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);
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);
102 shm_destroy(ns, shp);
106 void shm_exit_ns(struct ipc_namespace *ns)
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);
114 static int __init ipc_ns_init(void)
116 const int err = shm_init_ns(&init_ipc_ns);
117 WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
121 pure_initcall(ipc_ns_init);
123 void __init shm_init(void)
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",
129 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
131 IPC_SHM_IDS, sysvipc_shm_proc_show);
134 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
136 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
139 return ERR_CAST(ipcp);
141 return container_of(ipcp, struct shmid_kernel, shm_perm);
144 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
146 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
149 return ERR_CAST(ipcp);
151 return container_of(ipcp, struct shmid_kernel, shm_perm);
155 * shm_lock_(check_) routines are called in the paths where the rwsem
156 * is not necessarily held.
158 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
160 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
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
169 return container_of(ipcp, struct shmid_kernel, shm_perm);
172 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
175 ipc_lock_object(&ipcp->shm_perm);
178 static void shm_rcu_free(struct rcu_head *head)
180 struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
182 struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
184 security_shm_free(shp);
188 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
190 list_del(&s->shm_clist);
191 ipc_rmid(&shm_ids(ns), &s->shm_perm);
195 static int __shm_open(struct vm_area_struct *vma)
197 struct file *file = vma->vm_file;
198 struct shm_file_data *sfd = shm_file_data(file);
199 struct shmid_kernel *shp;
201 shp = shm_lock(sfd->ns, sfd->id);
206 shp->shm_atim = ktime_get_real_seconds();
207 shp->shm_lprid = task_tgid_vnr(current);
213 /* This is called by fork, once for every shm attach. */
214 static void shm_open(struct vm_area_struct *vma)
216 int err = __shm_open(vma);
218 * We raced in the idr lookup or with shm_destroy().
219 * Either way, the ID is busted.
225 * shm_destroy - free the struct shmid_kernel
228 * @shp: struct to free
230 * It has to be called with shp and shm_ids.rwsem (writer) locked,
231 * but returns with shp unlocked and freed.
233 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
235 struct file *shm_file;
237 shm_file = shp->shm_file;
238 shp->shm_file = NULL;
239 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
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)),
248 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
252 * shm_may_destroy - identifies whether shm segment should be destroyed now
254 * Returns true if and only if there are no active users of the segment and
255 * one of the following is true:
257 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
259 * 2) sysctl kernel.shm_rmid_forced is set to 1.
261 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
263 return (shp->shm_nattch == 0) &&
264 (ns->shm_rmid_forced ||
265 (shp->shm_perm.mode & SHM_DEST));
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.
274 static void shm_close(struct vm_area_struct *vma)
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;
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);
286 * We raced in the idr lookup or with shm_destroy().
287 * Either way, the ID is busted.
289 if (WARN_ON_ONCE(IS_ERR(shp)))
290 goto done; /* no-op */
292 shp->shm_lprid = task_tgid_vnr(current);
293 shp->shm_dtim = ktime_get_real_seconds();
295 if (shm_may_destroy(ns, shp))
296 shm_destroy(ns, shp);
300 up_write(&shm_ids(ns).rwsem);
303 /* Called with ns->shm_ids(ns).rwsem locked */
304 static int shm_try_destroy_orphaned(int id, void *p, void *data)
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);
311 * We want to destroy segments without users and with already
312 * exit'ed originating process.
314 * As shp->* are changed under rwsem, it's safe to skip shp locking.
316 if (shp->shm_creator != NULL)
319 if (shm_may_destroy(ns, shp)) {
320 shm_lock_by_ptr(shp);
321 shm_destroy(ns, shp);
326 void shm_destroy_orphaned(struct ipc_namespace *ns)
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);
334 /* Locking assumes this will only be called with task == current */
335 void exit_shm(struct task_struct *task)
337 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
338 struct shmid_kernel *shp, *n;
340 if (list_empty(&task->sysvshm.shm_clist))
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
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;
353 * Only under read lock but we are only called on current
354 * so no entry on the list will be shared.
356 list_del(&task->sysvshm.shm_clist);
357 up_read(&shm_ids(ns).rwsem);
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.
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;
370 if (shm_may_destroy(ns, shp)) {
371 shm_lock_by_ptr(shp);
372 shm_destroy(ns, shp);
376 /* Remove the list head from any segments still attached. */
377 list_del(&task->sysvshm.shm_clist);
378 up_write(&shm_ids(ns).rwsem);
381 static int shm_fault(struct vm_fault *vmf)
383 struct file *file = vmf->vma->vm_file;
384 struct shm_file_data *sfd = shm_file_data(file);
386 return sfd->vm_ops->fault(vmf);
389 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
391 struct file *file = vma->vm_file;
392 struct shm_file_data *sfd = shm_file_data(file);
394 if (sfd->vm_ops && sfd->vm_ops->split)
395 return sfd->vm_ops->split(vma, addr);
401 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
403 struct file *file = vma->vm_file;
404 struct shm_file_data *sfd = shm_file_data(file);
407 if (sfd->vm_ops->set_policy)
408 err = sfd->vm_ops->set_policy(vma, new);
412 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
415 struct file *file = vma->vm_file;
416 struct shm_file_data *sfd = shm_file_data(file);
417 struct mempolicy *pol = NULL;
419 if (sfd->vm_ops->get_policy)
420 pol = sfd->vm_ops->get_policy(vma, addr);
421 else if (vma->vm_policy)
422 pol = vma->vm_policy;
428 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
430 struct shm_file_data *sfd = shm_file_data(file);
434 * In case of remap_file_pages() emulation, the file can represent
435 * removed IPC ID: propogate shm_lock() error to caller.
437 ret = __shm_open(vma);
441 ret = call_mmap(sfd->file, vma);
446 sfd->vm_ops = vma->vm_ops;
448 WARN_ON(!sfd->vm_ops->fault);
450 vma->vm_ops = &shm_vm_ops;
454 static int shm_release(struct inode *ino, struct file *file)
456 struct shm_file_data *sfd = shm_file_data(file);
459 shm_file_data(file) = NULL;
464 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
466 struct shm_file_data *sfd = shm_file_data(file);
468 if (!sfd->file->f_op->fsync)
470 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
473 static long shm_fallocate(struct file *file, int mode, loff_t offset,
476 struct shm_file_data *sfd = shm_file_data(file);
478 if (!sfd->file->f_op->fallocate)
480 return sfd->file->f_op->fallocate(file, mode, offset, len);
483 static unsigned long shm_get_unmapped_area(struct file *file,
484 unsigned long addr, unsigned long len, unsigned long pgoff,
487 struct shm_file_data *sfd = shm_file_data(file);
489 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
493 static const struct file_operations shm_file_operations = {
496 .release = shm_release,
497 .get_unmapped_area = shm_get_unmapped_area,
498 .llseek = noop_llseek,
499 .fallocate = shm_fallocate,
503 * shm_file_operations_huge is now identical to shm_file_operations,
504 * but we keep it distinct for the sake of is_file_shm_hugepages().
506 static const struct file_operations shm_file_operations_huge = {
509 .release = shm_release,
510 .get_unmapped_area = shm_get_unmapped_area,
511 .llseek = noop_llseek,
512 .fallocate = shm_fallocate,
515 bool is_file_shm_hugepages(struct file *file)
517 return file->f_op == &shm_file_operations_huge;
520 static const struct vm_operations_struct shm_vm_ops = {
521 .open = shm_open, /* callback for a new vm-area open */
522 .close = shm_close, /* callback for when the vm-area is released */
525 #if defined(CONFIG_NUMA)
526 .set_policy = shm_set_policy,
527 .get_policy = shm_get_policy,
532 * newseg - Create a new shared memory segment
534 * @params: ptr to the structure that contains key, size and shmflg
536 * Called with shm_ids.rwsem held as a writer.
538 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
540 key_t key = params->key;
541 int shmflg = params->flg;
542 size_t size = params->u.size;
544 struct shmid_kernel *shp;
545 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
548 vm_flags_t acctflag = 0;
550 if (size < SHMMIN || size > ns->shm_ctlmax)
553 if (numpages << PAGE_SHIFT < size)
556 if (ns->shm_tot + numpages < ns->shm_tot ||
557 ns->shm_tot + numpages > ns->shm_ctlall)
560 shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
564 shp->shm_perm.key = key;
565 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
566 shp->mlock_user = NULL;
568 shp->shm_perm.security = NULL;
569 error = security_shm_alloc(shp);
575 sprintf(name, "SYSV%08x", key);
576 if (shmflg & SHM_HUGETLB) {
580 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
585 hugesize = ALIGN(size, huge_page_size(hs));
587 /* hugetlb_file_setup applies strict accounting */
588 if (shmflg & SHM_NORESERVE)
589 acctflag = VM_NORESERVE;
590 file = hugetlb_file_setup(name, hugesize, acctflag,
591 &shp->mlock_user, HUGETLB_SHMFS_INODE,
592 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
595 * Do not allow no accounting for OVERCOMMIT_NEVER, even
598 if ((shmflg & SHM_NORESERVE) &&
599 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
600 acctflag = VM_NORESERVE;
601 file = shmem_kernel_file_setup(name, size, acctflag);
603 error = PTR_ERR(file);
607 shp->shm_cprid = task_tgid_vnr(current);
609 shp->shm_atim = shp->shm_dtim = 0;
610 shp->shm_ctim = ktime_get_real_seconds();
611 shp->shm_segsz = size;
613 shp->shm_file = file;
614 shp->shm_creator = current;
616 /* ipc_addid() locks shp upon success. */
617 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
621 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
624 * shmid gets reported as "inode#" in /proc/pid/maps.
625 * proc-ps tools use this. Changing this will break them.
627 file_inode(file)->i_ino = shp->shm_perm.id;
629 ns->shm_tot += numpages;
630 error = shp->shm_perm.id;
632 ipc_unlock_object(&shp->shm_perm);
637 if (is_file_hugepages(file) && shp->mlock_user)
638 user_shm_unlock(size, shp->mlock_user);
641 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
646 * Called with shm_ids.rwsem and ipcp locked.
648 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
650 struct shmid_kernel *shp;
652 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
653 return security_shm_associate(shp, shmflg);
657 * Called with shm_ids.rwsem and ipcp locked.
659 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
660 struct ipc_params *params)
662 struct shmid_kernel *shp;
664 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
665 if (shp->shm_segsz < params->u.size)
671 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
673 struct ipc_namespace *ns;
674 static const struct ipc_ops shm_ops = {
676 .associate = shm_security,
677 .more_checks = shm_more_checks,
679 struct ipc_params shm_params;
681 ns = current->nsproxy->ipc_ns;
683 shm_params.key = key;
684 shm_params.flg = shmflg;
685 shm_params.u.size = size;
687 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
690 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
694 return copy_to_user(buf, in, sizeof(*in));
699 memset(&out, 0, sizeof(out));
700 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
701 out.shm_segsz = in->shm_segsz;
702 out.shm_atime = in->shm_atime;
703 out.shm_dtime = in->shm_dtime;
704 out.shm_ctime = in->shm_ctime;
705 out.shm_cpid = in->shm_cpid;
706 out.shm_lpid = in->shm_lpid;
707 out.shm_nattch = in->shm_nattch;
709 return copy_to_user(buf, &out, sizeof(out));
716 static inline unsigned long
717 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
721 if (copy_from_user(out, buf, sizeof(*out)))
726 struct shmid_ds tbuf_old;
728 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
731 out->shm_perm.uid = tbuf_old.shm_perm.uid;
732 out->shm_perm.gid = tbuf_old.shm_perm.gid;
733 out->shm_perm.mode = tbuf_old.shm_perm.mode;
742 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
746 return copy_to_user(buf, in, sizeof(*in));
751 if (in->shmmax > INT_MAX)
752 out.shmmax = INT_MAX;
754 out.shmmax = (int)in->shmmax;
756 out.shmmin = in->shmmin;
757 out.shmmni = in->shmmni;
758 out.shmseg = in->shmseg;
759 out.shmall = in->shmall;
761 return copy_to_user(buf, &out, sizeof(out));
769 * Calculate and add used RSS and swap pages of a shm.
770 * Called with shm_ids.rwsem held as a reader
772 static void shm_add_rss_swap(struct shmid_kernel *shp,
773 unsigned long *rss_add, unsigned long *swp_add)
777 inode = file_inode(shp->shm_file);
779 if (is_file_hugepages(shp->shm_file)) {
780 struct address_space *mapping = inode->i_mapping;
781 struct hstate *h = hstate_file(shp->shm_file);
782 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
785 struct shmem_inode_info *info = SHMEM_I(inode);
787 spin_lock_irq(&info->lock);
788 *rss_add += inode->i_mapping->nrpages;
789 *swp_add += info->swapped;
790 spin_unlock_irq(&info->lock);
792 *rss_add += inode->i_mapping->nrpages;
798 * Called with shm_ids.rwsem held as a reader
800 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
809 in_use = shm_ids(ns).in_use;
811 for (total = 0, next_id = 0; total < in_use; next_id++) {
812 struct kern_ipc_perm *ipc;
813 struct shmid_kernel *shp;
815 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
818 shp = container_of(ipc, struct shmid_kernel, shm_perm);
820 shm_add_rss_swap(shp, rss, swp);
827 * This function handles some shmctl commands which require the rwsem
828 * to be held in write mode.
829 * NOTE: no locks must be held, the rwsem is taken inside this function.
831 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
832 struct shmid64_ds *shmid64)
834 struct kern_ipc_perm *ipcp;
835 struct shmid_kernel *shp;
838 down_write(&shm_ids(ns).rwsem);
841 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
842 &shmid64->shm_perm, 0);
848 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
850 err = security_shm_shmctl(shp, cmd);
856 ipc_lock_object(&shp->shm_perm);
857 /* do_shm_rmid unlocks the ipc object and rcu */
858 do_shm_rmid(ns, ipcp);
861 ipc_lock_object(&shp->shm_perm);
862 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
865 shp->shm_ctim = ktime_get_real_seconds();
873 ipc_unlock_object(&shp->shm_perm);
877 up_write(&shm_ids(ns).rwsem);
881 static int shmctl_ipc_info(struct ipc_namespace *ns,
882 struct shminfo64 *shminfo)
884 int err = security_shm_shmctl(NULL, IPC_INFO);
886 memset(shminfo, 0, sizeof(*shminfo));
887 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
888 shminfo->shmmax = ns->shm_ctlmax;
889 shminfo->shmall = ns->shm_ctlall;
890 shminfo->shmmin = SHMMIN;
891 down_read(&shm_ids(ns).rwsem);
892 err = ipc_get_maxid(&shm_ids(ns));
893 up_read(&shm_ids(ns).rwsem);
900 static int shmctl_shm_info(struct ipc_namespace *ns,
901 struct shm_info *shm_info)
903 int err = security_shm_shmctl(NULL, SHM_INFO);
905 memset(shm_info, 0, sizeof(*shm_info));
906 down_read(&shm_ids(ns).rwsem);
907 shm_info->used_ids = shm_ids(ns).in_use;
908 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
909 shm_info->shm_tot = ns->shm_tot;
910 shm_info->swap_attempts = 0;
911 shm_info->swap_successes = 0;
912 err = ipc_get_maxid(&shm_ids(ns));
913 up_read(&shm_ids(ns).rwsem);
920 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
921 int cmd, struct shmid64_ds *tbuf)
923 struct shmid_kernel *shp;
927 memset(tbuf, 0, sizeof(*tbuf));
930 if (cmd == SHM_STAT) {
931 shp = shm_obtain_object(ns, shmid);
936 id = shp->shm_perm.id;
938 shp = shm_obtain_object_check(ns, shmid);
946 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
949 err = security_shm_shmctl(shp, cmd);
953 ipc_lock_object(&shp->shm_perm);
955 if (!ipc_valid_object(&shp->shm_perm)) {
956 ipc_unlock_object(&shp->shm_perm);
961 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
962 tbuf->shm_segsz = shp->shm_segsz;
963 tbuf->shm_atime = shp->shm_atim;
964 tbuf->shm_dtime = shp->shm_dtim;
965 tbuf->shm_ctime = shp->shm_ctim;
966 tbuf->shm_cpid = shp->shm_cprid;
967 tbuf->shm_lpid = shp->shm_lprid;
968 tbuf->shm_nattch = shp->shm_nattch;
970 ipc_unlock_object(&shp->shm_perm);
979 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
981 struct shmid_kernel *shp;
982 struct file *shm_file;
986 shp = shm_obtain_object_check(ns, shmid);
992 audit_ipc_obj(&(shp->shm_perm));
993 err = security_shm_shmctl(shp, cmd);
997 ipc_lock_object(&shp->shm_perm);
999 /* check if shm_destroy() is tearing down shp */
1000 if (!ipc_valid_object(&shp->shm_perm)) {
1005 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1006 kuid_t euid = current_euid();
1008 if (!uid_eq(euid, shp->shm_perm.uid) &&
1009 !uid_eq(euid, shp->shm_perm.cuid)) {
1013 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1019 shm_file = shp->shm_file;
1020 if (is_file_hugepages(shm_file))
1023 if (cmd == SHM_LOCK) {
1024 struct user_struct *user = current_user();
1026 err = shmem_lock(shm_file, 1, user);
1027 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1028 shp->shm_perm.mode |= SHM_LOCKED;
1029 shp->mlock_user = user;
1035 if (!(shp->shm_perm.mode & SHM_LOCKED))
1037 shmem_lock(shm_file, 0, shp->mlock_user);
1038 shp->shm_perm.mode &= ~SHM_LOCKED;
1039 shp->mlock_user = NULL;
1041 ipc_unlock_object(&shp->shm_perm);
1043 shmem_unlock_mapping(shm_file->f_mapping);
1049 ipc_unlock_object(&shp->shm_perm);
1055 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1058 struct ipc_namespace *ns;
1059 struct shmid64_ds sem64;
1061 if (cmd < 0 || shmid < 0)
1064 version = ipc_parse_version(&cmd);
1065 ns = current->nsproxy->ipc_ns;
1069 struct shminfo64 shminfo;
1070 err = shmctl_ipc_info(ns, &shminfo);
1073 if (copy_shminfo_to_user(buf, &shminfo, version))
1078 struct shm_info shm_info;
1079 err = shmctl_shm_info(ns, &shm_info);
1082 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1088 err = shmctl_stat(ns, shmid, cmd, &sem64);
1091 if (copy_shmid_to_user(buf, &sem64, version))
1096 if (copy_shmid_from_user(&sem64, buf, version))
1100 return shmctl_down(ns, shmid, cmd, &sem64);
1103 return shmctl_do_lock(ns, shmid, cmd);
1109 #ifdef CONFIG_COMPAT
1111 struct compat_shmid_ds {
1112 struct compat_ipc_perm shm_perm;
1114 compat_time_t shm_atime;
1115 compat_time_t shm_dtime;
1116 compat_time_t shm_ctime;
1117 compat_ipc_pid_t shm_cpid;
1118 compat_ipc_pid_t shm_lpid;
1119 unsigned short shm_nattch;
1120 unsigned short shm_unused;
1121 compat_uptr_t shm_unused2;
1122 compat_uptr_t shm_unused3;
1125 struct compat_shminfo64 {
1126 compat_ulong_t shmmax;
1127 compat_ulong_t shmmin;
1128 compat_ulong_t shmmni;
1129 compat_ulong_t shmseg;
1130 compat_ulong_t shmall;
1131 compat_ulong_t __unused1;
1132 compat_ulong_t __unused2;
1133 compat_ulong_t __unused3;
1134 compat_ulong_t __unused4;
1137 struct compat_shm_info {
1138 compat_int_t used_ids;
1139 compat_ulong_t shm_tot, shm_rss, shm_swp;
1140 compat_ulong_t swap_attempts, swap_successes;
1143 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1146 if (in->shmmax > INT_MAX)
1147 in->shmmax = INT_MAX;
1148 if (version == IPC_64) {
1149 struct compat_shminfo64 info;
1150 memset(&info, 0, sizeof(info));
1151 info.shmmax = in->shmmax;
1152 info.shmmin = in->shmmin;
1153 info.shmmni = in->shmmni;
1154 info.shmseg = in->shmseg;
1155 info.shmall = in->shmall;
1156 return copy_to_user(buf, &info, sizeof(info));
1158 struct shminfo info;
1159 memset(&info, 0, sizeof(info));
1160 info.shmmax = in->shmmax;
1161 info.shmmin = in->shmmin;
1162 info.shmmni = in->shmmni;
1163 info.shmseg = in->shmseg;
1164 info.shmall = in->shmall;
1165 return copy_to_user(buf, &info, sizeof(info));
1169 static int put_compat_shm_info(struct shm_info *ip,
1170 struct compat_shm_info __user *uip)
1172 struct compat_shm_info info;
1174 memset(&info, 0, sizeof(info));
1175 info.used_ids = ip->used_ids;
1176 info.shm_tot = ip->shm_tot;
1177 info.shm_rss = ip->shm_rss;
1178 info.shm_swp = ip->shm_swp;
1179 info.swap_attempts = ip->swap_attempts;
1180 info.swap_successes = ip->swap_successes;
1181 return copy_to_user(uip, &info, sizeof(info));
1184 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1187 if (version == IPC_64) {
1188 struct compat_shmid64_ds v;
1189 memset(&v, 0, sizeof(v));
1190 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1191 v.shm_atime = in->shm_atime;
1192 v.shm_dtime = in->shm_dtime;
1193 v.shm_ctime = in->shm_ctime;
1194 v.shm_segsz = in->shm_segsz;
1195 v.shm_nattch = in->shm_nattch;
1196 v.shm_cpid = in->shm_cpid;
1197 v.shm_lpid = in->shm_lpid;
1198 return copy_to_user(buf, &v, sizeof(v));
1200 struct compat_shmid_ds v;
1201 memset(&v, 0, sizeof(v));
1202 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1203 v.shm_perm.key = in->shm_perm.key;
1204 v.shm_atime = in->shm_atime;
1205 v.shm_dtime = in->shm_dtime;
1206 v.shm_ctime = in->shm_ctime;
1207 v.shm_segsz = in->shm_segsz;
1208 v.shm_nattch = in->shm_nattch;
1209 v.shm_cpid = in->shm_cpid;
1210 v.shm_lpid = in->shm_lpid;
1211 return copy_to_user(buf, &v, sizeof(v));
1215 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1218 memset(out, 0, sizeof(*out));
1219 if (version == IPC_64) {
1220 struct compat_shmid64_ds __user *p = buf;
1221 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1223 struct compat_shmid_ds __user *p = buf;
1224 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1228 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1230 struct ipc_namespace *ns;
1231 struct shmid64_ds sem64;
1232 int version = compat_ipc_parse_version(&cmd);
1235 ns = current->nsproxy->ipc_ns;
1237 if (cmd < 0 || shmid < 0)
1242 struct shminfo64 shminfo;
1243 err = shmctl_ipc_info(ns, &shminfo);
1246 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1251 struct shm_info shm_info;
1252 err = shmctl_shm_info(ns, &shm_info);
1255 if (put_compat_shm_info(&shm_info, uptr))
1261 err = shmctl_stat(ns, shmid, cmd, &sem64);
1264 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1269 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1273 return shmctl_down(ns, shmid, cmd, &sem64);
1276 return shmctl_do_lock(ns, shmid, cmd);
1286 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1288 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1289 * "raddr" thing points to kernel space, and there has to be a wrapper around
1292 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1293 ulong *raddr, unsigned long shmlba)
1295 struct shmid_kernel *shp;
1296 unsigned long addr = (unsigned long)shmaddr;
1300 unsigned long flags = MAP_SHARED;
1303 struct ipc_namespace *ns;
1304 struct shm_file_data *sfd;
1307 unsigned long populate = 0;
1314 if (addr & (shmlba - 1)) {
1316 * Round down to the nearest multiple of shmlba.
1317 * For sane do_mmap_pgoff() parameters, avoid
1318 * round downs that trigger nil-page and MAP_FIXED.
1320 if ((shmflg & SHM_RND) && addr >= shmlba)
1321 addr &= ~(shmlba - 1);
1323 #ifndef __ARCH_FORCE_SHMLBA
1324 if (addr & ~PAGE_MASK)
1330 } else if ((shmflg & SHM_REMAP))
1333 if (shmflg & SHM_RDONLY) {
1336 f_mode = FMODE_READ;
1338 prot = PROT_READ | PROT_WRITE;
1339 acc_mode = S_IRUGO | S_IWUGO;
1340 f_mode = FMODE_READ | FMODE_WRITE;
1342 if (shmflg & SHM_EXEC) {
1344 acc_mode |= S_IXUGO;
1348 * We cannot rely on the fs check since SYSV IPC does have an
1349 * additional creator id...
1351 ns = current->nsproxy->ipc_ns;
1353 shp = shm_obtain_object_check(ns, shmid);
1360 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1363 err = security_shm_shmat(shp, shmaddr, shmflg);
1367 ipc_lock_object(&shp->shm_perm);
1369 /* check if shm_destroy() is tearing down shp */
1370 if (!ipc_valid_object(&shp->shm_perm)) {
1371 ipc_unlock_object(&shp->shm_perm);
1376 path = shp->shm_file->f_path;
1379 size = i_size_read(d_inode(path.dentry));
1380 ipc_unlock_object(&shp->shm_perm);
1384 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1390 file = alloc_file(&path, f_mode,
1391 is_file_hugepages(shp->shm_file) ?
1392 &shm_file_operations_huge :
1393 &shm_file_operations);
1394 err = PTR_ERR(file);
1401 file->private_data = sfd;
1402 file->f_mapping = shp->shm_file->f_mapping;
1403 sfd->id = shp->shm_perm.id;
1404 sfd->ns = get_ipc_ns(ns);
1405 sfd->file = shp->shm_file;
1408 err = security_mmap_file(file, prot, flags);
1412 if (down_write_killable(¤t->mm->mmap_sem)) {
1417 if (addr && !(shmflg & SHM_REMAP)) {
1419 if (addr + size < addr)
1422 if (find_vma_intersection(current->mm, addr, addr + size))
1426 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1429 if (IS_ERR_VALUE(addr))
1432 up_write(¤t->mm->mmap_sem);
1434 mm_populate(addr, populate);
1440 down_write(&shm_ids(ns).rwsem);
1441 shp = shm_lock(ns, shmid);
1443 if (shm_may_destroy(ns, shp))
1444 shm_destroy(ns, shp);
1447 up_write(&shm_ids(ns).rwsem);
1456 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1461 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1464 force_successful_syscall_return();
1468 #ifdef CONFIG_COMPAT
1470 #ifndef COMPAT_SHMLBA
1471 #define COMPAT_SHMLBA SHMLBA
1474 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1479 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1482 force_successful_syscall_return();
1488 * detach and kill segment if marked destroyed.
1489 * The work is done in shm_close.
1491 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1493 struct mm_struct *mm = current->mm;
1494 struct vm_area_struct *vma;
1495 unsigned long addr = (unsigned long)shmaddr;
1496 int retval = -EINVAL;
1500 struct vm_area_struct *next;
1503 if (addr & ~PAGE_MASK)
1506 if (down_write_killable(&mm->mmap_sem))
1510 * This function tries to be smart and unmap shm segments that
1511 * were modified by partial mlock or munmap calls:
1512 * - It first determines the size of the shm segment that should be
1513 * unmapped: It searches for a vma that is backed by shm and that
1514 * started at address shmaddr. It records it's size and then unmaps
1516 * - Then it unmaps all shm vmas that started at shmaddr and that
1517 * are within the initially determined size and that are from the
1518 * same shm segment from which we determined the size.
1519 * Errors from do_munmap are ignored: the function only fails if
1520 * it's called with invalid parameters or if it's called to unmap
1521 * a part of a vma. Both calls in this function are for full vmas,
1522 * the parameters are directly copied from the vma itself and always
1523 * valid - therefore do_munmap cannot fail. (famous last words?)
1526 * If it had been mremap()'d, the starting address would not
1527 * match the usual checks anyway. So assume all vma's are
1528 * above the starting address given.
1530 vma = find_vma(mm, addr);
1534 next = vma->vm_next;
1537 * Check if the starting address would match, i.e. it's
1538 * a fragment created by mprotect() and/or munmap(), or it
1539 * otherwise it starts at this address with no hassles.
1541 if ((vma->vm_ops == &shm_vm_ops) &&
1542 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1545 * Record the file of the shm segment being
1546 * unmapped. With mremap(), someone could place
1547 * page from another segment but with equal offsets
1548 * in the range we are unmapping.
1550 file = vma->vm_file;
1551 size = i_size_read(file_inode(vma->vm_file));
1552 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1554 * We discovered the size of the shm segment, so
1555 * break out of here and fall through to the next
1556 * loop that uses the size information to stop
1557 * searching for matching vma's.
1567 * We need look no further than the maximum address a fragment
1568 * could possibly have landed at. Also cast things to loff_t to
1569 * prevent overflows and make comparisons vs. equal-width types.
1571 size = PAGE_ALIGN(size);
1572 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1573 next = vma->vm_next;
1575 /* finding a matching vma now does not alter retval */
1576 if ((vma->vm_ops == &shm_vm_ops) &&
1577 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1578 (vma->vm_file == file))
1579 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1583 #else /* CONFIG_MMU */
1584 /* under NOMMU conditions, the exact address to be destroyed must be
1587 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1588 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1594 up_write(&mm->mmap_sem);
1598 #ifdef CONFIG_PROC_FS
1599 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1601 struct user_namespace *user_ns = seq_user_ns(s);
1602 struct kern_ipc_perm *ipcp = it;
1603 struct shmid_kernel *shp;
1604 unsigned long rss = 0, swp = 0;
1606 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1607 shm_add_rss_swap(shp, &rss, &swp);
1609 #if BITS_PER_LONG <= 32
1610 #define SIZE_SPEC "%10lu"
1612 #define SIZE_SPEC "%21lu"
1616 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1617 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1618 SIZE_SPEC " " SIZE_SPEC "\n",
1626 from_kuid_munged(user_ns, shp->shm_perm.uid),
1627 from_kgid_munged(user_ns, shp->shm_perm.gid),
1628 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1629 from_kgid_munged(user_ns, shp->shm_perm.cgid),