3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
4 * Many improvements/fixes by Bruno Haible.
5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
16 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 * Pavel Emelianov <xemul@openvz.org>
23 * Better ipc lock (kern_ipc_perm.lock) handling
24 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
27 #include <linux/slab.h>
29 #include <linux/hugetlb.h>
30 #include <linux/shm.h>
31 #include <linux/init.h>
32 #include <linux/file.h>
33 #include <linux/mman.h>
34 #include <linux/shmem_fs.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/capability.h>
39 #include <linux/ptrace.h>
40 #include <linux/seq_file.h>
41 #include <linux/rwsem.h>
42 #include <linux/nsproxy.h>
43 #include <linux/mount.h>
44 #include <linux/ipc_namespace.h>
46 #include <linux/uaccess.h>
50 struct shm_file_data {
52 struct ipc_namespace *ns;
54 const struct vm_operations_struct *vm_ops;
57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
59 static const struct file_operations shm_file_operations;
60 static const struct vm_operations_struct shm_vm_ops;
62 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
64 #define shm_unlock(shp) \
65 ipc_unlock(&(shp)->shm_perm)
67 static int newseg(struct ipc_namespace *, struct ipc_params *);
68 static void shm_open(struct vm_area_struct *vma);
69 static void shm_close(struct vm_area_struct *vma);
70 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
75 void shm_init_ns(struct ipc_namespace *ns)
77 ns->shm_ctlmax = SHMMAX;
78 ns->shm_ctlall = SHMALL;
79 ns->shm_ctlmni = SHMMNI;
80 ns->shm_rmid_forced = 0;
82 ipc_init_ids(&shm_ids(ns));
86 * Called with shm_ids.rwsem (writer) and the shp structure locked.
87 * Only shm_ids.rwsem remains locked on exit.
89 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
91 struct shmid_kernel *shp;
92 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
94 if (shp->shm_nattch) {
95 shp->shm_perm.mode |= SHM_DEST;
96 /* Do not find it any more */
97 shp->shm_perm.key = IPC_PRIVATE;
100 shm_destroy(ns, shp);
104 void shm_exit_ns(struct ipc_namespace *ns)
106 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
107 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
111 static int __init ipc_ns_init(void)
113 shm_init_ns(&init_ipc_ns);
117 pure_initcall(ipc_ns_init);
119 void __init shm_init(void)
121 ipc_init_proc_interface("sysvipc/shm",
122 #if BITS_PER_LONG <= 32
123 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
125 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
127 IPC_SHM_IDS, sysvipc_shm_proc_show);
130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
132 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
135 return ERR_CAST(ipcp);
137 return container_of(ipcp, struct shmid_kernel, shm_perm);
140 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
142 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
145 return ERR_CAST(ipcp);
147 return container_of(ipcp, struct shmid_kernel, shm_perm);
151 * shm_lock_(check_) routines are called in the paths where the rwsem
152 * is not necessarily held.
154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
156 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
159 * Callers of shm_lock() must validate the status of the returned ipc
160 * object pointer (as returned by ipc_lock()), and error out as
165 return container_of(ipcp, struct shmid_kernel, shm_perm);
168 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
171 ipc_lock_object(&ipcp->shm_perm);
174 static void shm_rcu_free(struct rcu_head *head)
176 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
177 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
179 security_shm_free(shp);
183 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
185 list_del(&s->shm_clist);
186 ipc_rmid(&shm_ids(ns), &s->shm_perm);
190 static int __shm_open(struct vm_area_struct *vma)
192 struct file *file = vma->vm_file;
193 struct shm_file_data *sfd = shm_file_data(file);
194 struct shmid_kernel *shp;
196 shp = shm_lock(sfd->ns, sfd->id);
201 if (shp->shm_file != sfd->file) {
207 shp->shm_atim = get_seconds();
208 shp->shm_lprid = task_tgid_vnr(current);
214 /* This is called by fork, once for every shm attach. */
215 static void shm_open(struct vm_area_struct *vma)
217 int err = __shm_open(vma);
219 * We raced in the idr lookup or with shm_destroy().
220 * Either way, the ID is busted.
226 * shm_destroy - free the struct shmid_kernel
229 * @shp: struct to free
231 * It has to be called with shp and shm_ids.rwsem (writer) locked,
232 * but returns with shp unlocked and freed.
234 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
236 struct file *shm_file;
238 shm_file = shp->shm_file;
239 shp->shm_file = NULL;
240 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
243 if (!is_file_hugepages(shm_file))
244 shmem_lock(shm_file, 0, shp->mlock_user);
245 else if (shp->mlock_user)
246 user_shm_unlock(i_size_read(file_inode(shm_file)),
249 ipc_rcu_putref(shp, shm_rcu_free);
253 * shm_may_destroy - identifies whether shm segment should be destroyed now
255 * Returns true if and only if there are no active users of the segment and
256 * one of the following is true:
258 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
260 * 2) sysctl kernel.shm_rmid_forced is set to 1.
262 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
264 return (shp->shm_nattch == 0) &&
265 (ns->shm_rmid_forced ||
266 (shp->shm_perm.mode & SHM_DEST));
270 * remove the attach descriptor vma.
271 * free memory for segment if it is marked destroyed.
272 * The descriptor has already been removed from the current->mm->mmap list
273 * and will later be kfree()d.
275 static void shm_close(struct vm_area_struct *vma)
277 struct file *file = vma->vm_file;
278 struct shm_file_data *sfd = shm_file_data(file);
279 struct shmid_kernel *shp;
280 struct ipc_namespace *ns = sfd->ns;
282 down_write(&shm_ids(ns).rwsem);
283 /* remove from the list of attaches of the shm segment */
284 shp = shm_lock(ns, sfd->id);
287 * We raced in the idr lookup or with shm_destroy().
288 * Either way, the ID is busted.
290 if (WARN_ON_ONCE(IS_ERR(shp)))
291 goto done; /* no-op */
293 shp->shm_lprid = task_tgid_vnr(current);
294 shp->shm_dtim = get_seconds();
296 if (shm_may_destroy(ns, shp))
297 shm_destroy(ns, shp);
301 up_write(&shm_ids(ns).rwsem);
304 /* Called with ns->shm_ids(ns).rwsem locked */
305 static int shm_try_destroy_orphaned(int id, void *p, void *data)
307 struct ipc_namespace *ns = data;
308 struct kern_ipc_perm *ipcp = p;
309 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
312 * We want to destroy segments without users and with already
313 * exit'ed originating process.
315 * As shp->* are changed under rwsem, it's safe to skip shp locking.
317 if (shp->shm_creator != NULL)
320 if (shm_may_destroy(ns, shp)) {
321 shm_lock_by_ptr(shp);
322 shm_destroy(ns, shp);
327 void shm_destroy_orphaned(struct ipc_namespace *ns)
329 down_write(&shm_ids(ns).rwsem);
330 if (shm_ids(ns).in_use)
331 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
332 up_write(&shm_ids(ns).rwsem);
335 /* Locking assumes this will only be called with task == current */
336 void exit_shm(struct task_struct *task)
338 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
339 struct shmid_kernel *shp, *n;
341 if (list_empty(&task->sysvshm.shm_clist))
345 * If kernel.shm_rmid_forced is not set then only keep track of
346 * which shmids are orphaned, so that a later set of the sysctl
349 if (!ns->shm_rmid_forced) {
350 down_read(&shm_ids(ns).rwsem);
351 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
352 shp->shm_creator = NULL;
354 * Only under read lock but we are only called on current
355 * so no entry on the list will be shared.
357 list_del(&task->sysvshm.shm_clist);
358 up_read(&shm_ids(ns).rwsem);
363 * Destroy all already created segments, that were not yet mapped,
364 * and mark any mapped as orphan to cover the sysctl toggling.
365 * Destroy is skipped if shm_may_destroy() returns false.
367 down_write(&shm_ids(ns).rwsem);
368 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
369 shp->shm_creator = NULL;
371 if (shm_may_destroy(ns, shp)) {
372 shm_lock_by_ptr(shp);
373 shm_destroy(ns, shp);
377 /* Remove the list head from any segments still attached. */
378 list_del(&task->sysvshm.shm_clist);
379 up_write(&shm_ids(ns).rwsem);
382 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
384 struct file *file = vma->vm_file;
385 struct shm_file_data *sfd = shm_file_data(file);
387 return sfd->vm_ops->fault(vma, vmf);
390 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
392 struct file *file = vma->vm_file;
393 struct shm_file_data *sfd = shm_file_data(file);
395 if (sfd->vm_ops && sfd->vm_ops->split)
396 return sfd->vm_ops->split(vma, addr);
402 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
404 struct file *file = vma->vm_file;
405 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 an
435 * IPC ID that was removed, and possibly even reused by another shm
436 * segment already. Propagate this case as an error to caller.
438 ret =__shm_open(vma);
442 ret = sfd->file->f_op->mmap(sfd->file, vma);
447 sfd->vm_ops = vma->vm_ops;
449 WARN_ON(!sfd->vm_ops->fault);
451 vma->vm_ops = &shm_vm_ops;
455 static int shm_release(struct inode *ino, struct file *file)
457 struct shm_file_data *sfd = shm_file_data(file);
461 shm_file_data(file) = NULL;
466 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
468 struct shm_file_data *sfd = shm_file_data(file);
470 if (!sfd->file->f_op->fsync)
472 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
475 static long shm_fallocate(struct file *file, int mode, loff_t offset,
478 struct shm_file_data *sfd = shm_file_data(file);
480 if (!sfd->file->f_op->fallocate)
482 return sfd->file->f_op->fallocate(file, mode, offset, len);
485 static unsigned long shm_get_unmapped_area(struct file *file,
486 unsigned long addr, unsigned long len, unsigned long pgoff,
489 struct shm_file_data *sfd = shm_file_data(file);
490 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
494 static const struct file_operations shm_file_operations = {
497 .release = shm_release,
498 .get_unmapped_area = shm_get_unmapped_area,
499 .llseek = noop_llseek,
500 .fallocate = shm_fallocate,
504 * shm_file_operations_huge is now identical to shm_file_operations,
505 * but we keep it distinct for the sake of is_file_shm_hugepages().
507 static const struct file_operations shm_file_operations_huge = {
510 .release = shm_release,
511 .get_unmapped_area = shm_get_unmapped_area,
512 .llseek = noop_llseek,
513 .fallocate = shm_fallocate,
516 bool is_file_shm_hugepages(struct file *file)
518 return file->f_op == &shm_file_operations_huge;
521 static const struct vm_operations_struct shm_vm_ops = {
522 .open = shm_open, /* callback for a new vm-area open */
523 .close = shm_close, /* callback for when the vm-area is released */
526 #if defined(CONFIG_NUMA)
527 .set_policy = shm_set_policy,
528 .get_policy = shm_get_policy,
533 * newseg - Create a new shared memory segment
535 * @params: ptr to the structure that contains key, size and shmflg
537 * Called with shm_ids.rwsem held as a writer.
539 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
541 key_t key = params->key;
542 int shmflg = params->flg;
543 size_t size = params->u.size;
545 struct shmid_kernel *shp;
546 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
550 vm_flags_t acctflag = 0;
552 if (size < SHMMIN || size > ns->shm_ctlmax)
555 if (numpages << PAGE_SHIFT < size)
558 if (ns->shm_tot + numpages < ns->shm_tot ||
559 ns->shm_tot + numpages > ns->shm_ctlall)
562 shp = ipc_rcu_alloc(sizeof(*shp));
566 shp->shm_perm.key = key;
567 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
568 shp->mlock_user = NULL;
570 shp->shm_perm.security = NULL;
571 error = security_shm_alloc(shp);
573 ipc_rcu_putref(shp, ipc_rcu_free);
577 sprintf(name, "SYSV%08x", key);
578 if (shmflg & SHM_HUGETLB) {
582 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
587 hugesize = ALIGN(size, huge_page_size(hs));
589 /* hugetlb_file_setup applies strict accounting */
590 if (shmflg & SHM_NORESERVE)
591 acctflag = VM_NORESERVE;
592 file = hugetlb_file_setup(name, hugesize, acctflag,
593 &shp->mlock_user, HUGETLB_SHMFS_INODE,
594 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
597 * Do not allow no accounting for OVERCOMMIT_NEVER, even
600 if ((shmflg & SHM_NORESERVE) &&
601 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
602 acctflag = VM_NORESERVE;
603 file = shmem_kernel_file_setup(name, size, acctflag);
605 error = PTR_ERR(file);
609 shp->shm_cprid = task_tgid_vnr(current);
611 shp->shm_atim = shp->shm_dtim = 0;
612 shp->shm_ctim = get_seconds();
613 shp->shm_segsz = size;
615 shp->shm_file = file;
616 shp->shm_creator = current;
618 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
624 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
627 * shmid gets reported as "inode#" in /proc/pid/maps.
628 * proc-ps tools use this. Changing this will break them.
630 file_inode(file)->i_ino = shp->shm_perm.id;
632 ns->shm_tot += numpages;
633 error = shp->shm_perm.id;
635 ipc_unlock_object(&shp->shm_perm);
640 if (is_file_hugepages(file) && shp->mlock_user)
641 user_shm_unlock(size, shp->mlock_user);
644 ipc_rcu_putref(shp, shm_rcu_free);
649 * Called with shm_ids.rwsem and ipcp locked.
651 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
653 struct shmid_kernel *shp;
655 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
656 return security_shm_associate(shp, shmflg);
660 * Called with shm_ids.rwsem and ipcp locked.
662 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
663 struct ipc_params *params)
665 struct shmid_kernel *shp;
667 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
668 if (shp->shm_segsz < params->u.size)
674 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
676 struct ipc_namespace *ns;
677 static const struct ipc_ops shm_ops = {
679 .associate = shm_security,
680 .more_checks = shm_more_checks,
682 struct ipc_params shm_params;
684 ns = current->nsproxy->ipc_ns;
686 shm_params.key = key;
687 shm_params.flg = shmflg;
688 shm_params.u.size = size;
690 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
693 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
697 return copy_to_user(buf, in, sizeof(*in));
702 memset(&out, 0, sizeof(out));
703 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
704 out.shm_segsz = in->shm_segsz;
705 out.shm_atime = in->shm_atime;
706 out.shm_dtime = in->shm_dtime;
707 out.shm_ctime = in->shm_ctime;
708 out.shm_cpid = in->shm_cpid;
709 out.shm_lpid = in->shm_lpid;
710 out.shm_nattch = in->shm_nattch;
712 return copy_to_user(buf, &out, sizeof(out));
719 static inline unsigned long
720 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
724 if (copy_from_user(out, buf, sizeof(*out)))
729 struct shmid_ds tbuf_old;
731 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
734 out->shm_perm.uid = tbuf_old.shm_perm.uid;
735 out->shm_perm.gid = tbuf_old.shm_perm.gid;
736 out->shm_perm.mode = tbuf_old.shm_perm.mode;
745 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
749 return copy_to_user(buf, in, sizeof(*in));
754 if (in->shmmax > INT_MAX)
755 out.shmmax = INT_MAX;
757 out.shmmax = (int)in->shmmax;
759 out.shmmin = in->shmmin;
760 out.shmmni = in->shmmni;
761 out.shmseg = in->shmseg;
762 out.shmall = in->shmall;
764 return copy_to_user(buf, &out, sizeof(out));
772 * Calculate and add used RSS and swap pages of a shm.
773 * Called with shm_ids.rwsem held as a reader
775 static void shm_add_rss_swap(struct shmid_kernel *shp,
776 unsigned long *rss_add, unsigned long *swp_add)
780 inode = file_inode(shp->shm_file);
782 if (is_file_hugepages(shp->shm_file)) {
783 struct address_space *mapping = inode->i_mapping;
784 struct hstate *h = hstate_file(shp->shm_file);
785 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
788 struct shmem_inode_info *info = SHMEM_I(inode);
789 spin_lock_irq(&info->lock);
790 *rss_add += inode->i_mapping->nrpages;
791 *swp_add += info->swapped;
792 spin_unlock_irq(&info->lock);
794 *rss_add += inode->i_mapping->nrpages;
800 * Called with shm_ids.rwsem held as a reader
802 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
811 in_use = shm_ids(ns).in_use;
813 for (total = 0, next_id = 0; total < in_use; next_id++) {
814 struct kern_ipc_perm *ipc;
815 struct shmid_kernel *shp;
817 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
820 shp = container_of(ipc, struct shmid_kernel, shm_perm);
822 shm_add_rss_swap(shp, rss, swp);
829 * This function handles some shmctl commands which require the rwsem
830 * to be held in write mode.
831 * NOTE: no locks must be held, the rwsem is taken inside this function.
833 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
834 struct shmid_ds __user *buf, int version)
836 struct kern_ipc_perm *ipcp;
837 struct shmid64_ds shmid64;
838 struct shmid_kernel *shp;
841 if (cmd == IPC_SET) {
842 if (copy_shmid_from_user(&shmid64, buf, version))
846 down_write(&shm_ids(ns).rwsem);
849 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
850 &shmid64.shm_perm, 0);
856 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
858 err = security_shm_shmctl(shp, cmd);
864 ipc_lock_object(&shp->shm_perm);
865 /* do_shm_rmid unlocks the ipc object and rcu */
866 do_shm_rmid(ns, ipcp);
869 ipc_lock_object(&shp->shm_perm);
870 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
873 shp->shm_ctim = get_seconds();
881 ipc_unlock_object(&shp->shm_perm);
885 up_write(&shm_ids(ns).rwsem);
889 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
890 int cmd, int version, void __user *buf)
893 struct shmid_kernel *shp;
895 /* preliminary security checks for *_INFO */
896 if (cmd == IPC_INFO || cmd == SHM_INFO) {
897 err = security_shm_shmctl(NULL, cmd);
905 struct shminfo64 shminfo;
907 memset(&shminfo, 0, sizeof(shminfo));
908 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
909 shminfo.shmmax = ns->shm_ctlmax;
910 shminfo.shmall = ns->shm_ctlall;
912 shminfo.shmmin = SHMMIN;
913 if (copy_shminfo_to_user(buf, &shminfo, version))
916 down_read(&shm_ids(ns).rwsem);
917 err = ipc_get_maxid(&shm_ids(ns));
918 up_read(&shm_ids(ns).rwsem);
926 struct shm_info shm_info;
928 memset(&shm_info, 0, sizeof(shm_info));
929 down_read(&shm_ids(ns).rwsem);
930 shm_info.used_ids = shm_ids(ns).in_use;
931 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
932 shm_info.shm_tot = ns->shm_tot;
933 shm_info.swap_attempts = 0;
934 shm_info.swap_successes = 0;
935 err = ipc_get_maxid(&shm_ids(ns));
936 up_read(&shm_ids(ns).rwsem);
937 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
942 err = err < 0 ? 0 : err;
948 struct shmid64_ds tbuf;
952 if (cmd == SHM_STAT) {
953 shp = shm_obtain_object(ns, shmid);
958 result = shp->shm_perm.id;
960 shp = shm_obtain_object_check(ns, shmid);
969 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
972 err = security_shm_shmctl(shp, cmd);
976 memset(&tbuf, 0, sizeof(tbuf));
977 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
978 tbuf.shm_segsz = shp->shm_segsz;
979 tbuf.shm_atime = shp->shm_atim;
980 tbuf.shm_dtime = shp->shm_dtim;
981 tbuf.shm_ctime = shp->shm_ctim;
982 tbuf.shm_cpid = shp->shm_cprid;
983 tbuf.shm_lpid = shp->shm_lprid;
984 tbuf.shm_nattch = shp->shm_nattch;
987 if (copy_shmid_to_user(buf, &tbuf, version))
1003 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1005 struct shmid_kernel *shp;
1007 struct ipc_namespace *ns;
1009 if (cmd < 0 || shmid < 0)
1012 version = ipc_parse_version(&cmd);
1013 ns = current->nsproxy->ipc_ns;
1020 return shmctl_nolock(ns, shmid, cmd, version, buf);
1023 return shmctl_down(ns, shmid, cmd, buf, version);
1027 struct file *shm_file;
1030 shp = shm_obtain_object_check(ns, shmid);
1036 audit_ipc_obj(&(shp->shm_perm));
1037 err = security_shm_shmctl(shp, cmd);
1041 ipc_lock_object(&shp->shm_perm);
1043 /* check if shm_destroy() is tearing down shp */
1044 if (!ipc_valid_object(&shp->shm_perm)) {
1049 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1050 kuid_t euid = current_euid();
1051 if (!uid_eq(euid, shp->shm_perm.uid) &&
1052 !uid_eq(euid, shp->shm_perm.cuid)) {
1056 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1062 shm_file = shp->shm_file;
1063 if (is_file_hugepages(shm_file))
1066 if (cmd == SHM_LOCK) {
1067 struct user_struct *user = current_user();
1068 err = shmem_lock(shm_file, 1, user);
1069 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1070 shp->shm_perm.mode |= SHM_LOCKED;
1071 shp->mlock_user = user;
1077 if (!(shp->shm_perm.mode & SHM_LOCKED))
1079 shmem_lock(shm_file, 0, shp->mlock_user);
1080 shp->shm_perm.mode &= ~SHM_LOCKED;
1081 shp->mlock_user = NULL;
1083 ipc_unlock_object(&shp->shm_perm);
1085 shmem_unlock_mapping(shm_file->f_mapping);
1095 ipc_unlock_object(&shp->shm_perm);
1102 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1104 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1105 * "raddr" thing points to kernel space, and there has to be a wrapper around
1108 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1109 ulong *raddr, unsigned long shmlba)
1111 struct shmid_kernel *shp;
1116 unsigned long flags;
1119 struct ipc_namespace *ns;
1120 struct shm_file_data *sfd;
1123 unsigned long populate = 0;
1128 else if ((addr = (ulong)shmaddr)) {
1129 if (addr & (shmlba - 1)) {
1130 if (shmflg & SHM_RND) {
1131 addr &= ~(shmlba - 1); /* round down */
1134 * Ensure that the round-down is non-nil
1135 * when remapping. This can happen for
1136 * cases when addr < shmlba.
1138 if (!addr && (shmflg & SHM_REMAP))
1141 #ifndef __ARCH_FORCE_SHMLBA
1142 if (addr & ~PAGE_MASK)
1146 flags = MAP_SHARED | MAP_FIXED;
1148 if ((shmflg & SHM_REMAP))
1154 if (shmflg & SHM_RDONLY) {
1157 f_mode = FMODE_READ;
1159 prot = PROT_READ | PROT_WRITE;
1160 acc_mode = S_IRUGO | S_IWUGO;
1161 f_mode = FMODE_READ | FMODE_WRITE;
1163 if (shmflg & SHM_EXEC) {
1165 acc_mode |= S_IXUGO;
1169 * We cannot rely on the fs check since SYSV IPC does have an
1170 * additional creator id...
1172 ns = current->nsproxy->ipc_ns;
1174 shp = shm_obtain_object_check(ns, shmid);
1181 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1184 err = security_shm_shmat(shp, shmaddr, shmflg);
1188 ipc_lock_object(&shp->shm_perm);
1190 /* check if shm_destroy() is tearing down shp */
1191 if (!ipc_valid_object(&shp->shm_perm)) {
1192 ipc_unlock_object(&shp->shm_perm);
1197 path = shp->shm_file->f_path;
1200 size = i_size_read(d_inode(path.dentry));
1201 ipc_unlock_object(&shp->shm_perm);
1205 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1211 file = alloc_file(&path, f_mode,
1212 is_file_hugepages(shp->shm_file) ?
1213 &shm_file_operations_huge :
1214 &shm_file_operations);
1215 err = PTR_ERR(file);
1222 file->private_data = sfd;
1223 file->f_mapping = shp->shm_file->f_mapping;
1224 sfd->id = shp->shm_perm.id;
1225 sfd->ns = get_ipc_ns(ns);
1227 * We need to take a reference to the real shm file to prevent the
1228 * pointer from becoming stale in cases where the lifetime of the outer
1229 * file extends beyond that of the shm segment. It's not usually
1230 * possible, but it can happen during remap_file_pages() emulation as
1231 * that unmaps the memory, then does ->mmap() via file reference only.
1232 * We'll deny the ->mmap() if the shm segment was since removed, but to
1233 * detect shm ID reuse we need to compare the file pointers.
1235 sfd->file = get_file(shp->shm_file);
1238 err = security_mmap_file(file, prot, flags);
1242 if (down_write_killable(¤t->mm->mmap_sem)) {
1247 if (addr && !(shmflg & SHM_REMAP)) {
1249 if (addr + size < addr)
1252 if (find_vma_intersection(current->mm, addr, addr + size))
1256 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1259 if (IS_ERR_VALUE(addr))
1262 up_write(¤t->mm->mmap_sem);
1264 mm_populate(addr, populate);
1270 down_write(&shm_ids(ns).rwsem);
1271 shp = shm_lock(ns, shmid);
1273 if (shm_may_destroy(ns, shp))
1274 shm_destroy(ns, shp);
1277 up_write(&shm_ids(ns).rwsem);
1286 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1291 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1294 force_successful_syscall_return();
1299 * detach and kill segment if marked destroyed.
1300 * The work is done in shm_close.
1302 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1304 struct mm_struct *mm = current->mm;
1305 struct vm_area_struct *vma;
1306 unsigned long addr = (unsigned long)shmaddr;
1307 int retval = -EINVAL;
1311 struct vm_area_struct *next;
1314 if (addr & ~PAGE_MASK)
1317 if (down_write_killable(&mm->mmap_sem))
1321 * This function tries to be smart and unmap shm segments that
1322 * were modified by partial mlock or munmap calls:
1323 * - It first determines the size of the shm segment that should be
1324 * unmapped: It searches for a vma that is backed by shm and that
1325 * started at address shmaddr. It records it's size and then unmaps
1327 * - Then it unmaps all shm vmas that started at shmaddr and that
1328 * are within the initially determined size and that are from the
1329 * same shm segment from which we determined the size.
1330 * Errors from do_munmap are ignored: the function only fails if
1331 * it's called with invalid parameters or if it's called to unmap
1332 * a part of a vma. Both calls in this function are for full vmas,
1333 * the parameters are directly copied from the vma itself and always
1334 * valid - therefore do_munmap cannot fail. (famous last words?)
1337 * If it had been mremap()'d, the starting address would not
1338 * match the usual checks anyway. So assume all vma's are
1339 * above the starting address given.
1341 vma = find_vma(mm, addr);
1345 next = vma->vm_next;
1348 * Check if the starting address would match, i.e. it's
1349 * a fragment created by mprotect() and/or munmap(), or it
1350 * otherwise it starts at this address with no hassles.
1352 if ((vma->vm_ops == &shm_vm_ops) &&
1353 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1356 * Record the file of the shm segment being
1357 * unmapped. With mremap(), someone could place
1358 * page from another segment but with equal offsets
1359 * in the range we are unmapping.
1361 file = vma->vm_file;
1362 size = i_size_read(file_inode(vma->vm_file));
1363 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1365 * We discovered the size of the shm segment, so
1366 * break out of here and fall through to the next
1367 * loop that uses the size information to stop
1368 * searching for matching vma's.
1378 * We need look no further than the maximum address a fragment
1379 * could possibly have landed at. Also cast things to loff_t to
1380 * prevent overflows and make comparisons vs. equal-width types.
1382 size = PAGE_ALIGN(size);
1383 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1384 next = vma->vm_next;
1386 /* finding a matching vma now does not alter retval */
1387 if ((vma->vm_ops == &shm_vm_ops) &&
1388 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1389 (vma->vm_file == file))
1390 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1394 #else /* CONFIG_MMU */
1395 /* under NOMMU conditions, the exact address to be destroyed must be
1397 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1398 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1404 up_write(&mm->mmap_sem);
1408 #ifdef CONFIG_PROC_FS
1409 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1411 struct user_namespace *user_ns = seq_user_ns(s);
1412 struct shmid_kernel *shp = it;
1413 unsigned long rss = 0, swp = 0;
1415 shm_add_rss_swap(shp, &rss, &swp);
1417 #if BITS_PER_LONG <= 32
1418 #define SIZE_SPEC "%10lu"
1420 #define SIZE_SPEC "%21lu"
1424 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1425 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1426 SIZE_SPEC " " SIZE_SPEC "\n",
1434 from_kuid_munged(user_ns, shp->shm_perm.uid),
1435 from_kgid_munged(user_ns, shp->shm_perm.gid),
1436 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1437 from_kgid_munged(user_ns, shp->shm_perm.cgid),