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>
24 #include <linux/slab.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
43 #include <asm/uaccess.h>
47 struct shm_file_data {
49 struct ipc_namespace *ns;
51 const struct vm_operations_struct *vm_ops;
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56 static const struct file_operations shm_file_operations;
57 static const struct vm_operations_struct shm_vm_ops;
59 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
61 #define shm_unlock(shp) \
62 ipc_unlock(&(shp)->shm_perm)
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
72 void shm_init_ns(struct ipc_namespace *ns)
74 ns->shm_ctlmax = SHMMAX;
75 ns->shm_ctlall = SHMALL;
76 ns->shm_ctlmni = SHMMNI;
77 ns->shm_rmid_forced = 0;
79 ipc_init_ids(&shm_ids(ns));
83 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84 * Only shm_ids.rw_mutex remains locked on exit.
86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
88 struct shmid_kernel *shp;
89 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
92 shp->shm_perm.mode |= SHM_DEST;
93 /* Do not find it any more */
94 shp->shm_perm.key = IPC_PRIVATE;
101 void shm_exit_ns(struct ipc_namespace *ns)
103 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
104 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
108 static int __init ipc_ns_init(void)
110 shm_init_ns(&init_ipc_ns);
114 pure_initcall(ipc_ns_init);
116 void __init shm_init (void)
118 ipc_init_proc_interface("sysvipc/shm",
119 #if BITS_PER_LONG <= 32
120 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
122 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
124 IPC_SHM_IDS, sysvipc_shm_proc_show);
127 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
129 struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
132 return ERR_CAST(ipcp);
134 return container_of(ipcp, struct shmid_kernel, shm_perm);
137 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
139 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
142 return ERR_CAST(ipcp);
144 return container_of(ipcp, struct shmid_kernel, shm_perm);
148 * shm_lock_(check_) routines are called in the paths where the rw_mutex
149 * is not necessarily held.
151 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
153 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
156 return (struct shmid_kernel *)ipcp;
158 return container_of(ipcp, struct shmid_kernel, shm_perm);
161 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
164 ipc_lock_object(&ipcp->shm_perm);
167 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
170 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
173 return (struct shmid_kernel *)ipcp;
175 return container_of(ipcp, struct shmid_kernel, shm_perm);
178 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
180 ipc_rmid(&shm_ids(ns), &s->shm_perm);
184 /* This is called by fork, once for every shm attach. */
185 static void shm_open(struct vm_area_struct *vma)
187 struct file *file = vma->vm_file;
188 struct shm_file_data *sfd = shm_file_data(file);
189 struct shmid_kernel *shp;
191 shp = shm_lock(sfd->ns, sfd->id);
193 shp->shm_atim = get_seconds();
194 shp->shm_lprid = task_tgid_vnr(current);
200 * shm_destroy - free the struct shmid_kernel
203 * @shp: struct to free
205 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
206 * but returns with shp unlocked and freed.
208 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
210 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
213 if (!is_file_hugepages(shp->shm_file))
214 shmem_lock(shp->shm_file, 0, shp->mlock_user);
215 else if (shp->mlock_user)
216 user_shm_unlock(file_inode(shp->shm_file)->i_size,
218 fput (shp->shm_file);
219 security_shm_free(shp);
224 * shm_may_destroy - identifies whether shm segment should be destroyed now
226 * Returns true if and only if there are no active users of the segment and
227 * one of the following is true:
229 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
231 * 2) sysctl kernel.shm_rmid_forced is set to 1.
233 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
235 return (shp->shm_nattch == 0) &&
236 (ns->shm_rmid_forced ||
237 (shp->shm_perm.mode & SHM_DEST));
241 * remove the attach descriptor vma.
242 * free memory for segment if it is marked destroyed.
243 * The descriptor has already been removed from the current->mm->mmap list
244 * and will later be kfree()d.
246 static void shm_close(struct vm_area_struct *vma)
248 struct file * file = vma->vm_file;
249 struct shm_file_data *sfd = shm_file_data(file);
250 struct shmid_kernel *shp;
251 struct ipc_namespace *ns = sfd->ns;
253 down_write(&shm_ids(ns).rw_mutex);
254 /* remove from the list of attaches of the shm segment */
255 shp = shm_lock(ns, sfd->id);
257 shp->shm_lprid = task_tgid_vnr(current);
258 shp->shm_dtim = get_seconds();
260 if (shm_may_destroy(ns, shp))
261 shm_destroy(ns, shp);
264 up_write(&shm_ids(ns).rw_mutex);
267 /* Called with ns->shm_ids(ns).rw_mutex locked */
268 static int shm_try_destroy_current(int id, void *p, void *data)
270 struct ipc_namespace *ns = data;
271 struct kern_ipc_perm *ipcp = p;
272 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
274 if (shp->shm_creator != current)
278 * Mark it as orphaned to destroy the segment when
279 * kernel.shm_rmid_forced is changed.
280 * It is noop if the following shm_may_destroy() returns true.
282 shp->shm_creator = NULL;
285 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
286 * is not set, it shouldn't be deleted here.
288 if (!ns->shm_rmid_forced)
291 if (shm_may_destroy(ns, shp)) {
292 shm_lock_by_ptr(shp);
293 shm_destroy(ns, shp);
298 /* Called with ns->shm_ids(ns).rw_mutex locked */
299 static int shm_try_destroy_orphaned(int id, void *p, void *data)
301 struct ipc_namespace *ns = data;
302 struct kern_ipc_perm *ipcp = p;
303 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
306 * We want to destroy segments without users and with already
307 * exit'ed originating process.
309 * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
311 if (shp->shm_creator != NULL)
314 if (shm_may_destroy(ns, shp)) {
315 shm_lock_by_ptr(shp);
316 shm_destroy(ns, shp);
321 void shm_destroy_orphaned(struct ipc_namespace *ns)
323 down_write(&shm_ids(ns).rw_mutex);
324 if (shm_ids(ns).in_use)
325 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
326 up_write(&shm_ids(ns).rw_mutex);
330 void exit_shm(struct task_struct *task)
332 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
334 if (shm_ids(ns).in_use == 0)
337 /* Destroy all already created segments, but not mapped yet */
338 down_write(&shm_ids(ns).rw_mutex);
339 if (shm_ids(ns).in_use)
340 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
341 up_write(&shm_ids(ns).rw_mutex);
344 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
346 struct file *file = vma->vm_file;
347 struct shm_file_data *sfd = shm_file_data(file);
349 return sfd->vm_ops->fault(vma, vmf);
353 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
355 struct file *file = vma->vm_file;
356 struct shm_file_data *sfd = shm_file_data(file);
358 if (sfd->vm_ops->set_policy)
359 err = sfd->vm_ops->set_policy(vma, new);
363 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
366 struct file *file = vma->vm_file;
367 struct shm_file_data *sfd = shm_file_data(file);
368 struct mempolicy *pol = NULL;
370 if (sfd->vm_ops->get_policy)
371 pol = sfd->vm_ops->get_policy(vma, addr);
372 else if (vma->vm_policy)
373 pol = vma->vm_policy;
379 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
381 struct shm_file_data *sfd = shm_file_data(file);
384 ret = sfd->file->f_op->mmap(sfd->file, vma);
387 sfd->vm_ops = vma->vm_ops;
389 BUG_ON(!sfd->vm_ops->fault);
391 vma->vm_ops = &shm_vm_ops;
397 static int shm_release(struct inode *ino, struct file *file)
399 struct shm_file_data *sfd = shm_file_data(file);
402 shm_file_data(file) = NULL;
407 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
409 struct shm_file_data *sfd = shm_file_data(file);
411 if (!sfd->file->f_op->fsync)
413 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
416 static long shm_fallocate(struct file *file, int mode, loff_t offset,
419 struct shm_file_data *sfd = shm_file_data(file);
421 if (!sfd->file->f_op->fallocate)
423 return sfd->file->f_op->fallocate(file, mode, offset, len);
426 static unsigned long shm_get_unmapped_area(struct file *file,
427 unsigned long addr, unsigned long len, unsigned long pgoff,
430 struct shm_file_data *sfd = shm_file_data(file);
431 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
435 static const struct file_operations shm_file_operations = {
438 .release = shm_release,
440 .get_unmapped_area = shm_get_unmapped_area,
442 .llseek = noop_llseek,
443 .fallocate = shm_fallocate,
446 static const struct file_operations shm_file_operations_huge = {
449 .release = shm_release,
450 .get_unmapped_area = shm_get_unmapped_area,
451 .llseek = noop_llseek,
452 .fallocate = shm_fallocate,
455 int is_file_shm_hugepages(struct file *file)
457 return file->f_op == &shm_file_operations_huge;
460 static const struct vm_operations_struct shm_vm_ops = {
461 .open = shm_open, /* callback for a new vm-area open */
462 .close = shm_close, /* callback for when the vm-area is released */
464 #if defined(CONFIG_NUMA)
465 .set_policy = shm_set_policy,
466 .get_policy = shm_get_policy,
471 * newseg - Create a new shared memory segment
473 * @params: ptr to the structure that contains key, size and shmflg
475 * Called with shm_ids.rw_mutex held as a writer.
478 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
480 key_t key = params->key;
481 int shmflg = params->flg;
482 size_t size = params->u.size;
484 struct shmid_kernel *shp;
485 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
489 vm_flags_t acctflag = 0;
491 if (size < SHMMIN || size > ns->shm_ctlmax)
494 if (ns->shm_tot + numpages > ns->shm_ctlall)
497 shp = ipc_rcu_alloc(sizeof(*shp));
501 shp->shm_perm.key = key;
502 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
503 shp->mlock_user = NULL;
505 shp->shm_perm.security = NULL;
506 error = security_shm_alloc(shp);
512 sprintf (name, "SYSV%08x", key);
513 if (shmflg & SHM_HUGETLB) {
517 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
522 hugesize = ALIGN(size, huge_page_size(hs));
524 /* hugetlb_file_setup applies strict accounting */
525 if (shmflg & SHM_NORESERVE)
526 acctflag = VM_NORESERVE;
527 file = hugetlb_file_setup(name, hugesize, acctflag,
528 &shp->mlock_user, HUGETLB_SHMFS_INODE,
529 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
532 * Do not allow no accounting for OVERCOMMIT_NEVER, even
535 if ((shmflg & SHM_NORESERVE) &&
536 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
537 acctflag = VM_NORESERVE;
538 file = shmem_file_setup(name, size, acctflag);
540 error = PTR_ERR(file);
544 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
550 shp->shm_cprid = task_tgid_vnr(current);
552 shp->shm_atim = shp->shm_dtim = 0;
553 shp->shm_ctim = get_seconds();
554 shp->shm_segsz = size;
556 shp->shm_file = file;
557 shp->shm_creator = current;
560 * shmid gets reported as "inode#" in /proc/pid/maps.
561 * proc-ps tools use this. Changing this will break them.
563 file_inode(file)->i_ino = shp->shm_perm.id;
565 ns->shm_tot += numpages;
566 error = shp->shm_perm.id;
568 ipc_unlock_object(&shp->shm_perm);
573 if (is_file_hugepages(file) && shp->mlock_user)
574 user_shm_unlock(size, shp->mlock_user);
577 security_shm_free(shp);
583 * Called with shm_ids.rw_mutex and ipcp locked.
585 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
587 struct shmid_kernel *shp;
589 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
590 return security_shm_associate(shp, shmflg);
594 * Called with shm_ids.rw_mutex and ipcp locked.
596 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
597 struct ipc_params *params)
599 struct shmid_kernel *shp;
601 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
602 if (shp->shm_segsz < params->u.size)
608 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
610 struct ipc_namespace *ns;
611 struct ipc_ops shm_ops;
612 struct ipc_params shm_params;
614 ns = current->nsproxy->ipc_ns;
616 shm_ops.getnew = newseg;
617 shm_ops.associate = shm_security;
618 shm_ops.more_checks = shm_more_checks;
620 shm_params.key = key;
621 shm_params.flg = shmflg;
622 shm_params.u.size = size;
624 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
627 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
631 return copy_to_user(buf, in, sizeof(*in));
636 memset(&out, 0, sizeof(out));
637 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
638 out.shm_segsz = in->shm_segsz;
639 out.shm_atime = in->shm_atime;
640 out.shm_dtime = in->shm_dtime;
641 out.shm_ctime = in->shm_ctime;
642 out.shm_cpid = in->shm_cpid;
643 out.shm_lpid = in->shm_lpid;
644 out.shm_nattch = in->shm_nattch;
646 return copy_to_user(buf, &out, sizeof(out));
653 static inline unsigned long
654 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
658 if (copy_from_user(out, buf, sizeof(*out)))
663 struct shmid_ds tbuf_old;
665 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
668 out->shm_perm.uid = tbuf_old.shm_perm.uid;
669 out->shm_perm.gid = tbuf_old.shm_perm.gid;
670 out->shm_perm.mode = tbuf_old.shm_perm.mode;
679 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
683 return copy_to_user(buf, in, sizeof(*in));
688 if(in->shmmax > INT_MAX)
689 out.shmmax = INT_MAX;
691 out.shmmax = (int)in->shmmax;
693 out.shmmin = in->shmmin;
694 out.shmmni = in->shmmni;
695 out.shmseg = in->shmseg;
696 out.shmall = in->shmall;
698 return copy_to_user(buf, &out, sizeof(out));
706 * Calculate and add used RSS and swap pages of a shm.
707 * Called with shm_ids.rw_mutex held as a reader
709 static void shm_add_rss_swap(struct shmid_kernel *shp,
710 unsigned long *rss_add, unsigned long *swp_add)
714 inode = file_inode(shp->shm_file);
716 if (is_file_hugepages(shp->shm_file)) {
717 struct address_space *mapping = inode->i_mapping;
718 struct hstate *h = hstate_file(shp->shm_file);
719 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
722 struct shmem_inode_info *info = SHMEM_I(inode);
723 spin_lock(&info->lock);
724 *rss_add += inode->i_mapping->nrpages;
725 *swp_add += info->swapped;
726 spin_unlock(&info->lock);
728 *rss_add += inode->i_mapping->nrpages;
734 * Called with shm_ids.rw_mutex held as a reader
736 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
745 in_use = shm_ids(ns).in_use;
747 for (total = 0, next_id = 0; total < in_use; next_id++) {
748 struct kern_ipc_perm *ipc;
749 struct shmid_kernel *shp;
751 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
754 shp = container_of(ipc, struct shmid_kernel, shm_perm);
756 shm_add_rss_swap(shp, rss, swp);
763 * This function handles some shmctl commands which require the rw_mutex
764 * to be held in write mode.
765 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
767 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
768 struct shmid_ds __user *buf, int version)
770 struct kern_ipc_perm *ipcp;
771 struct shmid64_ds shmid64;
772 struct shmid_kernel *shp;
775 if (cmd == IPC_SET) {
776 if (copy_shmid_from_user(&shmid64, buf, version))
780 down_write(&shm_ids(ns).rw_mutex);
783 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
784 &shmid64.shm_perm, 0);
790 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
792 err = security_shm_shmctl(shp, cmd);
798 ipc_lock_object(&shp->shm_perm);
799 /* do_shm_rmid unlocks the ipc object and rcu */
800 do_shm_rmid(ns, ipcp);
803 ipc_lock_object(&shp->shm_perm);
804 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
807 shp->shm_ctim = get_seconds();
815 ipc_unlock_object(&shp->shm_perm);
819 up_write(&shm_ids(ns).rw_mutex);
823 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
824 int cmd, int version, void __user *buf)
827 struct shmid_kernel *shp;
829 /* preliminary security checks for *_INFO */
830 if (cmd == IPC_INFO || cmd == SHM_INFO) {
831 err = security_shm_shmctl(NULL, cmd);
839 struct shminfo64 shminfo;
841 memset(&shminfo, 0, sizeof(shminfo));
842 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
843 shminfo.shmmax = ns->shm_ctlmax;
844 shminfo.shmall = ns->shm_ctlall;
846 shminfo.shmmin = SHMMIN;
847 if(copy_shminfo_to_user (buf, &shminfo, version))
850 down_read(&shm_ids(ns).rw_mutex);
851 err = ipc_get_maxid(&shm_ids(ns));
852 up_read(&shm_ids(ns).rw_mutex);
860 struct shm_info shm_info;
862 memset(&shm_info, 0, sizeof(shm_info));
863 down_read(&shm_ids(ns).rw_mutex);
864 shm_info.used_ids = shm_ids(ns).in_use;
865 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
866 shm_info.shm_tot = ns->shm_tot;
867 shm_info.swap_attempts = 0;
868 shm_info.swap_successes = 0;
869 err = ipc_get_maxid(&shm_ids(ns));
870 up_read(&shm_ids(ns).rw_mutex);
871 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
876 err = err < 0 ? 0 : err;
882 struct shmid64_ds tbuf;
886 if (cmd == SHM_STAT) {
887 shp = shm_obtain_object(ns, shmid);
892 result = shp->shm_perm.id;
894 shp = shm_obtain_object_check(ns, shmid);
903 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
906 err = security_shm_shmctl(shp, cmd);
910 memset(&tbuf, 0, sizeof(tbuf));
911 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
912 tbuf.shm_segsz = shp->shm_segsz;
913 tbuf.shm_atime = shp->shm_atim;
914 tbuf.shm_dtime = shp->shm_dtim;
915 tbuf.shm_ctime = shp->shm_ctim;
916 tbuf.shm_cpid = shp->shm_cprid;
917 tbuf.shm_lpid = shp->shm_lprid;
918 tbuf.shm_nattch = shp->shm_nattch;
921 if (copy_shmid_to_user(buf, &tbuf, version))
937 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
939 struct shmid_kernel *shp;
941 struct ipc_namespace *ns;
943 if (cmd < 0 || shmid < 0)
946 version = ipc_parse_version(&cmd);
947 ns = current->nsproxy->ipc_ns;
954 return shmctl_nolock(ns, shmid, cmd, version, buf);
957 return shmctl_down(ns, shmid, cmd, buf, version);
961 struct file *shm_file;
964 shp = shm_obtain_object_check(ns, shmid);
970 audit_ipc_obj(&(shp->shm_perm));
971 err = security_shm_shmctl(shp, cmd);
975 ipc_lock_object(&shp->shm_perm);
976 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
977 kuid_t euid = current_euid();
979 if (!uid_eq(euid, shp->shm_perm.uid) &&
980 !uid_eq(euid, shp->shm_perm.cuid))
982 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
986 shm_file = shp->shm_file;
987 if (is_file_hugepages(shm_file))
990 if (cmd == SHM_LOCK) {
991 struct user_struct *user = current_user();
992 err = shmem_lock(shm_file, 1, user);
993 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
994 shp->shm_perm.mode |= SHM_LOCKED;
995 shp->mlock_user = user;
1001 if (!(shp->shm_perm.mode & SHM_LOCKED))
1003 shmem_lock(shm_file, 0, shp->mlock_user);
1004 shp->shm_perm.mode &= ~SHM_LOCKED;
1005 shp->mlock_user = NULL;
1007 ipc_unlock_object(&shp->shm_perm);
1009 shmem_unlock_mapping(shm_file->f_mapping);
1019 ipc_unlock_object(&shp->shm_perm);
1026 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1028 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1029 * "raddr" thing points to kernel space, and there has to be a wrapper around
1032 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1033 unsigned long shmlba)
1035 struct shmid_kernel *shp;
1040 unsigned long flags;
1043 struct ipc_namespace *ns;
1044 struct shm_file_data *sfd;
1047 unsigned long populate = 0;
1052 else if ((addr = (ulong)shmaddr)) {
1053 if (addr & (shmlba - 1)) {
1054 if (shmflg & SHM_RND)
1055 addr &= ~(shmlba - 1); /* round down */
1057 #ifndef __ARCH_FORCE_SHMLBA
1058 if (addr & ~PAGE_MASK)
1062 flags = MAP_SHARED | MAP_FIXED;
1064 if ((shmflg & SHM_REMAP))
1070 if (shmflg & SHM_RDONLY) {
1073 f_mode = FMODE_READ;
1075 prot = PROT_READ | PROT_WRITE;
1076 acc_mode = S_IRUGO | S_IWUGO;
1077 f_mode = FMODE_READ | FMODE_WRITE;
1079 if (shmflg & SHM_EXEC) {
1081 acc_mode |= S_IXUGO;
1085 * We cannot rely on the fs check since SYSV IPC does have an
1086 * additional creator id...
1088 ns = current->nsproxy->ipc_ns;
1089 shp = shm_lock_check(ns, shmid);
1096 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1099 err = security_shm_shmat(shp, shmaddr, shmflg);
1103 path = shp->shm_file->f_path;
1106 size = i_size_read(path.dentry->d_inode);
1110 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1116 file = alloc_file(&path, f_mode,
1117 is_file_hugepages(shp->shm_file) ?
1118 &shm_file_operations_huge :
1119 &shm_file_operations);
1120 err = PTR_ERR(file);
1127 file->private_data = sfd;
1128 file->f_mapping = shp->shm_file->f_mapping;
1129 sfd->id = shp->shm_perm.id;
1130 sfd->ns = get_ipc_ns(ns);
1131 sfd->file = shp->shm_file;
1134 err = security_mmap_file(file, prot, flags);
1138 down_write(¤t->mm->mmap_sem);
1139 if (addr && !(shmflg & SHM_REMAP)) {
1141 if (find_vma_intersection(current->mm, addr, addr + size))
1144 * If shm segment goes below stack, make sure there is some
1145 * space left for the stack to grow (at least 4 pages).
1147 if (addr < current->mm->start_stack &&
1148 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1152 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1155 if (IS_ERR_VALUE(addr))
1158 up_write(¤t->mm->mmap_sem);
1160 mm_populate(addr, populate);
1166 down_write(&shm_ids(ns).rw_mutex);
1167 shp = shm_lock(ns, shmid);
1168 BUG_ON(IS_ERR(shp));
1170 if (shm_may_destroy(ns, shp))
1171 shm_destroy(ns, shp);
1174 up_write(&shm_ids(ns).rw_mutex);
1183 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1188 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1191 force_successful_syscall_return();
1196 * detach and kill segment if marked destroyed.
1197 * The work is done in shm_close.
1199 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1201 struct mm_struct *mm = current->mm;
1202 struct vm_area_struct *vma;
1203 unsigned long addr = (unsigned long)shmaddr;
1204 int retval = -EINVAL;
1207 struct vm_area_struct *next;
1210 if (addr & ~PAGE_MASK)
1213 down_write(&mm->mmap_sem);
1216 * This function tries to be smart and unmap shm segments that
1217 * were modified by partial mlock or munmap calls:
1218 * - It first determines the size of the shm segment that should be
1219 * unmapped: It searches for a vma that is backed by shm and that
1220 * started at address shmaddr. It records it's size and then unmaps
1222 * - Then it unmaps all shm vmas that started at shmaddr and that
1223 * are within the initially determined size.
1224 * Errors from do_munmap are ignored: the function only fails if
1225 * it's called with invalid parameters or if it's called to unmap
1226 * a part of a vma. Both calls in this function are for full vmas,
1227 * the parameters are directly copied from the vma itself and always
1228 * valid - therefore do_munmap cannot fail. (famous last words?)
1231 * If it had been mremap()'d, the starting address would not
1232 * match the usual checks anyway. So assume all vma's are
1233 * above the starting address given.
1235 vma = find_vma(mm, addr);
1239 next = vma->vm_next;
1242 * Check if the starting address would match, i.e. it's
1243 * a fragment created by mprotect() and/or munmap(), or it
1244 * otherwise it starts at this address with no hassles.
1246 if ((vma->vm_ops == &shm_vm_ops) &&
1247 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1250 size = file_inode(vma->vm_file)->i_size;
1251 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1253 * We discovered the size of the shm segment, so
1254 * break out of here and fall through to the next
1255 * loop that uses the size information to stop
1256 * searching for matching vma's.
1266 * We need look no further than the maximum address a fragment
1267 * could possibly have landed at. Also cast things to loff_t to
1268 * prevent overflows and make comparisons vs. equal-width types.
1270 size = PAGE_ALIGN(size);
1271 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1272 next = vma->vm_next;
1274 /* finding a matching vma now does not alter retval */
1275 if ((vma->vm_ops == &shm_vm_ops) &&
1276 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1278 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1282 #else /* CONFIG_MMU */
1283 /* under NOMMU conditions, the exact address to be destroyed must be
1286 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1287 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1293 up_write(&mm->mmap_sem);
1297 #ifdef CONFIG_PROC_FS
1298 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1300 struct user_namespace *user_ns = seq_user_ns(s);
1301 struct shmid_kernel *shp = it;
1302 unsigned long rss = 0, swp = 0;
1304 shm_add_rss_swap(shp, &rss, &swp);
1306 #if BITS_PER_LONG <= 32
1307 #define SIZE_SPEC "%10lu"
1309 #define SIZE_SPEC "%21lu"
1312 return seq_printf(s,
1313 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1314 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1315 SIZE_SPEC " " SIZE_SPEC "\n",
1323 from_kuid_munged(user_ns, shp->shm_perm.uid),
1324 from_kgid_munged(user_ns, shp->shm_perm.gid),
1325 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1326 from_kgid_munged(user_ns, shp->shm_perm.cgid),