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);
473 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
475 struct file *file = vma->vm_file;
476 struct shm_file_data *sfd = shm_file_data(file);
478 if (sfd->vm_ops && sfd->vm_ops->split)
479 return sfd->vm_ops->split(vma, addr);
485 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
487 struct file *file = vma->vm_file;
488 struct shm_file_data *sfd = shm_file_data(file);
490 if (sfd->vm_ops->set_policy)
491 err = sfd->vm_ops->set_policy(vma, new);
495 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
498 struct file *file = vma->vm_file;
499 struct shm_file_data *sfd = shm_file_data(file);
500 struct mempolicy *pol = NULL;
502 if (sfd->vm_ops->get_policy)
503 pol = sfd->vm_ops->get_policy(vma, addr);
504 else if (vma->vm_policy)
505 pol = vma->vm_policy;
511 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
513 struct shm_file_data *sfd = shm_file_data(file);
517 * In case of remap_file_pages() emulation, the file can represent an
518 * IPC ID that was removed, and possibly even reused by another shm
519 * segment already. Propagate this case as an error to caller.
521 ret =__shm_open(vma);
525 ret = sfd->file->f_op->mmap(sfd->file, vma);
530 sfd->vm_ops = vma->vm_ops;
532 WARN_ON(!sfd->vm_ops->fault);
534 vma->vm_ops = &shm_vm_ops;
538 static int shm_release(struct inode *ino, struct file *file)
540 struct shm_file_data *sfd = shm_file_data(file);
544 shm_file_data(file) = NULL;
549 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
551 struct shm_file_data *sfd = shm_file_data(file);
553 if (!sfd->file->f_op->fsync)
555 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
558 static long shm_fallocate(struct file *file, int mode, loff_t offset,
561 struct shm_file_data *sfd = shm_file_data(file);
563 if (!sfd->file->f_op->fallocate)
565 return sfd->file->f_op->fallocate(file, mode, offset, len);
568 static unsigned long shm_get_unmapped_area(struct file *file,
569 unsigned long addr, unsigned long len, unsigned long pgoff,
572 struct shm_file_data *sfd = shm_file_data(file);
573 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
577 static const struct file_operations shm_file_operations = {
580 .release = shm_release,
581 .get_unmapped_area = shm_get_unmapped_area,
582 .llseek = noop_llseek,
583 .fallocate = shm_fallocate,
587 * shm_file_operations_huge is now identical to shm_file_operations,
588 * but we keep it distinct for the sake of is_file_shm_hugepages().
590 static const struct file_operations shm_file_operations_huge = {
593 .release = shm_release,
594 .get_unmapped_area = shm_get_unmapped_area,
595 .llseek = noop_llseek,
596 .fallocate = shm_fallocate,
599 bool is_file_shm_hugepages(struct file *file)
601 return file->f_op == &shm_file_operations_huge;
604 static const struct vm_operations_struct shm_vm_ops = {
605 .open = shm_open, /* callback for a new vm-area open */
606 .close = shm_close, /* callback for when the vm-area is released */
609 #if defined(CONFIG_NUMA)
610 .set_policy = shm_set_policy,
611 .get_policy = shm_get_policy,
616 * newseg - Create a new shared memory segment
618 * @params: ptr to the structure that contains key, size and shmflg
620 * Called with shm_ids.rwsem held as a writer.
622 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
624 key_t key = params->key;
625 int shmflg = params->flg;
626 size_t size = params->u.size;
628 struct shmid_kernel *shp;
629 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
633 vm_flags_t acctflag = 0;
635 if (size < SHMMIN || size > ns->shm_ctlmax)
638 if (numpages << PAGE_SHIFT < size)
641 if (ns->shm_tot + numpages < ns->shm_tot ||
642 ns->shm_tot + numpages > ns->shm_ctlall)
645 shp = ipc_rcu_alloc(sizeof(*shp));
649 shp->shm_perm.key = key;
650 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
651 shp->mlock_user = NULL;
653 shp->shm_perm.security = NULL;
654 error = security_shm_alloc(shp);
656 ipc_rcu_putref(shp, ipc_rcu_free);
660 sprintf(name, "SYSV%08x", key);
661 if (shmflg & SHM_HUGETLB) {
665 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
670 hugesize = ALIGN(size, huge_page_size(hs));
672 /* hugetlb_file_setup applies strict accounting */
673 if (shmflg & SHM_NORESERVE)
674 acctflag = VM_NORESERVE;
675 file = hugetlb_file_setup(name, hugesize, acctflag,
676 &shp->mlock_user, HUGETLB_SHMFS_INODE,
677 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
680 * Do not allow no accounting for OVERCOMMIT_NEVER, even
683 if ((shmflg & SHM_NORESERVE) &&
684 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
685 acctflag = VM_NORESERVE;
686 file = shmem_kernel_file_setup(name, size, acctflag);
688 error = PTR_ERR(file);
692 shp->shm_cprid = task_tgid_vnr(current);
694 shp->shm_atim = shp->shm_dtim = 0;
695 shp->shm_ctim = get_seconds();
696 shp->shm_segsz = size;
698 shp->shm_file = file;
699 shp->shm_creator = current;
701 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
710 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
711 task_unlock(current);
714 * shmid gets reported as "inode#" in /proc/pid/maps.
715 * proc-ps tools use this. Changing this will break them.
717 file_inode(file)->i_ino = shp->shm_perm.id;
719 ns->shm_tot += numpages;
720 error = shp->shm_perm.id;
722 ipc_unlock_object(&shp->shm_perm);
727 if (is_file_hugepages(file) && shp->mlock_user)
728 user_shm_unlock(size, shp->mlock_user);
731 ipc_rcu_putref(shp, shm_rcu_free);
736 * Called with shm_ids.rwsem and ipcp locked.
738 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
740 struct shmid_kernel *shp;
742 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
743 return security_shm_associate(shp, shmflg);
747 * Called with shm_ids.rwsem and ipcp locked.
749 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
750 struct ipc_params *params)
752 struct shmid_kernel *shp;
754 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
755 if (shp->shm_segsz < params->u.size)
761 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
763 struct ipc_namespace *ns;
764 static const struct ipc_ops shm_ops = {
766 .associate = shm_security,
767 .more_checks = shm_more_checks,
769 struct ipc_params shm_params;
771 ns = current->nsproxy->ipc_ns;
773 shm_params.key = key;
774 shm_params.flg = shmflg;
775 shm_params.u.size = size;
777 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
780 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
784 return copy_to_user(buf, in, sizeof(*in));
789 memset(&out, 0, sizeof(out));
790 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
791 out.shm_segsz = in->shm_segsz;
792 out.shm_atime = in->shm_atime;
793 out.shm_dtime = in->shm_dtime;
794 out.shm_ctime = in->shm_ctime;
795 out.shm_cpid = in->shm_cpid;
796 out.shm_lpid = in->shm_lpid;
797 out.shm_nattch = in->shm_nattch;
799 return copy_to_user(buf, &out, sizeof(out));
806 static inline unsigned long
807 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
811 if (copy_from_user(out, buf, sizeof(*out)))
816 struct shmid_ds tbuf_old;
818 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
821 out->shm_perm.uid = tbuf_old.shm_perm.uid;
822 out->shm_perm.gid = tbuf_old.shm_perm.gid;
823 out->shm_perm.mode = tbuf_old.shm_perm.mode;
832 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
836 return copy_to_user(buf, in, sizeof(*in));
841 if (in->shmmax > INT_MAX)
842 out.shmmax = INT_MAX;
844 out.shmmax = (int)in->shmmax;
846 out.shmmin = in->shmmin;
847 out.shmmni = in->shmmni;
848 out.shmseg = in->shmseg;
849 out.shmall = in->shmall;
851 return copy_to_user(buf, &out, sizeof(out));
859 * Calculate and add used RSS and swap pages of a shm.
860 * Called with shm_ids.rwsem held as a reader
862 static void shm_add_rss_swap(struct shmid_kernel *shp,
863 unsigned long *rss_add, unsigned long *swp_add)
867 inode = file_inode(shp->shm_file);
869 if (is_file_hugepages(shp->shm_file)) {
870 struct address_space *mapping = inode->i_mapping;
871 struct hstate *h = hstate_file(shp->shm_file);
872 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
875 struct shmem_inode_info *info = SHMEM_I(inode);
876 spin_lock_irq(&info->lock);
877 *rss_add += inode->i_mapping->nrpages;
878 *swp_add += info->swapped;
879 spin_unlock_irq(&info->lock);
881 *rss_add += inode->i_mapping->nrpages;
887 * Called with shm_ids.rwsem held as a reader
889 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
898 in_use = shm_ids(ns).in_use;
900 for (total = 0, next_id = 0; total < in_use; next_id++) {
901 struct kern_ipc_perm *ipc;
902 struct shmid_kernel *shp;
904 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
907 shp = container_of(ipc, struct shmid_kernel, shm_perm);
909 shm_add_rss_swap(shp, rss, swp);
916 * This function handles some shmctl commands which require the rwsem
917 * to be held in write mode.
918 * NOTE: no locks must be held, the rwsem is taken inside this function.
920 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
921 struct shmid_ds __user *buf, int version)
923 struct kern_ipc_perm *ipcp;
924 struct shmid64_ds shmid64;
925 struct shmid_kernel *shp;
928 if (cmd == IPC_SET) {
929 if (copy_shmid_from_user(&shmid64, buf, version))
933 down_write(&shm_ids(ns).rwsem);
936 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
937 &shmid64.shm_perm, 0);
943 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
945 err = security_shm_shmctl(shp, cmd);
951 ipc_lock_object(&shp->shm_perm);
952 /* do_shm_rmid unlocks the ipc object and rcu */
953 do_shm_rmid(ns, ipcp);
956 ipc_lock_object(&shp->shm_perm);
957 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
960 shp->shm_ctim = get_seconds();
968 ipc_unlock_object(&shp->shm_perm);
972 up_write(&shm_ids(ns).rwsem);
976 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
977 int cmd, int version, void __user *buf)
980 struct shmid_kernel *shp;
982 /* preliminary security checks for *_INFO */
983 if (cmd == IPC_INFO || cmd == SHM_INFO) {
984 err = security_shm_shmctl(NULL, cmd);
992 struct shminfo64 shminfo;
994 memset(&shminfo, 0, sizeof(shminfo));
995 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
996 shminfo.shmmax = ns->shm_ctlmax;
997 shminfo.shmall = ns->shm_ctlall;
999 shminfo.shmmin = SHMMIN;
1000 if (copy_shminfo_to_user(buf, &shminfo, version))
1003 down_read(&shm_ids(ns).rwsem);
1004 err = ipc_get_maxid(&shm_ids(ns));
1005 up_read(&shm_ids(ns).rwsem);
1013 struct shm_info shm_info;
1015 memset(&shm_info, 0, sizeof(shm_info));
1016 down_read(&shm_ids(ns).rwsem);
1017 shm_info.used_ids = shm_ids(ns).in_use;
1018 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
1019 shm_info.shm_tot = ns->shm_tot;
1020 shm_info.swap_attempts = 0;
1021 shm_info.swap_successes = 0;
1022 err = ipc_get_maxid(&shm_ids(ns));
1023 up_read(&shm_ids(ns).rwsem);
1024 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
1029 err = err < 0 ? 0 : err;
1035 struct shmid64_ds tbuf;
1039 if (cmd == SHM_STAT) {
1040 shp = shm_obtain_object(ns, shmid);
1045 result = shp->shm_perm.id;
1047 shp = shm_obtain_object_check(ns, shmid);
1056 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1059 err = security_shm_shmctl(shp, cmd);
1063 memset(&tbuf, 0, sizeof(tbuf));
1064 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
1065 tbuf.shm_segsz = shp->shm_segsz;
1066 tbuf.shm_atime = shp->shm_atim;
1067 tbuf.shm_dtime = shp->shm_dtim;
1068 tbuf.shm_ctime = shp->shm_ctim;
1069 tbuf.shm_cpid = shp->shm_cprid;
1070 tbuf.shm_lpid = shp->shm_lprid;
1071 tbuf.shm_nattch = shp->shm_nattch;
1074 if (copy_shmid_to_user(buf, &tbuf, version))
1090 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1092 struct shmid_kernel *shp;
1094 struct ipc_namespace *ns;
1096 if (cmd < 0 || shmid < 0)
1099 version = ipc_parse_version(&cmd);
1100 ns = current->nsproxy->ipc_ns;
1107 return shmctl_nolock(ns, shmid, cmd, version, buf);
1110 return shmctl_down(ns, shmid, cmd, buf, version);
1114 struct file *shm_file;
1117 shp = shm_obtain_object_check(ns, shmid);
1123 audit_ipc_obj(&(shp->shm_perm));
1124 err = security_shm_shmctl(shp, cmd);
1128 ipc_lock_object(&shp->shm_perm);
1130 /* check if shm_destroy() is tearing down shp */
1131 if (!ipc_valid_object(&shp->shm_perm)) {
1136 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1137 kuid_t euid = current_euid();
1138 if (!uid_eq(euid, shp->shm_perm.uid) &&
1139 !uid_eq(euid, shp->shm_perm.cuid)) {
1143 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1149 shm_file = shp->shm_file;
1150 if (is_file_hugepages(shm_file))
1153 if (cmd == SHM_LOCK) {
1154 struct user_struct *user = current_user();
1155 err = shmem_lock(shm_file, 1, user);
1156 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1157 shp->shm_perm.mode |= SHM_LOCKED;
1158 shp->mlock_user = user;
1164 if (!(shp->shm_perm.mode & SHM_LOCKED))
1166 shmem_lock(shm_file, 0, shp->mlock_user);
1167 shp->shm_perm.mode &= ~SHM_LOCKED;
1168 shp->mlock_user = NULL;
1170 ipc_unlock_object(&shp->shm_perm);
1172 shmem_unlock_mapping(shm_file->f_mapping);
1182 ipc_unlock_object(&shp->shm_perm);
1189 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1191 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1192 * "raddr" thing points to kernel space, and there has to be a wrapper around
1195 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1196 ulong *raddr, unsigned long shmlba)
1198 struct shmid_kernel *shp;
1203 unsigned long flags;
1206 struct ipc_namespace *ns;
1207 struct shm_file_data *sfd;
1210 unsigned long populate = 0;
1215 else if ((addr = (ulong)shmaddr)) {
1216 if (addr & (shmlba - 1)) {
1217 if (shmflg & SHM_RND) {
1218 addr &= ~(shmlba - 1); /* round down */
1221 * Ensure that the round-down is non-nil
1222 * when remapping. This can happen for
1223 * cases when addr < shmlba.
1225 if (!addr && (shmflg & SHM_REMAP))
1228 #ifndef __ARCH_FORCE_SHMLBA
1229 if (addr & ~PAGE_MASK)
1233 flags = MAP_SHARED | MAP_FIXED;
1235 if ((shmflg & SHM_REMAP))
1241 if (shmflg & SHM_RDONLY) {
1244 f_mode = FMODE_READ;
1246 prot = PROT_READ | PROT_WRITE;
1247 acc_mode = S_IRUGO | S_IWUGO;
1248 f_mode = FMODE_READ | FMODE_WRITE;
1250 if (shmflg & SHM_EXEC) {
1252 acc_mode |= S_IXUGO;
1256 * We cannot rely on the fs check since SYSV IPC does have an
1257 * additional creator id...
1259 ns = current->nsproxy->ipc_ns;
1261 shp = shm_obtain_object_check(ns, shmid);
1268 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1271 err = security_shm_shmat(shp, shmaddr, shmflg);
1275 ipc_lock_object(&shp->shm_perm);
1277 /* check if shm_destroy() is tearing down shp */
1278 if (!ipc_valid_object(&shp->shm_perm)) {
1279 ipc_unlock_object(&shp->shm_perm);
1284 path = shp->shm_file->f_path;
1287 size = i_size_read(d_inode(path.dentry));
1288 ipc_unlock_object(&shp->shm_perm);
1292 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1298 file = alloc_file(&path, f_mode,
1299 is_file_hugepages(shp->shm_file) ?
1300 &shm_file_operations_huge :
1301 &shm_file_operations);
1302 err = PTR_ERR(file);
1309 file->private_data = sfd;
1310 file->f_mapping = shp->shm_file->f_mapping;
1311 sfd->id = shp->shm_perm.id;
1312 sfd->ns = get_ipc_ns(ns);
1314 * We need to take a reference to the real shm file to prevent the
1315 * pointer from becoming stale in cases where the lifetime of the outer
1316 * file extends beyond that of the shm segment. It's not usually
1317 * possible, but it can happen during remap_file_pages() emulation as
1318 * that unmaps the memory, then does ->mmap() via file reference only.
1319 * We'll deny the ->mmap() if the shm segment was since removed, but to
1320 * detect shm ID reuse we need to compare the file pointers.
1322 sfd->file = get_file(shp->shm_file);
1325 err = security_mmap_file(file, prot, flags);
1329 if (down_write_killable(¤t->mm->mmap_sem)) {
1334 if (addr && !(shmflg & SHM_REMAP)) {
1336 if (addr + size < addr)
1339 if (find_vma_intersection(current->mm, addr, addr + size))
1343 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1346 if (IS_ERR_VALUE(addr))
1349 up_write(¤t->mm->mmap_sem);
1351 mm_populate(addr, populate);
1357 down_write(&shm_ids(ns).rwsem);
1358 shp = shm_lock(ns, shmid);
1361 if (shm_may_destroy(shp))
1362 shm_destroy(ns, shp);
1365 up_write(&shm_ids(ns).rwsem);
1374 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1379 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1382 force_successful_syscall_return();
1387 * detach and kill segment if marked destroyed.
1388 * The work is done in shm_close.
1390 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1392 struct mm_struct *mm = current->mm;
1393 struct vm_area_struct *vma;
1394 unsigned long addr = (unsigned long)shmaddr;
1395 int retval = -EINVAL;
1399 struct vm_area_struct *next;
1402 if (addr & ~PAGE_MASK)
1405 if (down_write_killable(&mm->mmap_sem))
1409 * This function tries to be smart and unmap shm segments that
1410 * were modified by partial mlock or munmap calls:
1411 * - It first determines the size of the shm segment that should be
1412 * unmapped: It searches for a vma that is backed by shm and that
1413 * started at address shmaddr. It records it's size and then unmaps
1415 * - Then it unmaps all shm vmas that started at shmaddr and that
1416 * are within the initially determined size and that are from the
1417 * same shm segment from which we determined the size.
1418 * Errors from do_munmap are ignored: the function only fails if
1419 * it's called with invalid parameters or if it's called to unmap
1420 * a part of a vma. Both calls in this function are for full vmas,
1421 * the parameters are directly copied from the vma itself and always
1422 * valid - therefore do_munmap cannot fail. (famous last words?)
1425 * If it had been mremap()'d, the starting address would not
1426 * match the usual checks anyway. So assume all vma's are
1427 * above the starting address given.
1429 vma = find_vma(mm, addr);
1433 next = vma->vm_next;
1436 * Check if the starting address would match, i.e. it's
1437 * a fragment created by mprotect() and/or munmap(), or it
1438 * otherwise it starts at this address with no hassles.
1440 if ((vma->vm_ops == &shm_vm_ops) &&
1441 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1444 * Record the file of the shm segment being
1445 * unmapped. With mremap(), someone could place
1446 * page from another segment but with equal offsets
1447 * in the range we are unmapping.
1449 file = vma->vm_file;
1450 size = i_size_read(file_inode(vma->vm_file));
1451 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1453 * We discovered the size of the shm segment, so
1454 * break out of here and fall through to the next
1455 * loop that uses the size information to stop
1456 * searching for matching vma's.
1466 * We need look no further than the maximum address a fragment
1467 * could possibly have landed at. Also cast things to loff_t to
1468 * prevent overflows and make comparisons vs. equal-width types.
1470 size = PAGE_ALIGN(size);
1471 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1472 next = vma->vm_next;
1474 /* finding a matching vma now does not alter retval */
1475 if ((vma->vm_ops == &shm_vm_ops) &&
1476 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1477 (vma->vm_file == file))
1478 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1482 #else /* CONFIG_MMU */
1483 /* under NOMMU conditions, the exact address to be destroyed must be
1485 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1486 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1492 up_write(&mm->mmap_sem);
1496 #ifdef CONFIG_PROC_FS
1497 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1499 struct user_namespace *user_ns = seq_user_ns(s);
1500 struct shmid_kernel *shp = it;
1501 unsigned long rss = 0, swp = 0;
1503 shm_add_rss_swap(shp, &rss, &swp);
1505 #if BITS_PER_LONG <= 32
1506 #define SIZE_SPEC "%10lu"
1508 #define SIZE_SPEC "%21lu"
1512 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1513 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1514 SIZE_SPEC " " SIZE_SPEC "\n",
1522 from_kuid_munged(user_ns, shp->shm_perm.uid),
1523 from_kgid_munged(user_ns, shp->shm_perm.gid),
1524 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1525 from_kgid_munged(user_ns, shp->shm_perm.cgid),