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);
93 WARN_ON(ns != shp->ns);
95 if (shp->shm_nattch) {
96 shp->shm_perm.mode |= SHM_DEST;
97 /* Do not find it any more */
98 shp->shm_perm.key = IPC_PRIVATE;
101 shm_destroy(ns, shp);
105 void shm_exit_ns(struct ipc_namespace *ns)
107 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
108 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
112 static int __init ipc_ns_init(void)
114 shm_init_ns(&init_ipc_ns);
118 pure_initcall(ipc_ns_init);
120 void __init shm_init(void)
122 ipc_init_proc_interface("sysvipc/shm",
123 #if BITS_PER_LONG <= 32
124 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
126 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
128 IPC_SHM_IDS, sysvipc_shm_proc_show);
131 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
133 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
136 return ERR_CAST(ipcp);
138 return container_of(ipcp, struct shmid_kernel, shm_perm);
141 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
143 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
146 return ERR_CAST(ipcp);
148 return container_of(ipcp, struct shmid_kernel, shm_perm);
152 * shm_lock_(check_) routines are called in the paths where the rwsem
153 * is not necessarily held.
155 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
157 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
160 * Callers of shm_lock() must validate the status of the returned ipc
161 * object pointer (as returned by ipc_lock()), and error out as
166 return container_of(ipcp, struct shmid_kernel, shm_perm);
169 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
172 ipc_lock_object(&ipcp->shm_perm);
175 static void shm_rcu_free(struct rcu_head *head)
177 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
178 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
180 security_shm_free(shp);
185 * It has to be called with shp locked.
186 * It must be called before ipc_rmid()
188 static inline void shm_clist_rm(struct shmid_kernel *shp)
190 struct task_struct *creator;
192 /* ensure that shm_creator does not disappear */
196 * A concurrent exit_shm may do a list_del_init() as well.
197 * Just do nothing if exit_shm already did the work
199 if (!list_empty(&shp->shm_clist)) {
201 * shp->shm_creator is guaranteed to be valid *only*
202 * if shp->shm_clist is not empty.
204 creator = shp->shm_creator;
208 * list_del_init() is a nop if the entry was already removed
211 list_del_init(&shp->shm_clist);
212 task_unlock(creator);
217 static inline void shm_rmid(struct shmid_kernel *s)
220 ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
224 static int __shm_open(struct vm_area_struct *vma)
226 struct file *file = vma->vm_file;
227 struct shm_file_data *sfd = shm_file_data(file);
228 struct shmid_kernel *shp;
230 shp = shm_lock(sfd->ns, sfd->id);
235 if (shp->shm_file != sfd->file) {
241 shp->shm_atim = get_seconds();
242 shp->shm_lprid = task_tgid_vnr(current);
248 /* This is called by fork, once for every shm attach. */
249 static void shm_open(struct vm_area_struct *vma)
251 int err = __shm_open(vma);
253 * We raced in the idr lookup or with shm_destroy().
254 * Either way, the ID is busted.
260 * shm_destroy - free the struct shmid_kernel
263 * @shp: struct to free
265 * It has to be called with shp and shm_ids.rwsem (writer) locked,
266 * but returns with shp unlocked and freed.
268 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
270 struct file *shm_file;
272 shm_file = shp->shm_file;
273 shp->shm_file = NULL;
274 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
277 if (!is_file_hugepages(shm_file))
278 shmem_lock(shm_file, 0, shp->mlock_user);
279 else if (shp->mlock_user)
280 user_shm_unlock(i_size_read(file_inode(shm_file)),
283 ipc_rcu_putref(shp, shm_rcu_free);
287 * shm_may_destroy - identifies whether shm segment should be destroyed now
289 * Returns true if and only if there are no active users of the segment and
290 * one of the following is true:
292 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
294 * 2) sysctl kernel.shm_rmid_forced is set to 1.
296 static bool shm_may_destroy(struct shmid_kernel *shp)
298 return (shp->shm_nattch == 0) &&
299 (shp->ns->shm_rmid_forced ||
300 (shp->shm_perm.mode & SHM_DEST));
304 * remove the attach descriptor vma.
305 * free memory for segment if it is marked destroyed.
306 * The descriptor has already been removed from the current->mm->mmap list
307 * and will later be kfree()d.
309 static void shm_close(struct vm_area_struct *vma)
311 struct file *file = vma->vm_file;
312 struct shm_file_data *sfd = shm_file_data(file);
313 struct shmid_kernel *shp;
314 struct ipc_namespace *ns = sfd->ns;
316 down_write(&shm_ids(ns).rwsem);
317 /* remove from the list of attaches of the shm segment */
318 shp = shm_lock(ns, sfd->id);
321 * We raced in the idr lookup or with shm_destroy().
322 * Either way, the ID is busted.
324 if (WARN_ON_ONCE(IS_ERR(shp)))
325 goto done; /* no-op */
327 shp->shm_lprid = task_tgid_vnr(current);
328 shp->shm_dtim = get_seconds();
330 if (shm_may_destroy(shp))
331 shm_destroy(ns, shp);
335 up_write(&shm_ids(ns).rwsem);
338 /* Called with ns->shm_ids(ns).rwsem locked */
339 static int shm_try_destroy_orphaned(int id, void *p, void *data)
341 struct ipc_namespace *ns = data;
342 struct kern_ipc_perm *ipcp = p;
343 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
346 * We want to destroy segments without users and with already
347 * exit'ed originating process.
349 * As shp->* are changed under rwsem, it's safe to skip shp locking.
351 if (!list_empty(&shp->shm_clist))
354 if (shm_may_destroy(shp)) {
355 shm_lock_by_ptr(shp);
356 shm_destroy(ns, shp);
361 void shm_destroy_orphaned(struct ipc_namespace *ns)
363 down_write(&shm_ids(ns).rwsem);
364 if (shm_ids(ns).in_use)
365 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
366 up_write(&shm_ids(ns).rwsem);
369 /* Locking assumes this will only be called with task == current */
370 void exit_shm(struct task_struct *task)
373 struct shmid_kernel *shp;
374 struct ipc_namespace *ns;
378 if (list_empty(&task->sysvshm.shm_clist)) {
383 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
387 * 1) Get pointer to the ipc namespace. It is worth to say
388 * that this pointer is guaranteed to be valid because
389 * shp lifetime is always shorter than namespace lifetime
390 * in which shp lives.
391 * We taken task_lock it means that shp won't be freed.
396 * 2) If kernel.shm_rmid_forced is not set then only keep track of
397 * which shmids are orphaned, so that a later set of the sysctl
400 if (!ns->shm_rmid_forced)
401 goto unlink_continue;
404 * 3) get a reference to the namespace.
405 * The refcount could be already 0. If it is 0, then
406 * the shm objects will be free by free_ipc_work().
408 ns = get_ipc_ns_not_zero(ns);
411 list_del_init(&shp->shm_clist);
417 * 4) get a reference to shp.
418 * This cannot fail: shm_clist_rm() is called before
419 * ipc_rmid(), thus the refcount cannot be 0.
421 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
424 * 5) unlink the shm segment from the list of segments
425 * created by current.
426 * This must be done last. After unlinking,
427 * only the refcounts obtained above prevent IPC_RMID
428 * from destroying the segment or the namespace.
430 list_del_init(&shp->shm_clist);
435 * 6) we have all references
436 * Thus lock & if needed destroy shp.
438 down_write(&shm_ids(ns).rwsem);
439 shm_lock_by_ptr(shp);
441 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
442 * safe to call ipc_rcu_putref here
444 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
446 if (ipc_valid_object(&shp->shm_perm)) {
447 if (shm_may_destroy(shp))
448 shm_destroy(ns, shp);
453 * Someone else deleted the shp from namespace
454 * idr/kht while we have waited.
455 * Just unlock and continue.
460 up_write(&shm_ids(ns).rwsem);
461 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
465 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
467 struct file *file = vma->vm_file;
468 struct shm_file_data *sfd = shm_file_data(file);
470 return sfd->vm_ops->fault(vma, vmf);
474 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
476 struct file *file = vma->vm_file;
477 struct shm_file_data *sfd = shm_file_data(file);
479 if (sfd->vm_ops->set_policy)
480 err = sfd->vm_ops->set_policy(vma, new);
484 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
487 struct file *file = vma->vm_file;
488 struct shm_file_data *sfd = shm_file_data(file);
489 struct mempolicy *pol = NULL;
491 if (sfd->vm_ops->get_policy)
492 pol = sfd->vm_ops->get_policy(vma, addr);
493 else if (vma->vm_policy)
494 pol = vma->vm_policy;
500 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
502 struct shm_file_data *sfd = shm_file_data(file);
506 * In case of remap_file_pages() emulation, the file can represent an
507 * IPC ID that was removed, and possibly even reused by another shm
508 * segment already. Propagate this case as an error to caller.
510 ret =__shm_open(vma);
514 ret = sfd->file->f_op->mmap(sfd->file, vma);
519 sfd->vm_ops = vma->vm_ops;
521 WARN_ON(!sfd->vm_ops->fault);
523 vma->vm_ops = &shm_vm_ops;
527 static int shm_release(struct inode *ino, struct file *file)
529 struct shm_file_data *sfd = shm_file_data(file);
533 shm_file_data(file) = NULL;
538 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
540 struct shm_file_data *sfd = shm_file_data(file);
542 if (!sfd->file->f_op->fsync)
544 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
547 static long shm_fallocate(struct file *file, int mode, loff_t offset,
550 struct shm_file_data *sfd = shm_file_data(file);
552 if (!sfd->file->f_op->fallocate)
554 return sfd->file->f_op->fallocate(file, mode, offset, len);
557 static unsigned long shm_get_unmapped_area(struct file *file,
558 unsigned long addr, unsigned long len, unsigned long pgoff,
561 struct shm_file_data *sfd = shm_file_data(file);
562 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
566 static const struct file_operations shm_file_operations = {
569 .release = shm_release,
571 .get_unmapped_area = shm_get_unmapped_area,
573 .llseek = noop_llseek,
574 .fallocate = shm_fallocate,
577 static const struct file_operations shm_file_operations_huge = {
580 .release = shm_release,
581 .get_unmapped_area = shm_get_unmapped_area,
582 .llseek = noop_llseek,
583 .fallocate = shm_fallocate,
586 int is_file_shm_hugepages(struct file *file)
588 return file->f_op == &shm_file_operations_huge;
591 static const struct vm_operations_struct shm_vm_ops = {
592 .open = shm_open, /* callback for a new vm-area open */
593 .close = shm_close, /* callback for when the vm-area is released */
595 #if defined(CONFIG_NUMA)
596 .set_policy = shm_set_policy,
597 .get_policy = shm_get_policy,
602 * newseg - Create a new shared memory segment
604 * @params: ptr to the structure that contains key, size and shmflg
606 * Called with shm_ids.rwsem held as a writer.
608 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
610 key_t key = params->key;
611 int shmflg = params->flg;
612 size_t size = params->u.size;
614 struct shmid_kernel *shp;
615 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
619 vm_flags_t acctflag = 0;
621 if (size < SHMMIN || size > ns->shm_ctlmax)
624 if (numpages << PAGE_SHIFT < size)
627 if (ns->shm_tot + numpages < ns->shm_tot ||
628 ns->shm_tot + numpages > ns->shm_ctlall)
631 shp = ipc_rcu_alloc(sizeof(*shp));
635 shp->shm_perm.key = key;
636 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
637 shp->mlock_user = NULL;
639 shp->shm_perm.security = NULL;
640 error = security_shm_alloc(shp);
642 ipc_rcu_putref(shp, ipc_rcu_free);
646 sprintf(name, "SYSV%08x", key);
647 if (shmflg & SHM_HUGETLB) {
651 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
656 hugesize = ALIGN(size, huge_page_size(hs));
658 /* hugetlb_file_setup applies strict accounting */
659 if (shmflg & SHM_NORESERVE)
660 acctflag = VM_NORESERVE;
661 file = hugetlb_file_setup(name, hugesize, acctflag,
662 &shp->mlock_user, HUGETLB_SHMFS_INODE,
663 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
666 * Do not allow no accounting for OVERCOMMIT_NEVER, even
669 if ((shmflg & SHM_NORESERVE) &&
670 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
671 acctflag = VM_NORESERVE;
672 file = shmem_kernel_file_setup(name, size, acctflag);
674 error = PTR_ERR(file);
678 shp->shm_cprid = task_tgid_vnr(current);
680 shp->shm_atim = shp->shm_dtim = 0;
681 shp->shm_ctim = get_seconds();
682 shp->shm_segsz = size;
684 shp->shm_file = file;
685 shp->shm_creator = current;
687 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
696 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
697 task_unlock(current);
700 * shmid gets reported as "inode#" in /proc/pid/maps.
701 * proc-ps tools use this. Changing this will break them.
703 file_inode(file)->i_ino = shp->shm_perm.id;
705 ns->shm_tot += numpages;
706 error = shp->shm_perm.id;
708 ipc_unlock_object(&shp->shm_perm);
713 if (is_file_hugepages(file) && shp->mlock_user)
714 user_shm_unlock(size, shp->mlock_user);
717 ipc_rcu_putref(shp, shm_rcu_free);
722 * Called with shm_ids.rwsem and ipcp locked.
724 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
726 struct shmid_kernel *shp;
728 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
729 return security_shm_associate(shp, shmflg);
733 * Called with shm_ids.rwsem and ipcp locked.
735 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
736 struct ipc_params *params)
738 struct shmid_kernel *shp;
740 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
741 if (shp->shm_segsz < params->u.size)
747 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
749 struct ipc_namespace *ns;
750 static const struct ipc_ops shm_ops = {
752 .associate = shm_security,
753 .more_checks = shm_more_checks,
755 struct ipc_params shm_params;
757 ns = current->nsproxy->ipc_ns;
759 shm_params.key = key;
760 shm_params.flg = shmflg;
761 shm_params.u.size = size;
763 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
766 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
770 return copy_to_user(buf, in, sizeof(*in));
775 memset(&out, 0, sizeof(out));
776 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
777 out.shm_segsz = in->shm_segsz;
778 out.shm_atime = in->shm_atime;
779 out.shm_dtime = in->shm_dtime;
780 out.shm_ctime = in->shm_ctime;
781 out.shm_cpid = in->shm_cpid;
782 out.shm_lpid = in->shm_lpid;
783 out.shm_nattch = in->shm_nattch;
785 return copy_to_user(buf, &out, sizeof(out));
792 static inline unsigned long
793 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
797 if (copy_from_user(out, buf, sizeof(*out)))
802 struct shmid_ds tbuf_old;
804 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
807 out->shm_perm.uid = tbuf_old.shm_perm.uid;
808 out->shm_perm.gid = tbuf_old.shm_perm.gid;
809 out->shm_perm.mode = tbuf_old.shm_perm.mode;
818 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
822 return copy_to_user(buf, in, sizeof(*in));
827 if (in->shmmax > INT_MAX)
828 out.shmmax = INT_MAX;
830 out.shmmax = (int)in->shmmax;
832 out.shmmin = in->shmmin;
833 out.shmmni = in->shmmni;
834 out.shmseg = in->shmseg;
835 out.shmall = in->shmall;
837 return copy_to_user(buf, &out, sizeof(out));
845 * Calculate and add used RSS and swap pages of a shm.
846 * Called with shm_ids.rwsem held as a reader
848 static void shm_add_rss_swap(struct shmid_kernel *shp,
849 unsigned long *rss_add, unsigned long *swp_add)
853 inode = file_inode(shp->shm_file);
855 if (is_file_hugepages(shp->shm_file)) {
856 struct address_space *mapping = inode->i_mapping;
857 struct hstate *h = hstate_file(shp->shm_file);
858 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
861 struct shmem_inode_info *info = SHMEM_I(inode);
862 spin_lock(&info->lock);
863 *rss_add += inode->i_mapping->nrpages;
864 *swp_add += info->swapped;
865 spin_unlock(&info->lock);
867 *rss_add += inode->i_mapping->nrpages;
873 * Called with shm_ids.rwsem held as a reader
875 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
884 in_use = shm_ids(ns).in_use;
886 for (total = 0, next_id = 0; total < in_use; next_id++) {
887 struct kern_ipc_perm *ipc;
888 struct shmid_kernel *shp;
890 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
893 shp = container_of(ipc, struct shmid_kernel, shm_perm);
895 shm_add_rss_swap(shp, rss, swp);
902 * This function handles some shmctl commands which require the rwsem
903 * to be held in write mode.
904 * NOTE: no locks must be held, the rwsem is taken inside this function.
906 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
907 struct shmid_ds __user *buf, int version)
909 struct kern_ipc_perm *ipcp;
910 struct shmid64_ds shmid64;
911 struct shmid_kernel *shp;
914 if (cmd == IPC_SET) {
915 if (copy_shmid_from_user(&shmid64, buf, version))
919 down_write(&shm_ids(ns).rwsem);
922 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
923 &shmid64.shm_perm, 0);
929 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
931 err = security_shm_shmctl(shp, cmd);
937 ipc_lock_object(&shp->shm_perm);
938 /* do_shm_rmid unlocks the ipc object and rcu */
939 do_shm_rmid(ns, ipcp);
942 ipc_lock_object(&shp->shm_perm);
943 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
946 shp->shm_ctim = get_seconds();
954 ipc_unlock_object(&shp->shm_perm);
958 up_write(&shm_ids(ns).rwsem);
962 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
963 int cmd, int version, void __user *buf)
966 struct shmid_kernel *shp;
968 /* preliminary security checks for *_INFO */
969 if (cmd == IPC_INFO || cmd == SHM_INFO) {
970 err = security_shm_shmctl(NULL, cmd);
978 struct shminfo64 shminfo;
980 memset(&shminfo, 0, sizeof(shminfo));
981 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
982 shminfo.shmmax = ns->shm_ctlmax;
983 shminfo.shmall = ns->shm_ctlall;
985 shminfo.shmmin = SHMMIN;
986 if (copy_shminfo_to_user(buf, &shminfo, version))
989 down_read(&shm_ids(ns).rwsem);
990 err = ipc_get_maxid(&shm_ids(ns));
991 up_read(&shm_ids(ns).rwsem);
999 struct shm_info shm_info;
1001 memset(&shm_info, 0, sizeof(shm_info));
1002 down_read(&shm_ids(ns).rwsem);
1003 shm_info.used_ids = shm_ids(ns).in_use;
1004 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
1005 shm_info.shm_tot = ns->shm_tot;
1006 shm_info.swap_attempts = 0;
1007 shm_info.swap_successes = 0;
1008 err = ipc_get_maxid(&shm_ids(ns));
1009 up_read(&shm_ids(ns).rwsem);
1010 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
1015 err = err < 0 ? 0 : err;
1021 struct shmid64_ds tbuf;
1025 if (cmd == SHM_STAT) {
1026 shp = shm_obtain_object(ns, shmid);
1031 result = shp->shm_perm.id;
1033 shp = shm_obtain_object_check(ns, shmid);
1042 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1045 err = security_shm_shmctl(shp, cmd);
1049 memset(&tbuf, 0, sizeof(tbuf));
1050 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
1051 tbuf.shm_segsz = shp->shm_segsz;
1052 tbuf.shm_atime = shp->shm_atim;
1053 tbuf.shm_dtime = shp->shm_dtim;
1054 tbuf.shm_ctime = shp->shm_ctim;
1055 tbuf.shm_cpid = shp->shm_cprid;
1056 tbuf.shm_lpid = shp->shm_lprid;
1057 tbuf.shm_nattch = shp->shm_nattch;
1060 if (copy_shmid_to_user(buf, &tbuf, version))
1076 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1078 struct shmid_kernel *shp;
1080 struct ipc_namespace *ns;
1082 if (cmd < 0 || shmid < 0)
1085 version = ipc_parse_version(&cmd);
1086 ns = current->nsproxy->ipc_ns;
1093 return shmctl_nolock(ns, shmid, cmd, version, buf);
1096 return shmctl_down(ns, shmid, cmd, buf, version);
1100 struct file *shm_file;
1103 shp = shm_obtain_object_check(ns, shmid);
1109 audit_ipc_obj(&(shp->shm_perm));
1110 err = security_shm_shmctl(shp, cmd);
1114 ipc_lock_object(&shp->shm_perm);
1116 /* check if shm_destroy() is tearing down shp */
1117 if (!ipc_valid_object(&shp->shm_perm)) {
1122 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1123 kuid_t euid = current_euid();
1124 if (!uid_eq(euid, shp->shm_perm.uid) &&
1125 !uid_eq(euid, shp->shm_perm.cuid)) {
1129 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1135 shm_file = shp->shm_file;
1136 if (is_file_hugepages(shm_file))
1139 if (cmd == SHM_LOCK) {
1140 struct user_struct *user = current_user();
1141 err = shmem_lock(shm_file, 1, user);
1142 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1143 shp->shm_perm.mode |= SHM_LOCKED;
1144 shp->mlock_user = user;
1150 if (!(shp->shm_perm.mode & SHM_LOCKED))
1152 shmem_lock(shm_file, 0, shp->mlock_user);
1153 shp->shm_perm.mode &= ~SHM_LOCKED;
1154 shp->mlock_user = NULL;
1156 ipc_unlock_object(&shp->shm_perm);
1158 shmem_unlock_mapping(shm_file->f_mapping);
1168 ipc_unlock_object(&shp->shm_perm);
1175 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1177 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1178 * "raddr" thing points to kernel space, and there has to be a wrapper around
1181 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1182 ulong *raddr, unsigned long shmlba)
1184 struct shmid_kernel *shp;
1189 unsigned long flags;
1192 struct ipc_namespace *ns;
1193 struct shm_file_data *sfd;
1196 unsigned long populate = 0;
1201 else if ((addr = (ulong)shmaddr)) {
1202 if (addr & (shmlba - 1)) {
1203 if (shmflg & SHM_RND) {
1204 addr &= ~(shmlba - 1); /* round down */
1207 * Ensure that the round-down is non-nil
1208 * when remapping. This can happen for
1209 * cases when addr < shmlba.
1211 if (!addr && (shmflg & SHM_REMAP))
1214 #ifndef __ARCH_FORCE_SHMLBA
1215 if (addr & ~PAGE_MASK)
1219 flags = MAP_SHARED | MAP_FIXED;
1221 if ((shmflg & SHM_REMAP))
1227 if (shmflg & SHM_RDONLY) {
1230 f_mode = FMODE_READ;
1232 prot = PROT_READ | PROT_WRITE;
1233 acc_mode = S_IRUGO | S_IWUGO;
1234 f_mode = FMODE_READ | FMODE_WRITE;
1236 if (shmflg & SHM_EXEC) {
1238 acc_mode |= S_IXUGO;
1242 * We cannot rely on the fs check since SYSV IPC does have an
1243 * additional creator id...
1245 ns = current->nsproxy->ipc_ns;
1247 shp = shm_obtain_object_check(ns, shmid);
1254 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1257 err = security_shm_shmat(shp, shmaddr, shmflg);
1261 ipc_lock_object(&shp->shm_perm);
1263 /* check if shm_destroy() is tearing down shp */
1264 if (!ipc_valid_object(&shp->shm_perm)) {
1265 ipc_unlock_object(&shp->shm_perm);
1270 path = shp->shm_file->f_path;
1273 size = i_size_read(d_inode(path.dentry));
1274 ipc_unlock_object(&shp->shm_perm);
1278 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1284 file = alloc_file(&path, f_mode,
1285 is_file_hugepages(shp->shm_file) ?
1286 &shm_file_operations_huge :
1287 &shm_file_operations);
1288 err = PTR_ERR(file);
1295 file->private_data = sfd;
1296 file->f_mapping = shp->shm_file->f_mapping;
1297 sfd->id = shp->shm_perm.id;
1298 sfd->ns = get_ipc_ns(ns);
1300 * We need to take a reference to the real shm file to prevent the
1301 * pointer from becoming stale in cases where the lifetime of the outer
1302 * file extends beyond that of the shm segment. It's not usually
1303 * possible, but it can happen during remap_file_pages() emulation as
1304 * that unmaps the memory, then does ->mmap() via file reference only.
1305 * We'll deny the ->mmap() if the shm segment was since removed, but to
1306 * detect shm ID reuse we need to compare the file pointers.
1308 sfd->file = get_file(shp->shm_file);
1311 err = security_mmap_file(file, prot, flags);
1315 down_write(¤t->mm->mmap_sem);
1316 if (addr && !(shmflg & SHM_REMAP)) {
1318 if (addr + size < addr)
1321 if (find_vma_intersection(current->mm, addr, addr + size))
1325 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1328 if (IS_ERR_VALUE(addr))
1331 up_write(¤t->mm->mmap_sem);
1333 mm_populate(addr, populate);
1339 down_write(&shm_ids(ns).rwsem);
1340 shp = shm_lock(ns, shmid);
1343 if (shm_may_destroy(shp))
1344 shm_destroy(ns, shp);
1347 up_write(&shm_ids(ns).rwsem);
1356 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1361 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1364 force_successful_syscall_return();
1369 * detach and kill segment if marked destroyed.
1370 * The work is done in shm_close.
1372 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1374 struct mm_struct *mm = current->mm;
1375 struct vm_area_struct *vma;
1376 unsigned long addr = (unsigned long)shmaddr;
1377 int retval = -EINVAL;
1381 struct vm_area_struct *next;
1384 if (addr & ~PAGE_MASK)
1387 down_write(&mm->mmap_sem);
1390 * This function tries to be smart and unmap shm segments that
1391 * were modified by partial mlock or munmap calls:
1392 * - It first determines the size of the shm segment that should be
1393 * unmapped: It searches for a vma that is backed by shm and that
1394 * started at address shmaddr. It records it's size and then unmaps
1396 * - Then it unmaps all shm vmas that started at shmaddr and that
1397 * are within the initially determined size and that are from the
1398 * same shm segment from which we determined the size.
1399 * Errors from do_munmap are ignored: the function only fails if
1400 * it's called with invalid parameters or if it's called to unmap
1401 * a part of a vma. Both calls in this function are for full vmas,
1402 * the parameters are directly copied from the vma itself and always
1403 * valid - therefore do_munmap cannot fail. (famous last words?)
1406 * If it had been mremap()'d, the starting address would not
1407 * match the usual checks anyway. So assume all vma's are
1408 * above the starting address given.
1410 vma = find_vma(mm, addr);
1414 next = vma->vm_next;
1417 * Check if the starting address would match, i.e. it's
1418 * a fragment created by mprotect() and/or munmap(), or it
1419 * otherwise it starts at this address with no hassles.
1421 if ((vma->vm_ops == &shm_vm_ops) &&
1422 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1425 * Record the file of the shm segment being
1426 * unmapped. With mremap(), someone could place
1427 * page from another segment but with equal offsets
1428 * in the range we are unmapping.
1430 file = vma->vm_file;
1431 size = i_size_read(file_inode(vma->vm_file));
1432 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1434 * We discovered the size of the shm segment, so
1435 * break out of here and fall through to the next
1436 * loop that uses the size information to stop
1437 * searching for matching vma's.
1447 * We need look no further than the maximum address a fragment
1448 * could possibly have landed at. Also cast things to loff_t to
1449 * prevent overflows and make comparisons vs. equal-width types.
1451 size = PAGE_ALIGN(size);
1452 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1453 next = vma->vm_next;
1455 /* finding a matching vma now does not alter retval */
1456 if ((vma->vm_ops == &shm_vm_ops) &&
1457 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1458 (vma->vm_file == file))
1459 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1463 #else /* CONFIG_MMU */
1464 /* under NOMMU conditions, the exact address to be destroyed must be
1466 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1467 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1473 up_write(&mm->mmap_sem);
1477 #ifdef CONFIG_PROC_FS
1478 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1480 struct user_namespace *user_ns = seq_user_ns(s);
1481 struct shmid_kernel *shp = it;
1482 unsigned long rss = 0, swp = 0;
1484 shm_add_rss_swap(shp, &rss, &swp);
1486 #if BITS_PER_LONG <= 32
1487 #define SIZE_SPEC "%10lu"
1489 #define SIZE_SPEC "%21lu"
1493 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1494 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1495 SIZE_SPEC " " SIZE_SPEC "\n",
1503 from_kuid_munged(user_ns, shp->shm_perm.uid),
1504 from_kgid_munged(user_ns, shp->shm_perm.gid),
1505 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1506 from_kgid_munged(user_ns, shp->shm_perm.cgid),