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);
390 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
392 struct file *file = vma->vm_file;
393 struct shm_file_data *sfd = shm_file_data(file);
396 if (sfd->vm_ops->set_policy)
397 err = sfd->vm_ops->set_policy(vma, new);
401 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
404 struct file *file = vma->vm_file;
405 struct shm_file_data *sfd = shm_file_data(file);
406 struct mempolicy *pol = NULL;
408 if (sfd->vm_ops->get_policy)
409 pol = sfd->vm_ops->get_policy(vma, addr);
410 else if (vma->vm_policy)
411 pol = vma->vm_policy;
417 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
419 struct shm_file_data *sfd = shm_file_data(file);
423 * In case of remap_file_pages() emulation, the file can represent
424 * removed IPC ID: propogate shm_lock() error to caller.
426 ret = __shm_open(vma);
430 ret = call_mmap(sfd->file, vma);
435 sfd->vm_ops = vma->vm_ops;
437 WARN_ON(!sfd->vm_ops->fault);
439 vma->vm_ops = &shm_vm_ops;
443 static int shm_release(struct inode *ino, struct file *file)
445 struct shm_file_data *sfd = shm_file_data(file);
448 shm_file_data(file) = NULL;
453 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
455 struct shm_file_data *sfd = shm_file_data(file);
457 if (!sfd->file->f_op->fsync)
459 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
462 static long shm_fallocate(struct file *file, int mode, loff_t offset,
465 struct shm_file_data *sfd = shm_file_data(file);
467 if (!sfd->file->f_op->fallocate)
469 return sfd->file->f_op->fallocate(file, mode, offset, len);
472 static unsigned long shm_get_unmapped_area(struct file *file,
473 unsigned long addr, unsigned long len, unsigned long pgoff,
476 struct shm_file_data *sfd = shm_file_data(file);
478 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
482 static const struct file_operations shm_file_operations = {
485 .release = shm_release,
486 .get_unmapped_area = shm_get_unmapped_area,
487 .llseek = noop_llseek,
488 .fallocate = shm_fallocate,
492 * shm_file_operations_huge is now identical to shm_file_operations,
493 * but we keep it distinct for the sake of is_file_shm_hugepages().
495 static const struct file_operations shm_file_operations_huge = {
498 .release = shm_release,
499 .get_unmapped_area = shm_get_unmapped_area,
500 .llseek = noop_llseek,
501 .fallocate = shm_fallocate,
504 bool is_file_shm_hugepages(struct file *file)
506 return file->f_op == &shm_file_operations_huge;
509 static const struct vm_operations_struct shm_vm_ops = {
510 .open = shm_open, /* callback for a new vm-area open */
511 .close = shm_close, /* callback for when the vm-area is released */
513 #if defined(CONFIG_NUMA)
514 .set_policy = shm_set_policy,
515 .get_policy = shm_get_policy,
520 * newseg - Create a new shared memory segment
522 * @params: ptr to the structure that contains key, size and shmflg
524 * Called with shm_ids.rwsem held as a writer.
526 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
528 key_t key = params->key;
529 int shmflg = params->flg;
530 size_t size = params->u.size;
532 struct shmid_kernel *shp;
533 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
536 vm_flags_t acctflag = 0;
538 if (size < SHMMIN || size > ns->shm_ctlmax)
541 if (numpages << PAGE_SHIFT < size)
544 if (ns->shm_tot + numpages < ns->shm_tot ||
545 ns->shm_tot + numpages > ns->shm_ctlall)
548 shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
552 shp->shm_perm.key = key;
553 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
554 shp->mlock_user = NULL;
556 shp->shm_perm.security = NULL;
557 error = security_shm_alloc(shp);
563 sprintf(name, "SYSV%08x", key);
564 if (shmflg & SHM_HUGETLB) {
568 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
573 hugesize = ALIGN(size, huge_page_size(hs));
575 /* hugetlb_file_setup applies strict accounting */
576 if (shmflg & SHM_NORESERVE)
577 acctflag = VM_NORESERVE;
578 file = hugetlb_file_setup(name, hugesize, acctflag,
579 &shp->mlock_user, HUGETLB_SHMFS_INODE,
580 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
583 * Do not allow no accounting for OVERCOMMIT_NEVER, even
586 if ((shmflg & SHM_NORESERVE) &&
587 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
588 acctflag = VM_NORESERVE;
589 file = shmem_kernel_file_setup(name, size, acctflag);
591 error = PTR_ERR(file);
595 shp->shm_cprid = task_tgid_vnr(current);
597 shp->shm_atim = shp->shm_dtim = 0;
598 shp->shm_ctim = ktime_get_real_seconds();
599 shp->shm_segsz = size;
601 shp->shm_file = file;
602 shp->shm_creator = current;
604 /* ipc_addid() locks shp upon success. */
605 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
609 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
612 * shmid gets reported as "inode#" in /proc/pid/maps.
613 * proc-ps tools use this. Changing this will break them.
615 file_inode(file)->i_ino = shp->shm_perm.id;
617 ns->shm_tot += numpages;
618 error = shp->shm_perm.id;
620 ipc_unlock_object(&shp->shm_perm);
625 if (is_file_hugepages(file) && shp->mlock_user)
626 user_shm_unlock(size, shp->mlock_user);
629 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
634 * Called with shm_ids.rwsem and ipcp locked.
636 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
638 struct shmid_kernel *shp;
640 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
641 return security_shm_associate(shp, shmflg);
645 * Called with shm_ids.rwsem and ipcp locked.
647 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
648 struct ipc_params *params)
650 struct shmid_kernel *shp;
652 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
653 if (shp->shm_segsz < params->u.size)
659 long ksys_shmget(key_t key, size_t size, int shmflg)
661 struct ipc_namespace *ns;
662 static const struct ipc_ops shm_ops = {
664 .associate = shm_security,
665 .more_checks = shm_more_checks,
667 struct ipc_params shm_params;
669 ns = current->nsproxy->ipc_ns;
671 shm_params.key = key;
672 shm_params.flg = shmflg;
673 shm_params.u.size = size;
675 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
678 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
680 return ksys_shmget(key, size, shmflg);
683 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
687 return copy_to_user(buf, in, sizeof(*in));
692 memset(&out, 0, sizeof(out));
693 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
694 out.shm_segsz = in->shm_segsz;
695 out.shm_atime = in->shm_atime;
696 out.shm_dtime = in->shm_dtime;
697 out.shm_ctime = in->shm_ctime;
698 out.shm_cpid = in->shm_cpid;
699 out.shm_lpid = in->shm_lpid;
700 out.shm_nattch = in->shm_nattch;
702 return copy_to_user(buf, &out, sizeof(out));
709 static inline unsigned long
710 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
714 if (copy_from_user(out, buf, sizeof(*out)))
719 struct shmid_ds tbuf_old;
721 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
724 out->shm_perm.uid = tbuf_old.shm_perm.uid;
725 out->shm_perm.gid = tbuf_old.shm_perm.gid;
726 out->shm_perm.mode = tbuf_old.shm_perm.mode;
735 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
739 return copy_to_user(buf, in, sizeof(*in));
744 if (in->shmmax > INT_MAX)
745 out.shmmax = INT_MAX;
747 out.shmmax = (int)in->shmmax;
749 out.shmmin = in->shmmin;
750 out.shmmni = in->shmmni;
751 out.shmseg = in->shmseg;
752 out.shmall = in->shmall;
754 return copy_to_user(buf, &out, sizeof(out));
762 * Calculate and add used RSS and swap pages of a shm.
763 * Called with shm_ids.rwsem held as a reader
765 static void shm_add_rss_swap(struct shmid_kernel *shp,
766 unsigned long *rss_add, unsigned long *swp_add)
770 inode = file_inode(shp->shm_file);
772 if (is_file_hugepages(shp->shm_file)) {
773 struct address_space *mapping = inode->i_mapping;
774 struct hstate *h = hstate_file(shp->shm_file);
775 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
778 struct shmem_inode_info *info = SHMEM_I(inode);
780 spin_lock_irq(&info->lock);
781 *rss_add += inode->i_mapping->nrpages;
782 *swp_add += info->swapped;
783 spin_unlock_irq(&info->lock);
785 *rss_add += inode->i_mapping->nrpages;
791 * Called with shm_ids.rwsem held as a reader
793 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
802 in_use = shm_ids(ns).in_use;
804 for (total = 0, next_id = 0; total < in_use; next_id++) {
805 struct kern_ipc_perm *ipc;
806 struct shmid_kernel *shp;
808 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
811 shp = container_of(ipc, struct shmid_kernel, shm_perm);
813 shm_add_rss_swap(shp, rss, swp);
820 * This function handles some shmctl commands which require the rwsem
821 * to be held in write mode.
822 * NOTE: no locks must be held, the rwsem is taken inside this function.
824 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
825 struct shmid64_ds *shmid64)
827 struct kern_ipc_perm *ipcp;
828 struct shmid_kernel *shp;
831 down_write(&shm_ids(ns).rwsem);
834 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
835 &shmid64->shm_perm, 0);
841 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
843 err = security_shm_shmctl(shp, cmd);
849 ipc_lock_object(&shp->shm_perm);
850 /* do_shm_rmid unlocks the ipc object and rcu */
851 do_shm_rmid(ns, ipcp);
854 ipc_lock_object(&shp->shm_perm);
855 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
858 shp->shm_ctim = ktime_get_real_seconds();
866 ipc_unlock_object(&shp->shm_perm);
870 up_write(&shm_ids(ns).rwsem);
874 static int shmctl_ipc_info(struct ipc_namespace *ns,
875 struct shminfo64 *shminfo)
877 int err = security_shm_shmctl(NULL, IPC_INFO);
879 memset(shminfo, 0, sizeof(*shminfo));
880 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
881 shminfo->shmmax = ns->shm_ctlmax;
882 shminfo->shmall = ns->shm_ctlall;
883 shminfo->shmmin = SHMMIN;
884 down_read(&shm_ids(ns).rwsem);
885 err = ipc_get_maxid(&shm_ids(ns));
886 up_read(&shm_ids(ns).rwsem);
893 static int shmctl_shm_info(struct ipc_namespace *ns,
894 struct shm_info *shm_info)
896 int err = security_shm_shmctl(NULL, SHM_INFO);
898 memset(shm_info, 0, sizeof(*shm_info));
899 down_read(&shm_ids(ns).rwsem);
900 shm_info->used_ids = shm_ids(ns).in_use;
901 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
902 shm_info->shm_tot = ns->shm_tot;
903 shm_info->swap_attempts = 0;
904 shm_info->swap_successes = 0;
905 err = ipc_get_maxid(&shm_ids(ns));
906 up_read(&shm_ids(ns).rwsem);
913 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
914 int cmd, struct shmid64_ds *tbuf)
916 struct shmid_kernel *shp;
920 memset(tbuf, 0, sizeof(*tbuf));
923 if (cmd == SHM_STAT) {
924 shp = shm_obtain_object(ns, shmid);
929 id = shp->shm_perm.id;
931 shp = shm_obtain_object_check(ns, shmid);
939 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
942 err = security_shm_shmctl(shp, cmd);
946 ipc_lock_object(&shp->shm_perm);
948 if (!ipc_valid_object(&shp->shm_perm)) {
949 ipc_unlock_object(&shp->shm_perm);
954 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
955 tbuf->shm_segsz = shp->shm_segsz;
956 tbuf->shm_atime = shp->shm_atim;
957 tbuf->shm_dtime = shp->shm_dtim;
958 tbuf->shm_ctime = shp->shm_ctim;
959 tbuf->shm_cpid = shp->shm_cprid;
960 tbuf->shm_lpid = shp->shm_lprid;
961 tbuf->shm_nattch = shp->shm_nattch;
963 ipc_unlock_object(&shp->shm_perm);
972 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
974 struct shmid_kernel *shp;
975 struct file *shm_file;
979 shp = shm_obtain_object_check(ns, shmid);
985 audit_ipc_obj(&(shp->shm_perm));
986 err = security_shm_shmctl(shp, cmd);
990 ipc_lock_object(&shp->shm_perm);
992 /* check if shm_destroy() is tearing down shp */
993 if (!ipc_valid_object(&shp->shm_perm)) {
998 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
999 kuid_t euid = current_euid();
1001 if (!uid_eq(euid, shp->shm_perm.uid) &&
1002 !uid_eq(euid, shp->shm_perm.cuid)) {
1006 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1012 shm_file = shp->shm_file;
1013 if (is_file_hugepages(shm_file))
1016 if (cmd == SHM_LOCK) {
1017 struct user_struct *user = current_user();
1019 err = shmem_lock(shm_file, 1, user);
1020 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1021 shp->shm_perm.mode |= SHM_LOCKED;
1022 shp->mlock_user = user;
1028 if (!(shp->shm_perm.mode & SHM_LOCKED))
1030 shmem_lock(shm_file, 0, shp->mlock_user);
1031 shp->shm_perm.mode &= ~SHM_LOCKED;
1032 shp->mlock_user = NULL;
1034 ipc_unlock_object(&shp->shm_perm);
1036 shmem_unlock_mapping(shm_file->f_mapping);
1042 ipc_unlock_object(&shp->shm_perm);
1048 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1051 struct ipc_namespace *ns;
1052 struct shmid64_ds sem64;
1054 if (cmd < 0 || shmid < 0)
1057 version = ipc_parse_version(&cmd);
1058 ns = current->nsproxy->ipc_ns;
1062 struct shminfo64 shminfo;
1063 err = shmctl_ipc_info(ns, &shminfo);
1066 if (copy_shminfo_to_user(buf, &shminfo, version))
1071 struct shm_info shm_info;
1072 err = shmctl_shm_info(ns, &shm_info);
1075 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1081 err = shmctl_stat(ns, shmid, cmd, &sem64);
1084 if (copy_shmid_to_user(buf, &sem64, version))
1089 if (copy_shmid_from_user(&sem64, buf, version))
1093 return shmctl_down(ns, shmid, cmd, &sem64);
1096 return shmctl_do_lock(ns, shmid, cmd);
1102 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1104 return ksys_shmctl(shmid, cmd, buf);
1107 #ifdef CONFIG_COMPAT
1109 struct compat_shmid_ds {
1110 struct compat_ipc_perm shm_perm;
1112 compat_time_t shm_atime;
1113 compat_time_t shm_dtime;
1114 compat_time_t shm_ctime;
1115 compat_ipc_pid_t shm_cpid;
1116 compat_ipc_pid_t shm_lpid;
1117 unsigned short shm_nattch;
1118 unsigned short shm_unused;
1119 compat_uptr_t shm_unused2;
1120 compat_uptr_t shm_unused3;
1123 struct compat_shminfo64 {
1124 compat_ulong_t shmmax;
1125 compat_ulong_t shmmin;
1126 compat_ulong_t shmmni;
1127 compat_ulong_t shmseg;
1128 compat_ulong_t shmall;
1129 compat_ulong_t __unused1;
1130 compat_ulong_t __unused2;
1131 compat_ulong_t __unused3;
1132 compat_ulong_t __unused4;
1135 struct compat_shm_info {
1136 compat_int_t used_ids;
1137 compat_ulong_t shm_tot, shm_rss, shm_swp;
1138 compat_ulong_t swap_attempts, swap_successes;
1141 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1144 if (in->shmmax > INT_MAX)
1145 in->shmmax = INT_MAX;
1146 if (version == IPC_64) {
1147 struct compat_shminfo64 info;
1148 memset(&info, 0, sizeof(info));
1149 info.shmmax = in->shmmax;
1150 info.shmmin = in->shmmin;
1151 info.shmmni = in->shmmni;
1152 info.shmseg = in->shmseg;
1153 info.shmall = in->shmall;
1154 return copy_to_user(buf, &info, sizeof(info));
1156 struct shminfo info;
1157 memset(&info, 0, sizeof(info));
1158 info.shmmax = in->shmmax;
1159 info.shmmin = in->shmmin;
1160 info.shmmni = in->shmmni;
1161 info.shmseg = in->shmseg;
1162 info.shmall = in->shmall;
1163 return copy_to_user(buf, &info, sizeof(info));
1167 static int put_compat_shm_info(struct shm_info *ip,
1168 struct compat_shm_info __user *uip)
1170 struct compat_shm_info info;
1172 memset(&info, 0, sizeof(info));
1173 info.used_ids = ip->used_ids;
1174 info.shm_tot = ip->shm_tot;
1175 info.shm_rss = ip->shm_rss;
1176 info.shm_swp = ip->shm_swp;
1177 info.swap_attempts = ip->swap_attempts;
1178 info.swap_successes = ip->swap_successes;
1179 return copy_to_user(uip, &info, sizeof(info));
1182 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1185 if (version == IPC_64) {
1186 struct compat_shmid64_ds v;
1187 memset(&v, 0, sizeof(v));
1188 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1189 v.shm_atime = in->shm_atime;
1190 v.shm_dtime = in->shm_dtime;
1191 v.shm_ctime = in->shm_ctime;
1192 v.shm_segsz = in->shm_segsz;
1193 v.shm_nattch = in->shm_nattch;
1194 v.shm_cpid = in->shm_cpid;
1195 v.shm_lpid = in->shm_lpid;
1196 return copy_to_user(buf, &v, sizeof(v));
1198 struct compat_shmid_ds v;
1199 memset(&v, 0, sizeof(v));
1200 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1201 v.shm_perm.key = in->shm_perm.key;
1202 v.shm_atime = in->shm_atime;
1203 v.shm_dtime = in->shm_dtime;
1204 v.shm_ctime = in->shm_ctime;
1205 v.shm_segsz = in->shm_segsz;
1206 v.shm_nattch = in->shm_nattch;
1207 v.shm_cpid = in->shm_cpid;
1208 v.shm_lpid = in->shm_lpid;
1209 return copy_to_user(buf, &v, sizeof(v));
1213 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1216 memset(out, 0, sizeof(*out));
1217 if (version == IPC_64) {
1218 struct compat_shmid64_ds __user *p = buf;
1219 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1221 struct compat_shmid_ds __user *p = buf;
1222 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1226 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1228 struct ipc_namespace *ns;
1229 struct shmid64_ds sem64;
1230 int version = compat_ipc_parse_version(&cmd);
1233 ns = current->nsproxy->ipc_ns;
1235 if (cmd < 0 || shmid < 0)
1240 struct shminfo64 shminfo;
1241 err = shmctl_ipc_info(ns, &shminfo);
1244 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1249 struct shm_info shm_info;
1250 err = shmctl_shm_info(ns, &shm_info);
1253 if (put_compat_shm_info(&shm_info, uptr))
1259 err = shmctl_stat(ns, shmid, cmd, &sem64);
1262 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1267 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1271 return shmctl_down(ns, shmid, cmd, &sem64);
1274 return shmctl_do_lock(ns, shmid, cmd);
1282 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1284 return compat_ksys_shmctl(shmid, cmd, uptr);
1289 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1291 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1292 * "raddr" thing points to kernel space, and there has to be a wrapper around
1295 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1296 ulong *raddr, unsigned long shmlba)
1298 struct shmid_kernel *shp;
1299 unsigned long addr = (unsigned long)shmaddr;
1303 unsigned long flags = MAP_SHARED;
1306 struct ipc_namespace *ns;
1307 struct shm_file_data *sfd;
1310 unsigned long populate = 0;
1317 if (addr & (shmlba - 1)) {
1319 * Round down to the nearest multiple of shmlba.
1320 * For sane do_mmap_pgoff() parameters, avoid
1321 * round downs that trigger nil-page and MAP_FIXED.
1323 if ((shmflg & SHM_RND) && addr >= shmlba)
1324 addr &= ~(shmlba - 1);
1326 #ifndef __ARCH_FORCE_SHMLBA
1327 if (addr & ~PAGE_MASK)
1333 } else if ((shmflg & SHM_REMAP))
1336 if (shmflg & SHM_RDONLY) {
1339 f_mode = FMODE_READ;
1341 prot = PROT_READ | PROT_WRITE;
1342 acc_mode = S_IRUGO | S_IWUGO;
1343 f_mode = FMODE_READ | FMODE_WRITE;
1345 if (shmflg & SHM_EXEC) {
1347 acc_mode |= S_IXUGO;
1351 * We cannot rely on the fs check since SYSV IPC does have an
1352 * additional creator id...
1354 ns = current->nsproxy->ipc_ns;
1356 shp = shm_obtain_object_check(ns, shmid);
1363 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1366 err = security_shm_shmat(shp, shmaddr, shmflg);
1370 ipc_lock_object(&shp->shm_perm);
1372 /* check if shm_destroy() is tearing down shp */
1373 if (!ipc_valid_object(&shp->shm_perm)) {
1374 ipc_unlock_object(&shp->shm_perm);
1379 path = shp->shm_file->f_path;
1382 size = i_size_read(d_inode(path.dentry));
1383 ipc_unlock_object(&shp->shm_perm);
1387 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1393 file = alloc_file(&path, f_mode,
1394 is_file_hugepages(shp->shm_file) ?
1395 &shm_file_operations_huge :
1396 &shm_file_operations);
1397 err = PTR_ERR(file);
1404 file->private_data = sfd;
1405 file->f_mapping = shp->shm_file->f_mapping;
1406 sfd->id = shp->shm_perm.id;
1407 sfd->ns = get_ipc_ns(ns);
1408 sfd->file = shp->shm_file;
1411 err = security_mmap_file(file, prot, flags);
1415 if (down_write_killable(¤t->mm->mmap_sem)) {
1420 if (addr && !(shmflg & SHM_REMAP)) {
1422 if (addr + size < addr)
1425 if (find_vma_intersection(current->mm, addr, addr + size))
1429 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1432 if (IS_ERR_VALUE(addr))
1435 up_write(¤t->mm->mmap_sem);
1437 mm_populate(addr, populate);
1443 down_write(&shm_ids(ns).rwsem);
1444 shp = shm_lock(ns, shmid);
1446 if (shm_may_destroy(ns, shp))
1447 shm_destroy(ns, shp);
1450 up_write(&shm_ids(ns).rwsem);
1459 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1464 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1467 force_successful_syscall_return();
1471 #ifdef CONFIG_COMPAT
1473 #ifndef COMPAT_SHMLBA
1474 #define COMPAT_SHMLBA SHMLBA
1477 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1482 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1485 force_successful_syscall_return();
1491 * detach and kill segment if marked destroyed.
1492 * The work is done in shm_close.
1494 long ksys_shmdt(char __user *shmaddr)
1496 struct mm_struct *mm = current->mm;
1497 struct vm_area_struct *vma;
1498 unsigned long addr = (unsigned long)shmaddr;
1499 int retval = -EINVAL;
1503 struct vm_area_struct *next;
1506 if (addr & ~PAGE_MASK)
1509 if (down_write_killable(&mm->mmap_sem))
1513 * This function tries to be smart and unmap shm segments that
1514 * were modified by partial mlock or munmap calls:
1515 * - It first determines the size of the shm segment that should be
1516 * unmapped: It searches for a vma that is backed by shm and that
1517 * started at address shmaddr. It records it's size and then unmaps
1519 * - Then it unmaps all shm vmas that started at shmaddr and that
1520 * are within the initially determined size and that are from the
1521 * same shm segment from which we determined the size.
1522 * Errors from do_munmap are ignored: the function only fails if
1523 * it's called with invalid parameters or if it's called to unmap
1524 * a part of a vma. Both calls in this function are for full vmas,
1525 * the parameters are directly copied from the vma itself and always
1526 * valid - therefore do_munmap cannot fail. (famous last words?)
1529 * If it had been mremap()'d, the starting address would not
1530 * match the usual checks anyway. So assume all vma's are
1531 * above the starting address given.
1533 vma = find_vma(mm, addr);
1537 next = vma->vm_next;
1540 * Check if the starting address would match, i.e. it's
1541 * a fragment created by mprotect() and/or munmap(), or it
1542 * otherwise it starts at this address with no hassles.
1544 if ((vma->vm_ops == &shm_vm_ops) &&
1545 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1548 * Record the file of the shm segment being
1549 * unmapped. With mremap(), someone could place
1550 * page from another segment but with equal offsets
1551 * in the range we are unmapping.
1553 file = vma->vm_file;
1554 size = i_size_read(file_inode(vma->vm_file));
1555 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1557 * We discovered the size of the shm segment, so
1558 * break out of here and fall through to the next
1559 * loop that uses the size information to stop
1560 * searching for matching vma's.
1570 * We need look no further than the maximum address a fragment
1571 * could possibly have landed at. Also cast things to loff_t to
1572 * prevent overflows and make comparisons vs. equal-width types.
1574 size = PAGE_ALIGN(size);
1575 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1576 next = vma->vm_next;
1578 /* finding a matching vma now does not alter retval */
1579 if ((vma->vm_ops == &shm_vm_ops) &&
1580 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1581 (vma->vm_file == file))
1582 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1586 #else /* CONFIG_MMU */
1587 /* under NOMMU conditions, the exact address to be destroyed must be
1590 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1591 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1597 up_write(&mm->mmap_sem);
1601 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1603 return ksys_shmdt(shmaddr);
1606 #ifdef CONFIG_PROC_FS
1607 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1609 struct user_namespace *user_ns = seq_user_ns(s);
1610 struct kern_ipc_perm *ipcp = it;
1611 struct shmid_kernel *shp;
1612 unsigned long rss = 0, swp = 0;
1614 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1615 shm_add_rss_swap(shp, &rss, &swp);
1617 #if BITS_PER_LONG <= 32
1618 #define SIZE_SPEC "%10lu"
1620 #define SIZE_SPEC "%21lu"
1624 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1625 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1626 SIZE_SPEC " " SIZE_SPEC "\n",
1634 from_kuid_munged(user_ns, shp->shm_perm.uid),
1635 from_kgid_munged(user_ns, shp->shm_perm.gid),
1636 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1637 from_kgid_munged(user_ns, shp->shm_perm.cgid),