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 * We raced in the idr lookup or with shm_destroy(). Either way, the
162 BUG_ON(IS_ERR(ipcp));
164 return container_of(ipcp, struct shmid_kernel, shm_perm);
167 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
170 ipc_lock_object(&ipcp->shm_perm);
173 static void shm_rcu_free(struct rcu_head *head)
175 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
176 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
178 security_shm_free(shp);
182 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
184 list_del(&s->shm_clist);
185 ipc_rmid(&shm_ids(ns), &s->shm_perm);
189 /* This is called by fork, once for every shm attach. */
190 static void 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);
197 shp->shm_atim = get_seconds();
198 shp->shm_lprid = task_tgid_vnr(current);
204 * shm_destroy - free the struct shmid_kernel
207 * @shp: struct to free
209 * It has to be called with shp and shm_ids.rwsem (writer) locked,
210 * but returns with shp unlocked and freed.
212 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
214 struct file *shm_file;
216 shm_file = shp->shm_file;
217 shp->shm_file = NULL;
218 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
221 if (!is_file_hugepages(shm_file))
222 shmem_lock(shm_file, 0, shp->mlock_user);
223 else if (shp->mlock_user)
224 user_shm_unlock(i_size_read(file_inode(shm_file)),
227 ipc_rcu_putref(shp, shm_rcu_free);
231 * shm_may_destroy - identifies whether shm segment should be destroyed now
233 * Returns true if and only if there are no active users of the segment and
234 * one of the following is true:
236 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
238 * 2) sysctl kernel.shm_rmid_forced is set to 1.
240 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
242 return (shp->shm_nattch == 0) &&
243 (ns->shm_rmid_forced ||
244 (shp->shm_perm.mode & SHM_DEST));
248 * remove the attach descriptor vma.
249 * free memory for segment if it is marked destroyed.
250 * The descriptor has already been removed from the current->mm->mmap list
251 * and will later be kfree()d.
253 static void shm_close(struct vm_area_struct *vma)
255 struct file *file = vma->vm_file;
256 struct shm_file_data *sfd = shm_file_data(file);
257 struct shmid_kernel *shp;
258 struct ipc_namespace *ns = sfd->ns;
260 down_write(&shm_ids(ns).rwsem);
261 /* remove from the list of attaches of the shm segment */
262 shp = shm_lock(ns, sfd->id);
263 shp->shm_lprid = task_tgid_vnr(current);
264 shp->shm_dtim = get_seconds();
266 if (shm_may_destroy(ns, shp))
267 shm_destroy(ns, shp);
270 up_write(&shm_ids(ns).rwsem);
273 /* Called with ns->shm_ids(ns).rwsem locked */
274 static int shm_try_destroy_orphaned(int id, void *p, void *data)
276 struct ipc_namespace *ns = data;
277 struct kern_ipc_perm *ipcp = p;
278 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
281 * We want to destroy segments without users and with already
282 * exit'ed originating process.
284 * As shp->* are changed under rwsem, it's safe to skip shp locking.
286 if (shp->shm_creator != NULL)
289 if (shm_may_destroy(ns, shp)) {
290 shm_lock_by_ptr(shp);
291 shm_destroy(ns, shp);
296 void shm_destroy_orphaned(struct ipc_namespace *ns)
298 down_write(&shm_ids(ns).rwsem);
299 if (shm_ids(ns).in_use)
300 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
301 up_write(&shm_ids(ns).rwsem);
304 /* Locking assumes this will only be called with task == current */
305 void exit_shm(struct task_struct *task)
307 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
308 struct shmid_kernel *shp, *n;
310 if (list_empty(&task->sysvshm.shm_clist))
314 * If kernel.shm_rmid_forced is not set then only keep track of
315 * which shmids are orphaned, so that a later set of the sysctl
318 if (!ns->shm_rmid_forced) {
319 down_read(&shm_ids(ns).rwsem);
320 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
321 shp->shm_creator = NULL;
323 * Only under read lock but we are only called on current
324 * so no entry on the list will be shared.
326 list_del(&task->sysvshm.shm_clist);
327 up_read(&shm_ids(ns).rwsem);
332 * Destroy all already created segments, that were not yet mapped,
333 * and mark any mapped as orphan to cover the sysctl toggling.
334 * Destroy is skipped if shm_may_destroy() returns false.
336 down_write(&shm_ids(ns).rwsem);
337 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
338 shp->shm_creator = NULL;
340 if (shm_may_destroy(ns, shp)) {
341 shm_lock_by_ptr(shp);
342 shm_destroy(ns, shp);
346 /* Remove the list head from any segments still attached. */
347 list_del(&task->sysvshm.shm_clist);
348 up_write(&shm_ids(ns).rwsem);
351 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
353 struct file *file = vma->vm_file;
354 struct shm_file_data *sfd = shm_file_data(file);
356 return sfd->vm_ops->fault(vma, vmf);
360 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
362 struct file *file = vma->vm_file;
363 struct shm_file_data *sfd = shm_file_data(file);
365 if (sfd->vm_ops->set_policy)
366 err = sfd->vm_ops->set_policy(vma, new);
370 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
373 struct file *file = vma->vm_file;
374 struct shm_file_data *sfd = shm_file_data(file);
375 struct mempolicy *pol = NULL;
377 if (sfd->vm_ops->get_policy)
378 pol = sfd->vm_ops->get_policy(vma, addr);
379 else if (vma->vm_policy)
380 pol = vma->vm_policy;
386 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
388 struct shm_file_data *sfd = shm_file_data(file);
391 ret = sfd->file->f_op->mmap(sfd->file, vma);
394 sfd->vm_ops = vma->vm_ops;
396 BUG_ON(!sfd->vm_ops->fault);
398 vma->vm_ops = &shm_vm_ops;
404 static int shm_release(struct inode *ino, struct file *file)
406 struct shm_file_data *sfd = shm_file_data(file);
409 shm_file_data(file) = NULL;
414 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
416 struct shm_file_data *sfd = shm_file_data(file);
418 if (!sfd->file->f_op->fsync)
420 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
423 static long shm_fallocate(struct file *file, int mode, loff_t offset,
426 struct shm_file_data *sfd = shm_file_data(file);
428 if (!sfd->file->f_op->fallocate)
430 return sfd->file->f_op->fallocate(file, mode, offset, len);
433 static unsigned long shm_get_unmapped_area(struct file *file,
434 unsigned long addr, unsigned long len, unsigned long pgoff,
437 struct shm_file_data *sfd = shm_file_data(file);
438 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
442 static const struct file_operations shm_file_operations = {
445 .release = shm_release,
447 .get_unmapped_area = shm_get_unmapped_area,
449 .llseek = noop_llseek,
450 .fallocate = shm_fallocate,
453 static const struct file_operations shm_file_operations_huge = {
456 .release = shm_release,
457 .get_unmapped_area = shm_get_unmapped_area,
458 .llseek = noop_llseek,
459 .fallocate = shm_fallocate,
462 int is_file_shm_hugepages(struct file *file)
464 return file->f_op == &shm_file_operations_huge;
467 static const struct vm_operations_struct shm_vm_ops = {
468 .open = shm_open, /* callback for a new vm-area open */
469 .close = shm_close, /* callback for when the vm-area is released */
471 #if defined(CONFIG_NUMA)
472 .set_policy = shm_set_policy,
473 .get_policy = shm_get_policy,
478 * newseg - Create a new shared memory segment
480 * @params: ptr to the structure that contains key, size and shmflg
482 * Called with shm_ids.rwsem held as a writer.
484 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
486 key_t key = params->key;
487 int shmflg = params->flg;
488 size_t size = params->u.size;
490 struct shmid_kernel *shp;
491 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
495 vm_flags_t acctflag = 0;
497 if (size < SHMMIN || size > ns->shm_ctlmax)
500 if (numpages << PAGE_SHIFT < size)
503 if (ns->shm_tot + numpages < ns->shm_tot ||
504 ns->shm_tot + numpages > ns->shm_ctlall)
507 shp = ipc_rcu_alloc(sizeof(*shp));
511 shp->shm_perm.key = key;
512 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
513 shp->mlock_user = NULL;
515 shp->shm_perm.security = NULL;
516 error = security_shm_alloc(shp);
518 ipc_rcu_putref(shp, ipc_rcu_free);
522 sprintf(name, "SYSV%08x", key);
523 if (shmflg & SHM_HUGETLB) {
527 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
532 hugesize = ALIGN(size, huge_page_size(hs));
534 /* hugetlb_file_setup applies strict accounting */
535 if (shmflg & SHM_NORESERVE)
536 acctflag = VM_NORESERVE;
537 file = hugetlb_file_setup(name, hugesize, acctflag,
538 &shp->mlock_user, HUGETLB_SHMFS_INODE,
539 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
542 * Do not allow no accounting for OVERCOMMIT_NEVER, even
545 if ((shmflg & SHM_NORESERVE) &&
546 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
547 acctflag = VM_NORESERVE;
548 file = shmem_kernel_file_setup(name, size, acctflag);
550 error = PTR_ERR(file);
554 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
560 shp->shm_cprid = task_tgid_vnr(current);
562 shp->shm_atim = shp->shm_dtim = 0;
563 shp->shm_ctim = get_seconds();
564 shp->shm_segsz = size;
566 shp->shm_file = file;
567 shp->shm_creator = current;
568 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
571 * shmid gets reported as "inode#" in /proc/pid/maps.
572 * proc-ps tools use this. Changing this will break them.
574 file_inode(file)->i_ino = shp->shm_perm.id;
576 ns->shm_tot += numpages;
577 error = shp->shm_perm.id;
579 ipc_unlock_object(&shp->shm_perm);
584 if (is_file_hugepages(file) && shp->mlock_user)
585 user_shm_unlock(size, shp->mlock_user);
588 ipc_rcu_putref(shp, shm_rcu_free);
593 * Called with shm_ids.rwsem and ipcp locked.
595 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
597 struct shmid_kernel *shp;
599 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
600 return security_shm_associate(shp, shmflg);
604 * Called with shm_ids.rwsem and ipcp locked.
606 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
607 struct ipc_params *params)
609 struct shmid_kernel *shp;
611 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
612 if (shp->shm_segsz < params->u.size)
618 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
620 struct ipc_namespace *ns;
621 static const struct ipc_ops shm_ops = {
623 .associate = shm_security,
624 .more_checks = shm_more_checks,
626 struct ipc_params shm_params;
628 ns = current->nsproxy->ipc_ns;
630 shm_params.key = key;
631 shm_params.flg = shmflg;
632 shm_params.u.size = size;
634 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
637 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
641 return copy_to_user(buf, in, sizeof(*in));
646 memset(&out, 0, sizeof(out));
647 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
648 out.shm_segsz = in->shm_segsz;
649 out.shm_atime = in->shm_atime;
650 out.shm_dtime = in->shm_dtime;
651 out.shm_ctime = in->shm_ctime;
652 out.shm_cpid = in->shm_cpid;
653 out.shm_lpid = in->shm_lpid;
654 out.shm_nattch = in->shm_nattch;
656 return copy_to_user(buf, &out, sizeof(out));
663 static inline unsigned long
664 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
668 if (copy_from_user(out, buf, sizeof(*out)))
673 struct shmid_ds tbuf_old;
675 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
678 out->shm_perm.uid = tbuf_old.shm_perm.uid;
679 out->shm_perm.gid = tbuf_old.shm_perm.gid;
680 out->shm_perm.mode = tbuf_old.shm_perm.mode;
689 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
693 return copy_to_user(buf, in, sizeof(*in));
698 if (in->shmmax > INT_MAX)
699 out.shmmax = INT_MAX;
701 out.shmmax = (int)in->shmmax;
703 out.shmmin = in->shmmin;
704 out.shmmni = in->shmmni;
705 out.shmseg = in->shmseg;
706 out.shmall = in->shmall;
708 return copy_to_user(buf, &out, sizeof(out));
716 * Calculate and add used RSS and swap pages of a shm.
717 * Called with shm_ids.rwsem held as a reader
719 static void shm_add_rss_swap(struct shmid_kernel *shp,
720 unsigned long *rss_add, unsigned long *swp_add)
724 inode = file_inode(shp->shm_file);
726 if (is_file_hugepages(shp->shm_file)) {
727 struct address_space *mapping = inode->i_mapping;
728 struct hstate *h = hstate_file(shp->shm_file);
729 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
732 struct shmem_inode_info *info = SHMEM_I(inode);
733 spin_lock(&info->lock);
734 *rss_add += inode->i_mapping->nrpages;
735 *swp_add += info->swapped;
736 spin_unlock(&info->lock);
738 *rss_add += inode->i_mapping->nrpages;
744 * Called with shm_ids.rwsem held as a reader
746 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
755 in_use = shm_ids(ns).in_use;
757 for (total = 0, next_id = 0; total < in_use; next_id++) {
758 struct kern_ipc_perm *ipc;
759 struct shmid_kernel *shp;
761 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
764 shp = container_of(ipc, struct shmid_kernel, shm_perm);
766 shm_add_rss_swap(shp, rss, swp);
773 * This function handles some shmctl commands which require the rwsem
774 * to be held in write mode.
775 * NOTE: no locks must be held, the rwsem is taken inside this function.
777 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
778 struct shmid_ds __user *buf, int version)
780 struct kern_ipc_perm *ipcp;
781 struct shmid64_ds shmid64;
782 struct shmid_kernel *shp;
785 if (cmd == IPC_SET) {
786 if (copy_shmid_from_user(&shmid64, buf, version))
790 down_write(&shm_ids(ns).rwsem);
793 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
794 &shmid64.shm_perm, 0);
800 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
802 err = security_shm_shmctl(shp, cmd);
808 ipc_lock_object(&shp->shm_perm);
809 /* do_shm_rmid unlocks the ipc object and rcu */
810 do_shm_rmid(ns, ipcp);
813 ipc_lock_object(&shp->shm_perm);
814 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
817 shp->shm_ctim = get_seconds();
825 ipc_unlock_object(&shp->shm_perm);
829 up_write(&shm_ids(ns).rwsem);
833 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
834 int cmd, int version, void __user *buf)
837 struct shmid_kernel *shp;
839 /* preliminary security checks for *_INFO */
840 if (cmd == IPC_INFO || cmd == SHM_INFO) {
841 err = security_shm_shmctl(NULL, cmd);
849 struct shminfo64 shminfo;
851 memset(&shminfo, 0, sizeof(shminfo));
852 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
853 shminfo.shmmax = ns->shm_ctlmax;
854 shminfo.shmall = ns->shm_ctlall;
856 shminfo.shmmin = SHMMIN;
857 if (copy_shminfo_to_user(buf, &shminfo, version))
860 down_read(&shm_ids(ns).rwsem);
861 err = ipc_get_maxid(&shm_ids(ns));
862 up_read(&shm_ids(ns).rwsem);
870 struct shm_info shm_info;
872 memset(&shm_info, 0, sizeof(shm_info));
873 down_read(&shm_ids(ns).rwsem);
874 shm_info.used_ids = shm_ids(ns).in_use;
875 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
876 shm_info.shm_tot = ns->shm_tot;
877 shm_info.swap_attempts = 0;
878 shm_info.swap_successes = 0;
879 err = ipc_get_maxid(&shm_ids(ns));
880 up_read(&shm_ids(ns).rwsem);
881 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
886 err = err < 0 ? 0 : err;
892 struct shmid64_ds tbuf;
896 if (cmd == SHM_STAT) {
897 shp = shm_obtain_object(ns, shmid);
902 result = shp->shm_perm.id;
904 shp = shm_obtain_object_check(ns, shmid);
913 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
916 err = security_shm_shmctl(shp, cmd);
920 memset(&tbuf, 0, sizeof(tbuf));
921 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
922 tbuf.shm_segsz = shp->shm_segsz;
923 tbuf.shm_atime = shp->shm_atim;
924 tbuf.shm_dtime = shp->shm_dtim;
925 tbuf.shm_ctime = shp->shm_ctim;
926 tbuf.shm_cpid = shp->shm_cprid;
927 tbuf.shm_lpid = shp->shm_lprid;
928 tbuf.shm_nattch = shp->shm_nattch;
931 if (copy_shmid_to_user(buf, &tbuf, version))
947 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
949 struct shmid_kernel *shp;
951 struct ipc_namespace *ns;
953 if (cmd < 0 || shmid < 0)
956 version = ipc_parse_version(&cmd);
957 ns = current->nsproxy->ipc_ns;
964 return shmctl_nolock(ns, shmid, cmd, version, buf);
967 return shmctl_down(ns, shmid, cmd, buf, version);
971 struct file *shm_file;
974 shp = shm_obtain_object_check(ns, shmid);
980 audit_ipc_obj(&(shp->shm_perm));
981 err = security_shm_shmctl(shp, cmd);
985 ipc_lock_object(&shp->shm_perm);
987 /* check if shm_destroy() is tearing down shp */
988 if (!ipc_valid_object(&shp->shm_perm)) {
993 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
994 kuid_t euid = current_euid();
995 if (!uid_eq(euid, shp->shm_perm.uid) &&
996 !uid_eq(euid, shp->shm_perm.cuid)) {
1000 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1006 shm_file = shp->shm_file;
1007 if (is_file_hugepages(shm_file))
1010 if (cmd == SHM_LOCK) {
1011 struct user_struct *user = current_user();
1012 err = shmem_lock(shm_file, 1, user);
1013 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1014 shp->shm_perm.mode |= SHM_LOCKED;
1015 shp->mlock_user = user;
1021 if (!(shp->shm_perm.mode & SHM_LOCKED))
1023 shmem_lock(shm_file, 0, shp->mlock_user);
1024 shp->shm_perm.mode &= ~SHM_LOCKED;
1025 shp->mlock_user = NULL;
1027 ipc_unlock_object(&shp->shm_perm);
1029 shmem_unlock_mapping(shm_file->f_mapping);
1039 ipc_unlock_object(&shp->shm_perm);
1046 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1048 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1049 * "raddr" thing points to kernel space, and there has to be a wrapper around
1052 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1053 unsigned long shmlba)
1055 struct shmid_kernel *shp;
1060 unsigned long flags;
1063 struct ipc_namespace *ns;
1064 struct shm_file_data *sfd;
1067 unsigned long populate = 0;
1072 else if ((addr = (ulong)shmaddr)) {
1073 if (addr & (shmlba - 1)) {
1074 if (shmflg & SHM_RND)
1075 addr &= ~(shmlba - 1); /* round down */
1077 #ifndef __ARCH_FORCE_SHMLBA
1078 if (addr & ~PAGE_MASK)
1082 flags = MAP_SHARED | MAP_FIXED;
1084 if ((shmflg & SHM_REMAP))
1090 if (shmflg & SHM_RDONLY) {
1093 f_mode = FMODE_READ;
1095 prot = PROT_READ | PROT_WRITE;
1096 acc_mode = S_IRUGO | S_IWUGO;
1097 f_mode = FMODE_READ | FMODE_WRITE;
1099 if (shmflg & SHM_EXEC) {
1101 acc_mode |= S_IXUGO;
1105 * We cannot rely on the fs check since SYSV IPC does have an
1106 * additional creator id...
1108 ns = current->nsproxy->ipc_ns;
1110 shp = shm_obtain_object_check(ns, shmid);
1117 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1120 err = security_shm_shmat(shp, shmaddr, shmflg);
1124 ipc_lock_object(&shp->shm_perm);
1126 /* check if shm_destroy() is tearing down shp */
1127 if (!ipc_valid_object(&shp->shm_perm)) {
1128 ipc_unlock_object(&shp->shm_perm);
1133 path = shp->shm_file->f_path;
1136 size = i_size_read(d_inode(path.dentry));
1137 ipc_unlock_object(&shp->shm_perm);
1141 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1147 file = alloc_file(&path, f_mode,
1148 is_file_hugepages(shp->shm_file) ?
1149 &shm_file_operations_huge :
1150 &shm_file_operations);
1151 err = PTR_ERR(file);
1158 file->private_data = sfd;
1159 file->f_mapping = shp->shm_file->f_mapping;
1160 sfd->id = shp->shm_perm.id;
1161 sfd->ns = get_ipc_ns(ns);
1162 sfd->file = shp->shm_file;
1165 err = security_mmap_file(file, prot, flags);
1169 down_write(¤t->mm->mmap_sem);
1170 if (addr && !(shmflg & SHM_REMAP)) {
1172 if (addr + size < addr)
1175 if (find_vma_intersection(current->mm, addr, addr + size))
1179 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1182 if (IS_ERR_VALUE(addr))
1185 up_write(¤t->mm->mmap_sem);
1187 mm_populate(addr, populate);
1193 down_write(&shm_ids(ns).rwsem);
1194 shp = shm_lock(ns, shmid);
1196 if (shm_may_destroy(ns, shp))
1197 shm_destroy(ns, shp);
1200 up_write(&shm_ids(ns).rwsem);
1209 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1214 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1217 force_successful_syscall_return();
1222 * detach and kill segment if marked destroyed.
1223 * The work is done in shm_close.
1225 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1227 struct mm_struct *mm = current->mm;
1228 struct vm_area_struct *vma;
1229 unsigned long addr = (unsigned long)shmaddr;
1230 int retval = -EINVAL;
1234 struct vm_area_struct *next;
1237 if (addr & ~PAGE_MASK)
1240 down_write(&mm->mmap_sem);
1243 * This function tries to be smart and unmap shm segments that
1244 * were modified by partial mlock or munmap calls:
1245 * - It first determines the size of the shm segment that should be
1246 * unmapped: It searches for a vma that is backed by shm and that
1247 * started at address shmaddr. It records it's size and then unmaps
1249 * - Then it unmaps all shm vmas that started at shmaddr and that
1250 * are within the initially determined size and that are from the
1251 * same shm segment from which we determined the size.
1252 * Errors from do_munmap are ignored: the function only fails if
1253 * it's called with invalid parameters or if it's called to unmap
1254 * a part of a vma. Both calls in this function are for full vmas,
1255 * the parameters are directly copied from the vma itself and always
1256 * valid - therefore do_munmap cannot fail. (famous last words?)
1259 * If it had been mremap()'d, the starting address would not
1260 * match the usual checks anyway. So assume all vma's are
1261 * above the starting address given.
1263 vma = find_vma(mm, addr);
1267 next = vma->vm_next;
1270 * Check if the starting address would match, i.e. it's
1271 * a fragment created by mprotect() and/or munmap(), or it
1272 * otherwise it starts at this address with no hassles.
1274 if ((vma->vm_ops == &shm_vm_ops) &&
1275 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1278 * Record the file of the shm segment being
1279 * unmapped. With mremap(), someone could place
1280 * page from another segment but with equal offsets
1281 * in the range we are unmapping.
1283 file = vma->vm_file;
1284 size = i_size_read(file_inode(vma->vm_file));
1285 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1287 * We discovered the size of the shm segment, so
1288 * break out of here and fall through to the next
1289 * loop that uses the size information to stop
1290 * searching for matching vma's.
1300 * We need look no further than the maximum address a fragment
1301 * could possibly have landed at. Also cast things to loff_t to
1302 * prevent overflows and make comparisons vs. equal-width types.
1304 size = PAGE_ALIGN(size);
1305 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1306 next = vma->vm_next;
1308 /* finding a matching vma now does not alter retval */
1309 if ((vma->vm_ops == &shm_vm_ops) &&
1310 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1311 (vma->vm_file == file))
1312 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1316 #else /* CONFIG_MMU */
1317 /* under NOMMU conditions, the exact address to be destroyed must be
1319 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1320 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1326 up_write(&mm->mmap_sem);
1330 #ifdef CONFIG_PROC_FS
1331 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1333 struct user_namespace *user_ns = seq_user_ns(s);
1334 struct shmid_kernel *shp = it;
1335 unsigned long rss = 0, swp = 0;
1337 shm_add_rss_swap(shp, &rss, &swp);
1339 #if BITS_PER_LONG <= 32
1340 #define SIZE_SPEC "%10lu"
1342 #define SIZE_SPEC "%21lu"
1346 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1347 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1348 SIZE_SPEC " " SIZE_SPEC "\n",
1356 from_kuid_munged(user_ns, shp->shm_perm.uid),
1357 from_kgid_munged(user_ns, shp->shm_perm.gid),
1358 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1359 from_kgid_munged(user_ns, shp->shm_perm.cgid),