shm_file_data(file);
233 struct shmid_kernel *shp;
235 shp = shm_lock(sfd->ns, sfd->id);
240 if (shp->shm_file != sfd->file) {
246 shp->shm_atim = ktime_get_real_seconds();
247 shp->shm_lprid = task_tgid_vnr(current);
253 /* This is called by fork, once for every shm attach. */
254 static void shm_open(struct vm_area_struct *vma)
256 int err = __shm_open(vma);
258 * We raced in the idr lookup or with shm_destroy().
259 * Either way, the ID is busted.
265 * shm_destroy - free the struct shmid_kernel
268 * @shp: struct to free
270 * It has to be called with shp and shm_ids.rwsem (writer) locked,
271 * but returns with shp unlocked and freed.
273 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
275 struct file *shm_file;
277 shm_file = shp->shm_file;
278 shp->shm_file = NULL;
279 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
282 if (!is_file_hugepages(shm_file))
283 shmem_lock(shm_file, 0, shp->mlock_user);
284 else if (shp->mlock_user)
285 user_shm_unlock(i_size_read(file_inode(shm_file)),
288 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
292 * shm_may_destroy - identifies whether shm segment should be destroyed now
294 * Returns true if and only if there are no active users of the segment and
295 * one of the following is true:
297 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
299 * 2) sysctl kernel.shm_rmid_forced is set to 1.
301 static bool shm_may_destroy(struct shmid_kernel *shp)
303 return (shp->shm_nattch == 0) &&
304 (shp->ns->shm_rmid_forced ||
305 (shp->shm_perm.mode & SHM_DEST));
309 * remove the attach descriptor vma.
310 * free memory for segment if it is marked destroyed.
311 * The descriptor has already been removed from the current->mm->mmap list
312 * and will later be kfree()d.
314 static void shm_close(struct vm_area_struct *vma)
316 struct file *file = vma->vm_file;
317 struct shm_file_data *sfd = shm_file_data(file);
318 struct shmid_kernel *shp;
319 struct ipc_namespace *ns = sfd->ns;
321 down_write(&shm_ids(ns).rwsem);
322 /* remove from the list of attaches of the shm segment */
323 shp = shm_lock(ns, sfd->id);
326 * We raced in the idr lookup or with shm_destroy().
327 * Either way, the ID is busted.
329 if (WARN_ON_ONCE(IS_ERR(shp)))
330 goto done; /* no-op */
332 shp->shm_lprid = task_tgid_vnr(current);
333 shp->shm_dtim = ktime_get_real_seconds();
335 if (shm_may_destroy(shp))
336 shm_destroy(ns, shp);
340 up_write(&shm_ids(ns).rwsem);
343 /* Called with ns->shm_ids(ns).rwsem locked */
344 static int shm_try_destroy_orphaned(int id, void *p, void *data)
346 struct ipc_namespace *ns = data;
347 struct kern_ipc_perm *ipcp = p;
348 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
351 * We want to destroy segments without users and with already
352 * exit'ed originating process.
354 * As shp->* are changed under rwsem, it's safe to skip shp locking.
356 if (!list_empty(&shp->shm_clist))
359 if (shm_may_destroy(shp)) {
360 shm_lock_by_ptr(shp);
361 shm_destroy(ns, shp);
366 void shm_destroy_orphaned(struct ipc_namespace *ns)
368 down_write(&shm_ids(ns).rwsem);
369 if (shm_ids(ns).in_use)
370 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
371 up_write(&shm_ids(ns).rwsem);
374 /* Locking assumes this will only be called with task == current */
375 void exit_shm(struct task_struct *task)
378 struct shmid_kernel *shp;
379 struct ipc_namespace *ns;
383 if (list_empty(&task->sysvshm.shm_clist)) {
388 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
392 * 1) Get pointer to the ipc namespace. It is worth to say
393 * that this pointer is guaranteed to be valid because
394 * shp lifetime is always shorter than namespace lifetime
395 * in which shp lives.
396 * We taken task_lock it means that shp won't be freed.
401 * 2) If kernel.shm_rmid_forced is not set then only keep track of
402 * which shmids are orphaned, so that a later set of the sysctl
405 if (!ns->shm_rmid_forced)
406 goto unlink_continue;
409 * 3) get a reference to the namespace.
410 * The refcount could be already 0. If it is 0, then
411 * the shm objects will be free by free_ipc_work().
413 ns = get_ipc_ns_not_zero(ns);
416 list_del_init(&shp->shm_clist);
422 * 4) get a reference to shp.
423 * This cannot fail: shm_clist_rm() is called before
424 * ipc_rmid(), thus the refcount cannot be 0.
426 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
429 * 5) unlink the shm segment from the list of segments
430 * created by current.
431 * This must be done last. After unlinking,
432 * only the refcounts obtained above prevent IPC_RMID
433 * from destroying the segment or the namespace.
435 list_del_init(&shp->shm_clist);
440 * 6) we have all references
441 * Thus lock & if needed destroy shp.
443 down_write(&shm_ids(ns).rwsem);
444 shm_lock_by_ptr(shp);
446 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
447 * safe to call ipc_rcu_putref here
449 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
451 if (ipc_valid_object(&shp->shm_perm)) {
452 if (shm_may_destroy(shp))
453 shm_destroy(ns, shp);
458 * Someone else deleted the shp from namespace
459 * idr/kht while we have waited.
460 * Just unlock and continue.
465 up_write(&shm_ids(ns).rwsem);
466 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
470 static int shm_fault(struct vm_fault *vmf)
472 struct file *file = vmf->vma->vm_file;
473 struct shm_file_data *sfd = shm_file_data(file);
475 return sfd->vm_ops->fault(vmf);
478 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
480 struct file *file = vma->vm_file;
481 struct shm_file_data *sfd = shm_file_data(file);
483 if (sfd->vm_ops && sfd->vm_ops->split)
484 return sfd->vm_ops->split(vma, addr);
490 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
492 struct file *file = vma->vm_file;
493 struct shm_file_data *sfd = shm_file_data(file);
496 if (sfd->vm_ops->set_policy)
497 err = sfd->vm_ops->set_policy(vma, new);
501 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
504 struct file *file = vma->vm_file;
505 struct shm_file_data *sfd = shm_file_data(file);
506 struct mempolicy *pol = NULL;
508 if (sfd->vm_ops->get_policy)
509 pol = sfd->vm_ops->get_policy(vma, addr);
510 else if (vma->vm_policy)
511 pol = vma->vm_policy;
517 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
519 struct shm_file_data *sfd = shm_file_data(file);
523 * In case of remap_file_pages() emulation, the file can represent an
524 * IPC ID that was removed, and possibly even reused by another shm
525 * segment already. Propagate this case as an error to caller.
527 ret = __shm_open(vma);
531 ret = call_mmap(sfd->file, vma);
536 sfd->vm_ops = vma->vm_ops;
538 WARN_ON(!sfd->vm_ops->fault);
540 vma->vm_ops = &shm_vm_ops;
544 static int shm_release(struct inode *ino, struct file *file)
546 struct shm_file_data *sfd = shm_file_data(file);
550 shm_file_data(file) = NULL;
555 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
557 struct shm_file_data *sfd = shm_file_data(file);
559 if (!sfd->file->f_op->fsync)
561 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
564 static long shm_fallocate(struct file *file, int mode, loff_t offset,
567 struct shm_file_data *sfd = shm_file_data(file);
569 if (!sfd->file->f_op->fallocate)
571 return sfd->file->f_op->fallocate(file, mode, offset, len);
574 static unsigned long shm_get_unmapped_area(struct file *file,
575 unsigned long addr, unsigned long len, unsigned long pgoff,
578 struct shm_file_data *sfd = shm_file_data(file);
580 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
584 static const struct file_operations shm_file_operations = {
587 .release = shm_release,
588 .get_unmapped_area = shm_get_unmapped_area,
589 .llseek = noop_llseek,
590 .fallocate = shm_fallocate,
594 * shm_file_operations_huge is now identical to shm_file_operations,
595 * but we keep it distinct for the sake of is_file_shm_hugepages().
597 static const struct file_operations shm_file_operations_huge = {
600 .release = shm_release,
601 .get_unmapped_area = shm_get_unmapped_area,
602 .llseek = noop_llseek,
603 .fallocate = shm_fallocate,
606 bool is_file_shm_hugepages(struct file *file)
608 return file->f_op == &shm_file_operations_huge;
611 static const struct vm_operations_struct shm_vm_ops = {
612 .open = shm_open, /* callback for a new vm-area open */
613 .close = shm_close, /* callback for when the vm-area is released */
616 #if defined(CONFIG_NUMA)
617 .set_policy = shm_set_policy,
618 .get_policy = shm_get_policy,
623 * newseg - Create a new shared memory segment
625 * @params: ptr to the structure that contains key, size and shmflg
627 * Called with shm_ids.rwsem held as a writer.
629 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
631 key_t key = params->key;
632 int shmflg = params->flg;
633 size_t size = params->u.size;
635 struct shmid_kernel *shp;
636 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
639 vm_flags_t acctflag = 0;
641 if (size < SHMMIN || size > ns->shm_ctlmax)
644 if (numpages << PAGE_SHIFT < size)
647 if (ns->shm_tot + numpages < ns->shm_tot ||
648 ns->shm_tot + numpages > ns->shm_ctlall)
651 shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
655 shp->shm_perm.key = key;
656 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
657 shp->mlock_user = NULL;
659 shp->shm_perm.security = NULL;
660 error = security_shm_alloc(shp);
666 sprintf(name, "SYSV%08x", key);
667 if (shmflg & SHM_HUGETLB) {
671 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
676 hugesize = ALIGN(size, huge_page_size(hs));
678 /* hugetlb_file_setup applies strict accounting */
679 if (shmflg & SHM_NORESERVE)
680 acctflag = VM_NORESERVE;
681 file = hugetlb_file_setup(name, hugesize, acctflag,
682 &shp->mlock_user, HUGETLB_SHMFS_INODE,
683 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
686 * Do not allow no accounting for OVERCOMMIT_NEVER, even
689 if ((shmflg & SHM_NORESERVE) &&
690 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
691 acctflag = VM_NORESERVE;
692 file = shmem_kernel_file_setup(name, size, acctflag);
694 error = PTR_ERR(file);
698 shp->shm_cprid = task_tgid_vnr(current);
700 shp->shm_atim = shp->shm_dtim = 0;
701 shp->shm_ctim = ktime_get_real_seconds();
702 shp->shm_segsz = size;
704 shp->shm_file = file;
705 shp->shm_creator = current;
707 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
714 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
715 task_unlock(current);
718 * shmid gets reported as "inode#" in /proc/pid/maps.
719 * proc-ps tools use this. Changing this will break them.
721 file_inode(file)->i_ino = shp->shm_perm.id;
723 ns->shm_tot += numpages;
724 error = shp->shm_perm.id;
726 ipc_unlock_object(&shp->shm_perm);
731 if (is_file_hugepages(file) && shp->mlock_user)
732 user_shm_unlock(size, shp->mlock_user);
735 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
740 * Called with shm_ids.rwsem and ipcp locked.
742 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
744 struct shmid_kernel *shp;
746 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
747 return security_shm_associate(shp, shmflg);
751 * Called with shm_ids.rwsem and ipcp locked.
753 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
754 struct ipc_params *params)
756 struct shmid_kernel *shp;
758 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
759 if (shp->shm_segsz < params->u.size)
765 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
767 struct ipc_namespace *ns;
768 static const struct ipc_ops shm_ops = {
770 .associate = shm_security,
771 .more_checks = shm_more_checks,
773 struct ipc_params shm_params;
775 ns = current->nsproxy->ipc_ns;
777 shm_params.key = key;
778 shm_params.flg = shmflg;
779 shm_params.u.size = size;
781 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
784 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
788 return copy_to_user(buf, in, sizeof(*in));
793 memset(&out, 0, sizeof(out));
794 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
795 out.shm_segsz = in->shm_segsz;
796 out.shm_atime = in->shm_atime;
797 out.shm_dtime = in->shm_dtime;
798 out.shm_ctime = in->shm_ctime;
799 out.shm_cpid = in->shm_cpid;
800 out.shm_lpid = in->shm_lpid;
801 out.shm_nattch = in->shm_nattch;
803 return copy_to_user(buf, &out, sizeof(out));
810 static inline unsigned long
811 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
815 if (copy_from_user(out, buf, sizeof(*out)))
820 struct shmid_ds tbuf_old;
822 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
825 out->shm_perm.uid = tbuf_old.shm_perm.uid;
826 out->shm_perm.gid = tbuf_old.shm_perm.gid;
827 out->shm_perm.mode = tbuf_old.shm_perm.mode;
836 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
840 return copy_to_user(buf, in, sizeof(*in));
845 if (in->shmmax > INT_MAX)
846 out.shmmax = INT_MAX;
848 out.shmmax = (int)in->shmmax;
850 out.shmmin = in->shmmin;
851 out.shmmni = in->shmmni;
852 out.shmseg = in->shmseg;
853 out.shmall = in->shmall;
855 return copy_to_user(buf, &out, sizeof(out));
863 * Calculate and add used RSS and swap pages of a shm.
864 * Called with shm_ids.rwsem held as a reader
866 static void shm_add_rss_swap(struct shmid_kernel *shp,
867 unsigned long *rss_add, unsigned long *swp_add)
871 inode = file_inode(shp->shm_file);
873 if (is_file_hugepages(shp->shm_file)) {
874 struct address_space *mapping = inode->i_mapping;
875 struct hstate *h = hstate_file(shp->shm_file);
876 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
879 struct shmem_inode_info *info = SHMEM_I(inode);
881 spin_lock_irq(&info->lock);
882 *rss_add += inode->i_mapping->nrpages;
883 *swp_add += info->swapped;
884 spin_unlock_irq(&info->lock);
886 *rss_add += inode->i_mapping->nrpages;
892 * Called with shm_ids.rwsem held as a reader
894 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
903 in_use = shm_ids(ns).in_use;
905 for (total = 0, next_id = 0; total < in_use; next_id++) {
906 struct kern_ipc_perm *ipc;
907 struct shmid_kernel *shp;
909 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
912 shp = container_of(ipc, struct shmid_kernel, shm_perm);
914 shm_add_rss_swap(shp, rss, swp);
921 * This function handles some shmctl commands which require the rwsem
922 * to be held in write mode.
923 * NOTE: no locks must be held, the rwsem is taken inside this function.
925 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
926 struct shmid64_ds *shmid64)
928 struct kern_ipc_perm *ipcp;
929 struct shmid_kernel *shp;
932 down_write(&shm_ids(ns).rwsem);
935 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
936 &shmid64->shm_perm, 0);
942 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
944 err = security_shm_shmctl(shp, cmd);
950 ipc_lock_object(&shp->shm_perm);
951 /* do_shm_rmid unlocks the ipc object and rcu */
952 do_shm_rmid(ns, ipcp);
955 ipc_lock_object(&shp->shm_perm);
956 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
959 shp->shm_ctim = ktime_get_real_seconds();
967 ipc_unlock_object(&shp->shm_perm);
971 up_write(&shm_ids(ns).rwsem);
975 static int shmctl_ipc_info(struct ipc_namespace *ns,
976 struct shminfo64 *shminfo)
978 int err = security_shm_shmctl(NULL, IPC_INFO);
980 memset(shminfo, 0, sizeof(*shminfo));
981 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
982 shminfo->shmmax = ns->shm_ctlmax;
983 shminfo->shmall = ns->shm_ctlall;
984 shminfo->shmmin = SHMMIN;
985 down_read(&shm_ids(ns).rwsem);
986 err = ipc_get_maxid(&shm_ids(ns));
987 up_read(&shm_ids(ns).rwsem);
994 static int shmctl_shm_info(struct ipc_namespace *ns,
995 struct shm_info *shm_info)
997 int err = security_shm_shmctl(NULL, SHM_INFO);
999 memset(shm_info, 0, sizeof(*shm_info));
1000 down_read(&shm_ids(ns).rwsem);
1001 shm_info->used_ids = shm_ids(ns).in_use;
1002 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1003 shm_info->shm_tot = ns->shm_tot;
1004 shm_info->swap_attempts = 0;
1005 shm_info->swap_successes = 0;
1006 err = ipc_get_maxid(&shm_ids(ns));
1007 up_read(&shm_ids(ns).rwsem);
1014 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1015 int cmd, struct shmid64_ds *tbuf)
1017 struct shmid_kernel *shp;
1022 if (cmd == SHM_STAT) {
1023 shp = shm_obtain_object(ns, shmid);
1028 result = shp->shm_perm.id;
1030 shp = shm_obtain_object_check(ns, shmid);
1039 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1042 err = security_shm_shmctl(shp, cmd);
1046 memset(tbuf, 0, sizeof(*tbuf));
1047 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1048 tbuf->shm_segsz = shp->shm_segsz;
1049 tbuf->shm_atime = shp->shm_atim;
1050 tbuf->shm_dtime = shp->shm_dtim;
1051 tbuf->shm_ctime = shp->shm_ctim;
1052 tbuf->shm_cpid = shp->shm_cprid;
1053 tbuf->shm_lpid = shp->shm_lprid;
1054 tbuf->shm_nattch = shp->shm_nattch;
1063 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1065 struct shmid_kernel *shp;
1066 struct file *shm_file;
1070 shp = shm_obtain_object_check(ns, shmid);
1076 audit_ipc_obj(&(shp->shm_perm));
1077 err = security_shm_shmctl(shp, cmd);
1081 ipc_lock_object(&shp->shm_perm);
1083 /* check if shm_destroy() is tearing down shp */
1084 if (!ipc_valid_object(&shp->shm_perm)) {
1089 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1090 kuid_t euid = current_euid();
1092 if (!uid_eq(euid, shp->shm_perm.uid) &&
1093 !uid_eq(euid, shp->shm_perm.cuid)) {
1097 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1103 shm_file = shp->shm_file;
1104 if (is_file_hugepages(shm_file))
1107 if (cmd == SHM_LOCK) {
1108 struct user_struct *user = current_user();
1110 err = shmem_lock(shm_file, 1, user);
1111 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1112 shp->shm_perm.mode |= SHM_LOCKED;
1113 shp->mlock_user = user;
1119 if (!(shp->shm_perm.mode & SHM_LOCKED))
1121 shmem_lock(shm_file, 0, shp->mlock_user);
1122 shp->shm_perm.mode &= ~SHM_LOCKED;
1123 shp->mlock_user = NULL;
1125 ipc_unlock_object(&shp->shm_perm);
1127 shmem_unlock_mapping(shm_file->f_mapping);
1133 ipc_unlock_object(&shp->shm_perm);
1139 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1142 struct ipc_namespace *ns;
1143 struct shmid64_ds sem64;
1145 if (cmd < 0 || shmid < 0)
1148 version = ipc_parse_version(&cmd);
1149 ns = current->nsproxy->ipc_ns;
1153 struct shminfo64 shminfo;
1154 err = shmctl_ipc_info(ns, &shminfo);
1157 if (copy_shminfo_to_user(buf, &shminfo, version))
1162 struct shm_info shm_info;
1163 err = shmctl_shm_info(ns, &shm_info);
1166 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1172 err = shmctl_stat(ns, shmid, cmd, &sem64);
1175 if (copy_shmid_to_user(buf, &sem64, version))
1180 if (copy_shmid_from_user(&sem64, buf, version))
1184 return shmctl_down(ns, shmid, cmd, &sem64);
1187 return shmctl_do_lock(ns, shmid, cmd);
1193 #ifdef CONFIG_COMPAT
1195 struct compat_shmid_ds {
1196 struct compat_ipc_perm shm_perm;
1198 compat_time_t shm_atime;
1199 compat_time_t shm_dtime;
1200 compat_time_t shm_ctime;
1201 compat_ipc_pid_t shm_cpid;
1202 compat_ipc_pid_t shm_lpid;
1203 unsigned short shm_nattch;
1204 unsigned short shm_unused;
1205 compat_uptr_t shm_unused2;
1206 compat_uptr_t shm_unused3;
1209 struct compat_shminfo64 {
1210 compat_ulong_t shmmax;
1211 compat_ulong_t shmmin;
1212 compat_ulong_t shmmni;
1213 compat_ulong_t shmseg;
1214 compat_ulong_t shmall;
1215 compat_ulong_t __unused1;
1216 compat_ulong_t __unused2;
1217 compat_ulong_t __unused3;
1218 compat_ulong_t __unused4;
1221 struct compat_shm_info {
1222 compat_int_t used_ids;
1223 compat_ulong_t shm_tot, shm_rss, shm_swp;
1224 compat_ulong_t swap_attempts, swap_successes;
1227 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1230 if (in->shmmax > INT_MAX)
1231 in->shmmax = INT_MAX;
1232 if (version == IPC_64) {
1233 struct compat_shminfo64 info;
1234 memset(&info, 0, sizeof(info));
1235 info.shmmax = in->shmmax;
1236 info.shmmin = in->shmmin;
1237 info.shmmni = in->shmmni;
1238 info.shmseg = in->shmseg;
1239 info.shmall = in->shmall;
1240 return copy_to_user(buf, &info, sizeof(info));
1242 struct shminfo info;
1243 memset(&info, 0, sizeof(info));
1244 info.shmmax = in->shmmax;
1245 info.shmmin = in->shmmin;
1246 info.shmmni = in->shmmni;
1247 info.shmseg = in->shmseg;
1248 info.shmall = in->shmall;
1249 return copy_to_user(buf, &info, sizeof(info));
1253 static int put_compat_shm_info(struct shm_info *ip,
1254 struct compat_shm_info __user *uip)
1256 struct compat_shm_info info;
1258 memset(&info, 0, sizeof(info));
1259 info.used_ids = ip->used_ids;
1260 info.shm_tot = ip->shm_tot;
1261 info.shm_rss = ip->shm_rss;
1262 info.shm_swp = ip->shm_swp;
1263 info.swap_attempts = ip->swap_attempts;
1264 info.swap_successes = ip->swap_successes;
1265 return copy_to_user(uip, &info, sizeof(info));
1268 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1271 if (version == IPC_64) {
1272 struct compat_shmid64_ds v;
1273 memset(&v, 0, sizeof(v));
1274 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1275 v.shm_atime = in->shm_atime;
1276 v.shm_dtime = in->shm_dtime;
1277 v.shm_ctime = in->shm_ctime;
1278 v.shm_segsz = in->shm_segsz;
1279 v.shm_nattch = in->shm_nattch;
1280 v.shm_cpid = in->shm_cpid;
1281 v.shm_lpid = in->shm_lpid;
1282 return copy_to_user(buf, &v, sizeof(v));
1284 struct compat_shmid_ds v;
1285 memset(&v, 0, sizeof(v));
1286 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1287 v.shm_perm.key = in->shm_perm.key;
1288 v.shm_atime = in->shm_atime;
1289 v.shm_dtime = in->shm_dtime;
1290 v.shm_ctime = in->shm_ctime;
1291 v.shm_segsz = in->shm_segsz;
1292 v.shm_nattch = in->shm_nattch;
1293 v.shm_cpid = in->shm_cpid;
1294 v.shm_lpid = in->shm_lpid;
1295 return copy_to_user(buf, &v, sizeof(v));
1299 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1302 memset(out, 0, sizeof(*out));
1303 if (version == IPC_64) {
1304 struct compat_shmid64_ds *p = buf;
1305 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1307 struct compat_shmid_ds *p = buf;
1308 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1312 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1314 struct ipc_namespace *ns;
1315 struct shmid64_ds sem64;
1316 int version = compat_ipc_parse_version(&cmd);
1319 ns = current->nsproxy->ipc_ns;
1321 if (cmd < 0 || shmid < 0)
1326 struct shminfo64 shminfo;
1327 err = shmctl_ipc_info(ns, &shminfo);
1330 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1335 struct shm_info shm_info;
1336 err = shmctl_shm_info(ns, &shm_info);
1339 if (put_compat_shm_info(&shm_info, uptr))
1345 err = shmctl_stat(ns, shmid, cmd, &sem64);
1348 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1353 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1357 return shmctl_down(ns, shmid, cmd, &sem64);
1360 return shmctl_do_lock(ns, shmid, cmd);
1370 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1372 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1373 * "raddr" thing points to kernel space, and there has to be a wrapper around
1376 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1377 ulong *raddr, unsigned long shmlba)
1379 struct shmid_kernel *shp;
1380 unsigned long addr = (unsigned long)shmaddr;
1384 unsigned long flags = MAP_SHARED;
1387 struct ipc_namespace *ns;
1388 struct shm_file_data *sfd;
1391 unsigned long populate = 0;
1398 if (addr & (shmlba - 1)) {
1399 if (shmflg & SHM_RND) {
1400 addr &= ~(shmlba - 1); /* round down */
1403 * Ensure that the round-down is non-nil
1404 * when remapping. This can happen for
1405 * cases when addr < shmlba.
1407 if (!addr && (shmflg & SHM_REMAP))
1410 #ifndef __ARCH_FORCE_SHMLBA
1411 if (addr & ~PAGE_MASK)
1417 } else if ((shmflg & SHM_REMAP))
1420 if (shmflg & SHM_RDONLY) {
1423 f_mode = FMODE_READ;
1425 prot = PROT_READ | PROT_WRITE;
1426 acc_mode = S_IRUGO | S_IWUGO;
1427 f_mode = FMODE_READ | FMODE_WRITE;
1429 if (shmflg & SHM_EXEC) {
1431 acc_mode |= S_IXUGO;
1435 * We cannot rely on the fs check since SYSV IPC does have an
1436 * additional creator id...
1438 ns = current->nsproxy->ipc_ns;
1440 shp = shm_obtain_object_check(ns, shmid);
1447 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1450 err = security_shm_shmat(shp, shmaddr, shmflg);
1454 ipc_lock_object(&shp->shm_perm);
1456 /* check if shm_destroy() is tearing down shp */
1457 if (!ipc_valid_object(&shp->shm_perm)) {
1458 ipc_unlock_object(&shp->shm_perm);
1463 path = shp->shm_file->f_path;
1466 size = i_size_read(d_inode(path.dentry));
1467 ipc_unlock_object(&shp->shm_perm);
1471 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1477 file = alloc_file(&path, f_mode,
1478 is_file_hugepages(shp->shm_file) ?
1479 &shm_file_operations_huge :
1480 &shm_file_operations);
1481 err = PTR_ERR(file);
1488 file->private_data = sfd;
1489 file->f_mapping = shp->shm_file->f_mapping;
1490 sfd->id = shp->shm_perm.id;
1491 sfd->ns = get_ipc_ns(ns);
1493 * We need to take a reference to the real shm file to prevent the
1494 * pointer from becoming stale in cases where the lifetime of the outer
1495 * file extends beyond that of the shm segment. It's not usually
1496 * possible, but it can happen during remap_file_pages() emulation as
1497 * that unmaps the memory, then does ->mmap() via file reference only.
1498 * We'll deny the ->mmap() if the shm segment was since removed, but to
1499 * detect shm ID reuse we need to compare the file pointers.
1501 sfd->file = get_file(shp->shm_file);
1504 err = security_mmap_file(file, prot, flags);
1508 if (down_write_killable(¤t->mm->mmap_sem)) {
1513 if (addr && !(shmflg & SHM_REMAP)) {
1515 if (addr + size < addr)
1518 if (find_vma_intersection(current->mm, addr, addr + size))
1522 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1525 if (IS_ERR_VALUE(addr))
1528 up_write(¤t->mm->mmap_sem);
1530 mm_populate(addr, populate);
1536 down_write(&shm_ids(ns).rwsem);
1537 shp = shm_lock(ns, shmid);
1540 if (shm_may_destroy(shp))
1541 shm_destroy(ns, shp);
1544 up_write(&shm_ids(ns).rwsem);
1553 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1558 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1561 force_successful_syscall_return();
1565 #ifdef CONFIG_COMPAT
1567 #ifndef COMPAT_SHMLBA
1568 #define COMPAT_SHMLBA SHMLBA
1571 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1576 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1579 force_successful_syscall_return();
1585 * detach and kill segment if marked destroyed.
1586 * The work is done in shm_close.
1588 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1590 struct mm_struct *mm = current->mm;
1591 struct vm_area_struct *vma;
1592 unsigned long addr = (unsigned long)shmaddr;
1593 int retval = -EINVAL;
1597 struct vm_area_struct *next;
1600 if (addr & ~PAGE_MASK)
1603 if (down_write_killable(&mm->mmap_sem))
1607 * This function tries to be smart and unmap shm segments that
1608 * were modified by partial mlock or munmap calls:
1609 * - It first determines the size of the shm segment that should be
1610 * unmapped: It searches for a vma that is backed by shm and that
1611 * started at address shmaddr. It records it's size and then unmaps
1613 * - Then it unmaps all shm vmas that started at shmaddr and that
1614 * are within the initially determined size and that are from the
1615 * same shm segment from which we determined the size.
1616 * Errors from do_munmap are ignored: the function only fails if
1617 * it's called with invalid parameters or if it's called to unmap
1618 * a part of a vma. Both calls in this function are for full vmas,
1619 * the parameters are directly copied from the vma itself and always
1620 * valid - therefore do_munmap cannot fail. (famous last words?)
1623 * If it had been mremap()'d, the starting address would not
1624 * match the usual checks anyway. So assume all vma's are
1625 * above the starting address given.
1627 vma = find_vma(mm, addr);
1631 next = vma->vm_next;
1634 * Check if the starting address would match, i.e. it's
1635 * a fragment created by mprotect() and/or munmap(), or it
1636 * otherwise it starts at this address with no hassles.
1638 if ((vma->vm_ops == &shm_vm_ops) &&
1639 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1642 * Record the file of the shm segment being
1643 * unmapped. With mremap(), someone could place
1644 * page from another segment but with equal offsets
1645 * in the range we are unmapping.
1647 file = vma->vm_file;
1648 size = i_size_read(file_inode(vma->vm_file));
1649 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1651 * We discovered the size of the shm segment, so
1652 * break out of here and fall through to the next
1653 * loop that uses the size information to stop
1654 * searching for matching vma's.
1664 * We need look no further than the maximum address a fragment
1665 * could possibly have landed at. Also cast things to loff_t to
1666 * prevent overflows and make comparisons vs. equal-width types.
1668 size = PAGE_ALIGN(size);
1669 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1670 next = vma->vm_next;
1672 /* finding a matching vma now does not alter retval */
1673 if ((vma->vm_ops == &shm_vm_ops) &&
1674 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1675 (vma->vm_file == file))
1676 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1680 #else /* CONFIG_MMU */
1681 /* under NOMMU conditions, the exact address to be destroyed must be
1684 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1685 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1691 up_write(&mm->mmap_sem);
1695 #ifdef CONFIG_PROC_FS
1696 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1698 struct user_namespace *user_ns = seq_user_ns(s);
1699 struct kern_ipc_perm *ipcp = it;
1700 struct shmid_kernel *shp;
1701 unsigned long rss = 0, swp = 0;
1703 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1704 shm_add_rss_swap(shp, &rss, &swp);
1706 #if BITS_PER_LONG <= 32
1707 #define SIZE_SPEC "%10lu"
1709 #define SIZE_SPEC "%21lu"
1713 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1714 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1715 SIZE_SPEC " " SIZE_SPEC "\n",
1723 from_kuid_munged(user_ns, shp->shm_perm.uid),
1724 from_kgid_munged(user_ns, shp->shm_perm.gid),
1725 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1726 from_kgid_munged(user_ns, shp->shm_perm.cgid),