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 WARN_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 WARN_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 shp->shm_cprid = task_tgid_vnr(current);
556 shp->shm_atim = shp->shm_dtim = 0;
557 shp->shm_ctim = get_seconds();
558 shp->shm_segsz = size;
560 shp->shm_file = file;
561 shp->shm_creator = current;
563 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
569 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
572 * shmid gets reported as "inode#" in /proc/pid/maps.
573 * proc-ps tools use this. Changing this will break them.
575 file_inode(file)->i_ino = shp->shm_perm.id;
577 ns->shm_tot += numpages;
578 error = shp->shm_perm.id;
580 ipc_unlock_object(&shp->shm_perm);
585 if (is_file_hugepages(file) && shp->mlock_user)
586 user_shm_unlock(size, shp->mlock_user);
589 ipc_rcu_putref(shp, shm_rcu_free);
594 * Called with shm_ids.rwsem and ipcp locked.
596 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
598 struct shmid_kernel *shp;
600 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
601 return security_shm_associate(shp, shmflg);
605 * Called with shm_ids.rwsem and ipcp locked.
607 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
608 struct ipc_params *params)
610 struct shmid_kernel *shp;
612 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
613 if (shp->shm_segsz < params->u.size)
619 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
621 struct ipc_namespace *ns;
622 static const struct ipc_ops shm_ops = {
624 .associate = shm_security,
625 .more_checks = shm_more_checks,
627 struct ipc_params shm_params;
629 ns = current->nsproxy->ipc_ns;
631 shm_params.key = key;
632 shm_params.flg = shmflg;
633 shm_params.u.size = size;
635 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
638 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
642 return copy_to_user(buf, in, sizeof(*in));
647 memset(&out, 0, sizeof(out));
648 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
649 out.shm_segsz = in->shm_segsz;
650 out.shm_atime = in->shm_atime;
651 out.shm_dtime = in->shm_dtime;
652 out.shm_ctime = in->shm_ctime;
653 out.shm_cpid = in->shm_cpid;
654 out.shm_lpid = in->shm_lpid;
655 out.shm_nattch = in->shm_nattch;
657 return copy_to_user(buf, &out, sizeof(out));
664 static inline unsigned long
665 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
669 if (copy_from_user(out, buf, sizeof(*out)))
674 struct shmid_ds tbuf_old;
676 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
679 out->shm_perm.uid = tbuf_old.shm_perm.uid;
680 out->shm_perm.gid = tbuf_old.shm_perm.gid;
681 out->shm_perm.mode = tbuf_old.shm_perm.mode;
690 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
694 return copy_to_user(buf, in, sizeof(*in));
699 if (in->shmmax > INT_MAX)
700 out.shmmax = INT_MAX;
702 out.shmmax = (int)in->shmmax;
704 out.shmmin = in->shmmin;
705 out.shmmni = in->shmmni;
706 out.shmseg = in->shmseg;
707 out.shmall = in->shmall;
709 return copy_to_user(buf, &out, sizeof(out));
717 * Calculate and add used RSS and swap pages of a shm.
718 * Called with shm_ids.rwsem held as a reader
720 static void shm_add_rss_swap(struct shmid_kernel *shp,
721 unsigned long *rss_add, unsigned long *swp_add)
725 inode = file_inode(shp->shm_file);
727 if (is_file_hugepages(shp->shm_file)) {
728 struct address_space *mapping = inode->i_mapping;
729 struct hstate *h = hstate_file(shp->shm_file);
730 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
733 struct shmem_inode_info *info = SHMEM_I(inode);
734 spin_lock(&info->lock);
735 *rss_add += inode->i_mapping->nrpages;
736 *swp_add += info->swapped;
737 spin_unlock(&info->lock);
739 *rss_add += inode->i_mapping->nrpages;
745 * Called with shm_ids.rwsem held as a reader
747 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
756 in_use = shm_ids(ns).in_use;
758 for (total = 0, next_id = 0; total < in_use; next_id++) {
759 struct kern_ipc_perm *ipc;
760 struct shmid_kernel *shp;
762 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
765 shp = container_of(ipc, struct shmid_kernel, shm_perm);
767 shm_add_rss_swap(shp, rss, swp);
774 * This function handles some shmctl commands which require the rwsem
775 * to be held in write mode.
776 * NOTE: no locks must be held, the rwsem is taken inside this function.
778 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
779 struct shmid_ds __user *buf, int version)
781 struct kern_ipc_perm *ipcp;
782 struct shmid64_ds shmid64;
783 struct shmid_kernel *shp;
786 if (cmd == IPC_SET) {
787 if (copy_shmid_from_user(&shmid64, buf, version))
791 down_write(&shm_ids(ns).rwsem);
794 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
795 &shmid64.shm_perm, 0);
801 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
803 err = security_shm_shmctl(shp, cmd);
809 ipc_lock_object(&shp->shm_perm);
810 /* do_shm_rmid unlocks the ipc object and rcu */
811 do_shm_rmid(ns, ipcp);
814 ipc_lock_object(&shp->shm_perm);
815 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
818 shp->shm_ctim = get_seconds();
826 ipc_unlock_object(&shp->shm_perm);
830 up_write(&shm_ids(ns).rwsem);
834 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
835 int cmd, int version, void __user *buf)
838 struct shmid_kernel *shp;
840 /* preliminary security checks for *_INFO */
841 if (cmd == IPC_INFO || cmd == SHM_INFO) {
842 err = security_shm_shmctl(NULL, cmd);
850 struct shminfo64 shminfo;
852 memset(&shminfo, 0, sizeof(shminfo));
853 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
854 shminfo.shmmax = ns->shm_ctlmax;
855 shminfo.shmall = ns->shm_ctlall;
857 shminfo.shmmin = SHMMIN;
858 if (copy_shminfo_to_user(buf, &shminfo, version))
861 down_read(&shm_ids(ns).rwsem);
862 err = ipc_get_maxid(&shm_ids(ns));
863 up_read(&shm_ids(ns).rwsem);
871 struct shm_info shm_info;
873 memset(&shm_info, 0, sizeof(shm_info));
874 down_read(&shm_ids(ns).rwsem);
875 shm_info.used_ids = shm_ids(ns).in_use;
876 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
877 shm_info.shm_tot = ns->shm_tot;
878 shm_info.swap_attempts = 0;
879 shm_info.swap_successes = 0;
880 err = ipc_get_maxid(&shm_ids(ns));
881 up_read(&shm_ids(ns).rwsem);
882 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
887 err = err < 0 ? 0 : err;
893 struct shmid64_ds tbuf;
897 if (cmd == SHM_STAT) {
898 shp = shm_obtain_object(ns, shmid);
903 result = shp->shm_perm.id;
905 shp = shm_obtain_object_check(ns, shmid);
914 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
917 err = security_shm_shmctl(shp, cmd);
921 memset(&tbuf, 0, sizeof(tbuf));
922 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
923 tbuf.shm_segsz = shp->shm_segsz;
924 tbuf.shm_atime = shp->shm_atim;
925 tbuf.shm_dtime = shp->shm_dtim;
926 tbuf.shm_ctime = shp->shm_ctim;
927 tbuf.shm_cpid = shp->shm_cprid;
928 tbuf.shm_lpid = shp->shm_lprid;
929 tbuf.shm_nattch = shp->shm_nattch;
932 if (copy_shmid_to_user(buf, &tbuf, version))
948 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
950 struct shmid_kernel *shp;
952 struct ipc_namespace *ns;
954 if (cmd < 0 || shmid < 0)
957 version = ipc_parse_version(&cmd);
958 ns = current->nsproxy->ipc_ns;
965 return shmctl_nolock(ns, shmid, cmd, version, buf);
968 return shmctl_down(ns, shmid, cmd, buf, version);
972 struct file *shm_file;
975 shp = shm_obtain_object_check(ns, shmid);
981 audit_ipc_obj(&(shp->shm_perm));
982 err = security_shm_shmctl(shp, cmd);
986 ipc_lock_object(&shp->shm_perm);
988 /* check if shm_destroy() is tearing down shp */
989 if (!ipc_valid_object(&shp->shm_perm)) {
994 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
995 kuid_t euid = current_euid();
996 if (!uid_eq(euid, shp->shm_perm.uid) &&
997 !uid_eq(euid, shp->shm_perm.cuid)) {
1001 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1007 shm_file = shp->shm_file;
1008 if (is_file_hugepages(shm_file))
1011 if (cmd == SHM_LOCK) {
1012 struct user_struct *user = current_user();
1013 err = shmem_lock(shm_file, 1, user);
1014 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1015 shp->shm_perm.mode |= SHM_LOCKED;
1016 shp->mlock_user = user;
1022 if (!(shp->shm_perm.mode & SHM_LOCKED))
1024 shmem_lock(shm_file, 0, shp->mlock_user);
1025 shp->shm_perm.mode &= ~SHM_LOCKED;
1026 shp->mlock_user = NULL;
1028 ipc_unlock_object(&shp->shm_perm);
1030 shmem_unlock_mapping(shm_file->f_mapping);
1040 ipc_unlock_object(&shp->shm_perm);
1047 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1049 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1050 * "raddr" thing points to kernel space, and there has to be a wrapper around
1053 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1054 unsigned long shmlba)
1056 struct shmid_kernel *shp;
1061 unsigned long flags;
1064 struct ipc_namespace *ns;
1065 struct shm_file_data *sfd;
1068 unsigned long populate = 0;
1073 else if ((addr = (ulong)shmaddr)) {
1074 if (addr & (shmlba - 1)) {
1075 if (shmflg & SHM_RND)
1076 addr &= ~(shmlba - 1); /* round down */
1078 #ifndef __ARCH_FORCE_SHMLBA
1079 if (addr & ~PAGE_MASK)
1083 flags = MAP_SHARED | MAP_FIXED;
1085 if ((shmflg & SHM_REMAP))
1091 if (shmflg & SHM_RDONLY) {
1094 f_mode = FMODE_READ;
1096 prot = PROT_READ | PROT_WRITE;
1097 acc_mode = S_IRUGO | S_IWUGO;
1098 f_mode = FMODE_READ | FMODE_WRITE;
1100 if (shmflg & SHM_EXEC) {
1102 acc_mode |= S_IXUGO;
1106 * We cannot rely on the fs check since SYSV IPC does have an
1107 * additional creator id...
1109 ns = current->nsproxy->ipc_ns;
1111 shp = shm_obtain_object_check(ns, shmid);
1118 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1121 err = security_shm_shmat(shp, shmaddr, shmflg);
1125 ipc_lock_object(&shp->shm_perm);
1127 /* check if shm_destroy() is tearing down shp */
1128 if (!ipc_valid_object(&shp->shm_perm)) {
1129 ipc_unlock_object(&shp->shm_perm);
1134 path = shp->shm_file->f_path;
1137 size = i_size_read(d_inode(path.dentry));
1138 ipc_unlock_object(&shp->shm_perm);
1142 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1148 file = alloc_file(&path, f_mode,
1149 is_file_hugepages(shp->shm_file) ?
1150 &shm_file_operations_huge :
1151 &shm_file_operations);
1152 err = PTR_ERR(file);
1159 file->private_data = sfd;
1160 file->f_mapping = shp->shm_file->f_mapping;
1161 sfd->id = shp->shm_perm.id;
1162 sfd->ns = get_ipc_ns(ns);
1163 sfd->file = shp->shm_file;
1166 err = security_mmap_file(file, prot, flags);
1170 down_write(¤t->mm->mmap_sem);
1171 if (addr && !(shmflg & SHM_REMAP)) {
1173 if (addr + size < addr)
1176 if (find_vma_intersection(current->mm, addr, addr + size))
1180 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1183 if (IS_ERR_VALUE(addr))
1186 up_write(¤t->mm->mmap_sem);
1188 mm_populate(addr, populate);
1194 down_write(&shm_ids(ns).rwsem);
1195 shp = shm_lock(ns, shmid);
1197 if (shm_may_destroy(ns, shp))
1198 shm_destroy(ns, shp);
1201 up_write(&shm_ids(ns).rwsem);
1210 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1215 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1218 force_successful_syscall_return();
1223 * detach and kill segment if marked destroyed.
1224 * The work is done in shm_close.
1226 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1228 struct mm_struct *mm = current->mm;
1229 struct vm_area_struct *vma;
1230 unsigned long addr = (unsigned long)shmaddr;
1231 int retval = -EINVAL;
1235 struct vm_area_struct *next;
1238 if (addr & ~PAGE_MASK)
1241 down_write(&mm->mmap_sem);
1244 * This function tries to be smart and unmap shm segments that
1245 * were modified by partial mlock or munmap calls:
1246 * - It first determines the size of the shm segment that should be
1247 * unmapped: It searches for a vma that is backed by shm and that
1248 * started at address shmaddr. It records it's size and then unmaps
1250 * - Then it unmaps all shm vmas that started at shmaddr and that
1251 * are within the initially determined size and that are from the
1252 * same shm segment from which we determined the size.
1253 * Errors from do_munmap are ignored: the function only fails if
1254 * it's called with invalid parameters or if it's called to unmap
1255 * a part of a vma. Both calls in this function are for full vmas,
1256 * the parameters are directly copied from the vma itself and always
1257 * valid - therefore do_munmap cannot fail. (famous last words?)
1260 * If it had been mremap()'d, the starting address would not
1261 * match the usual checks anyway. So assume all vma's are
1262 * above the starting address given.
1264 vma = find_vma(mm, addr);
1268 next = vma->vm_next;
1271 * Check if the starting address would match, i.e. it's
1272 * a fragment created by mprotect() and/or munmap(), or it
1273 * otherwise it starts at this address with no hassles.
1275 if ((vma->vm_ops == &shm_vm_ops) &&
1276 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1279 * Record the file of the shm segment being
1280 * unmapped. With mremap(), someone could place
1281 * page from another segment but with equal offsets
1282 * in the range we are unmapping.
1284 file = vma->vm_file;
1285 size = i_size_read(file_inode(vma->vm_file));
1286 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1288 * We discovered the size of the shm segment, so
1289 * break out of here and fall through to the next
1290 * loop that uses the size information to stop
1291 * searching for matching vma's.
1301 * We need look no further than the maximum address a fragment
1302 * could possibly have landed at. Also cast things to loff_t to
1303 * prevent overflows and make comparisons vs. equal-width types.
1305 size = PAGE_ALIGN(size);
1306 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1307 next = vma->vm_next;
1309 /* finding a matching vma now does not alter retval */
1310 if ((vma->vm_ops == &shm_vm_ops) &&
1311 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1312 (vma->vm_file == file))
1313 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1317 #else /* CONFIG_MMU */
1318 /* under NOMMU conditions, the exact address to be destroyed must be
1320 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1321 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1327 up_write(&mm->mmap_sem);
1331 #ifdef CONFIG_PROC_FS
1332 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1334 struct user_namespace *user_ns = seq_user_ns(s);
1335 struct shmid_kernel *shp = it;
1336 unsigned long rss = 0, swp = 0;
1338 shm_add_rss_swap(shp, &rss, &swp);
1340 #if BITS_PER_LONG <= 32
1341 #define SIZE_SPEC "%10lu"
1343 #define SIZE_SPEC "%21lu"
1347 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1348 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1349 SIZE_SPEC " " SIZE_SPEC "\n",
1357 from_kuid_munged(user_ns, shp->shm_perm.uid),
1358 from_kgid_munged(user_ns, shp->shm_perm.gid),
1359 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1360 from_kgid_munged(user_ns, shp->shm_perm.cgid),