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 void __init shm_init (void)
110 shm_init_ns(&init_ipc_ns);
111 ipc_init_proc_interface("sysvipc/shm",
112 #if BITS_PER_LONG <= 32
113 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
115 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
117 IPC_SHM_IDS, sysvipc_shm_proc_show);
121 * shm_lock_(check_) routines are called in the paths where the rw_mutex
122 * is not necessarily held.
124 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
126 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
129 return (struct shmid_kernel *)ipcp;
131 return container_of(ipcp, struct shmid_kernel, shm_perm);
134 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
137 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
140 return (struct shmid_kernel *)ipcp;
142 return container_of(ipcp, struct shmid_kernel, shm_perm);
145 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
147 ipc_rmid(&shm_ids(ns), &s->shm_perm);
151 /* This is called by fork, once for every shm attach. */
152 static void shm_open(struct vm_area_struct *vma)
154 struct file *file = vma->vm_file;
155 struct shm_file_data *sfd = shm_file_data(file);
156 struct shmid_kernel *shp;
158 shp = shm_lock(sfd->ns, sfd->id);
160 shp->shm_atim = get_seconds();
161 shp->shm_lprid = task_tgid_vnr(current);
167 * shm_destroy - free the struct shmid_kernel
170 * @shp: struct to free
172 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
173 * but returns with shp unlocked and freed.
175 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
177 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
180 if (!is_file_hugepages(shp->shm_file))
181 shmem_lock(shp->shm_file, 0, shp->mlock_user);
182 else if (shp->mlock_user)
183 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
185 fput (shp->shm_file);
186 security_shm_free(shp);
191 * shm_may_destroy - identifies whether shm segment should be destroyed now
193 * Returns true if and only if there are no active users of the segment and
194 * one of the following is true:
196 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
198 * 2) sysctl kernel.shm_rmid_forced is set to 1.
200 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
202 return (shp->shm_nattch == 0) &&
203 (ns->shm_rmid_forced ||
204 (shp->shm_perm.mode & SHM_DEST));
208 * remove the attach descriptor vma.
209 * free memory for segment if it is marked destroyed.
210 * The descriptor has already been removed from the current->mm->mmap list
211 * and will later be kfree()d.
213 static void shm_close(struct vm_area_struct *vma)
215 struct file * file = vma->vm_file;
216 struct shm_file_data *sfd = shm_file_data(file);
217 struct shmid_kernel *shp;
218 struct ipc_namespace *ns = sfd->ns;
220 down_write(&shm_ids(ns).rw_mutex);
221 /* remove from the list of attaches of the shm segment */
222 shp = shm_lock(ns, sfd->id);
224 shp->shm_lprid = task_tgid_vnr(current);
225 shp->shm_dtim = get_seconds();
227 if (shm_may_destroy(ns, shp))
228 shm_destroy(ns, shp);
231 up_write(&shm_ids(ns).rw_mutex);
234 static int shm_try_destroy_current(int id, void *p, void *data)
236 struct ipc_namespace *ns = data;
237 struct shmid_kernel *shp = shm_lock(ns, id);
242 if (shp->shm_creator != current) {
248 * Mark it as orphaned to destroy the segment when
249 * kernel.shm_rmid_forced is changed.
250 * It is noop if the following shm_may_destroy() returns true.
252 shp->shm_creator = NULL;
255 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
256 * is not set, it shouldn't be deleted here.
258 if (!ns->shm_rmid_forced) {
263 if (shm_may_destroy(ns, shp))
264 shm_destroy(ns, shp);
270 static int shm_try_destroy_orphaned(int id, void *p, void *data)
272 struct ipc_namespace *ns = data;
273 struct shmid_kernel *shp = shm_lock(ns, id);
279 * We want to destroy segments without users and with already
280 * exit'ed originating process.
282 if (shp->shm_creator != NULL) {
287 if (shm_may_destroy(ns, shp))
288 shm_destroy(ns, shp);
294 void shm_destroy_orphaned(struct ipc_namespace *ns)
296 down_write(&shm_ids(ns).rw_mutex);
297 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
298 up_write(&shm_ids(ns).rw_mutex);
302 void exit_shm(struct task_struct *task)
304 struct nsproxy *nsp = task->nsproxy;
305 struct ipc_namespace *ns;
313 /* Destroy all already created segments, but not mapped yet */
314 down_write(&shm_ids(ns).rw_mutex);
315 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
316 up_write(&shm_ids(ns).rw_mutex);
319 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
321 struct file *file = vma->vm_file;
322 struct shm_file_data *sfd = shm_file_data(file);
324 return sfd->vm_ops->fault(vma, vmf);
328 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
330 struct file *file = vma->vm_file;
331 struct shm_file_data *sfd = shm_file_data(file);
333 if (sfd->vm_ops->set_policy)
334 err = sfd->vm_ops->set_policy(vma, new);
338 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
341 struct file *file = vma->vm_file;
342 struct shm_file_data *sfd = shm_file_data(file);
343 struct mempolicy *pol = NULL;
345 if (sfd->vm_ops->get_policy)
346 pol = sfd->vm_ops->get_policy(vma, addr);
347 else if (vma->vm_policy)
348 pol = vma->vm_policy;
354 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
356 struct shm_file_data *sfd = shm_file_data(file);
359 ret = sfd->file->f_op->mmap(sfd->file, vma);
362 sfd->vm_ops = vma->vm_ops;
364 BUG_ON(!sfd->vm_ops->fault);
366 vma->vm_ops = &shm_vm_ops;
372 static int shm_release(struct inode *ino, struct file *file)
374 struct shm_file_data *sfd = shm_file_data(file);
377 shm_file_data(file) = NULL;
382 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
384 struct shm_file_data *sfd = shm_file_data(file);
386 if (!sfd->file->f_op->fsync)
388 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
391 static unsigned long shm_get_unmapped_area(struct file *file,
392 unsigned long addr, unsigned long len, unsigned long pgoff,
395 struct shm_file_data *sfd = shm_file_data(file);
396 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
400 static const struct file_operations shm_file_operations = {
403 .release = shm_release,
405 .get_unmapped_area = shm_get_unmapped_area,
407 .llseek = noop_llseek,
410 static const struct file_operations shm_file_operations_huge = {
413 .release = shm_release,
414 .get_unmapped_area = shm_get_unmapped_area,
415 .llseek = noop_llseek,
418 int is_file_shm_hugepages(struct file *file)
420 return file->f_op == &shm_file_operations_huge;
423 static const struct vm_operations_struct shm_vm_ops = {
424 .open = shm_open, /* callback for a new vm-area open */
425 .close = shm_close, /* callback for when the vm-area is released */
427 #if defined(CONFIG_NUMA)
428 .set_policy = shm_set_policy,
429 .get_policy = shm_get_policy,
434 * newseg - Create a new shared memory segment
436 * @params: ptr to the structure that contains key, size and shmflg
438 * Called with shm_ids.rw_mutex held as a writer.
441 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
443 key_t key = params->key;
444 int shmflg = params->flg;
445 size_t size = params->u.size;
447 struct shmid_kernel *shp;
448 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
452 vm_flags_t acctflag = 0;
454 if (size < SHMMIN || size > ns->shm_ctlmax)
457 if (ns->shm_tot + numpages > ns->shm_ctlall)
460 shp = ipc_rcu_alloc(sizeof(*shp));
464 shp->shm_perm.key = key;
465 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
466 shp->mlock_user = NULL;
468 shp->shm_perm.security = NULL;
469 error = security_shm_alloc(shp);
475 sprintf (name, "SYSV%08x", key);
476 if (shmflg & SHM_HUGETLB) {
477 /* hugetlb_file_setup applies strict accounting */
478 if (shmflg & SHM_NORESERVE)
479 acctflag = VM_NORESERVE;
480 file = hugetlb_file_setup(name, size, acctflag,
481 &shp->mlock_user, HUGETLB_SHMFS_INODE);
484 * Do not allow no accounting for OVERCOMMIT_NEVER, even
487 if ((shmflg & SHM_NORESERVE) &&
488 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
489 acctflag = VM_NORESERVE;
490 file = shmem_file_setup(name, size, acctflag);
492 error = PTR_ERR(file);
496 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
502 shp->shm_cprid = task_tgid_vnr(current);
504 shp->shm_atim = shp->shm_dtim = 0;
505 shp->shm_ctim = get_seconds();
506 shp->shm_segsz = size;
508 shp->shm_file = file;
509 shp->shm_creator = current;
511 * shmid gets reported as "inode#" in /proc/pid/maps.
512 * proc-ps tools use this. Changing this will break them.
514 file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
516 ns->shm_tot += numpages;
517 error = shp->shm_perm.id;
522 if (is_file_hugepages(file) && shp->mlock_user)
523 user_shm_unlock(size, shp->mlock_user);
526 security_shm_free(shp);
532 * Called with shm_ids.rw_mutex and ipcp locked.
534 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
536 struct shmid_kernel *shp;
538 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
539 return security_shm_associate(shp, shmflg);
543 * Called with shm_ids.rw_mutex and ipcp locked.
545 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
546 struct ipc_params *params)
548 struct shmid_kernel *shp;
550 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
551 if (shp->shm_segsz < params->u.size)
557 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
559 struct ipc_namespace *ns;
560 struct ipc_ops shm_ops;
561 struct ipc_params shm_params;
563 ns = current->nsproxy->ipc_ns;
565 shm_ops.getnew = newseg;
566 shm_ops.associate = shm_security;
567 shm_ops.more_checks = shm_more_checks;
569 shm_params.key = key;
570 shm_params.flg = shmflg;
571 shm_params.u.size = size;
573 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
576 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
580 return copy_to_user(buf, in, sizeof(*in));
585 memset(&out, 0, sizeof(out));
586 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
587 out.shm_segsz = in->shm_segsz;
588 out.shm_atime = in->shm_atime;
589 out.shm_dtime = in->shm_dtime;
590 out.shm_ctime = in->shm_ctime;
591 out.shm_cpid = in->shm_cpid;
592 out.shm_lpid = in->shm_lpid;
593 out.shm_nattch = in->shm_nattch;
595 return copy_to_user(buf, &out, sizeof(out));
602 static inline unsigned long
603 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
607 if (copy_from_user(out, buf, sizeof(*out)))
612 struct shmid_ds tbuf_old;
614 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
617 out->shm_perm.uid = tbuf_old.shm_perm.uid;
618 out->shm_perm.gid = tbuf_old.shm_perm.gid;
619 out->shm_perm.mode = tbuf_old.shm_perm.mode;
628 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
632 return copy_to_user(buf, in, sizeof(*in));
637 if(in->shmmax > INT_MAX)
638 out.shmmax = INT_MAX;
640 out.shmmax = (int)in->shmmax;
642 out.shmmin = in->shmmin;
643 out.shmmni = in->shmmni;
644 out.shmseg = in->shmseg;
645 out.shmall = in->shmall;
647 return copy_to_user(buf, &out, sizeof(out));
655 * Calculate and add used RSS and swap pages of a shm.
656 * Called with shm_ids.rw_mutex held as a reader
658 static void shm_add_rss_swap(struct shmid_kernel *shp,
659 unsigned long *rss_add, unsigned long *swp_add)
663 inode = shp->shm_file->f_path.dentry->d_inode;
665 if (is_file_hugepages(shp->shm_file)) {
666 struct address_space *mapping = inode->i_mapping;
667 struct hstate *h = hstate_file(shp->shm_file);
668 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
671 struct shmem_inode_info *info = SHMEM_I(inode);
672 spin_lock(&info->lock);
673 *rss_add += inode->i_mapping->nrpages;
674 *swp_add += info->swapped;
675 spin_unlock(&info->lock);
677 *rss_add += inode->i_mapping->nrpages;
683 * Called with shm_ids.rw_mutex held as a reader
685 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
694 in_use = shm_ids(ns).in_use;
696 for (total = 0, next_id = 0; total < in_use; next_id++) {
697 struct kern_ipc_perm *ipc;
698 struct shmid_kernel *shp;
700 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
703 shp = container_of(ipc, struct shmid_kernel, shm_perm);
705 shm_add_rss_swap(shp, rss, swp);
712 * This function handles some shmctl commands which require the rw_mutex
713 * to be held in write mode.
714 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
716 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
717 struct shmid_ds __user *buf, int version)
719 struct kern_ipc_perm *ipcp;
720 struct shmid64_ds shmid64;
721 struct shmid_kernel *shp;
724 if (cmd == IPC_SET) {
725 if (copy_shmid_from_user(&shmid64, buf, version))
729 ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
730 &shmid64.shm_perm, 0);
732 return PTR_ERR(ipcp);
734 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
736 err = security_shm_shmctl(shp, cmd);
741 do_shm_rmid(ns, ipcp);
744 ipc_update_perm(&shmid64.shm_perm, ipcp);
745 shp->shm_ctim = get_seconds();
753 up_write(&shm_ids(ns).rw_mutex);
757 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
759 struct shmid_kernel *shp;
761 struct ipc_namespace *ns;
763 if (cmd < 0 || shmid < 0) {
768 version = ipc_parse_version(&cmd);
769 ns = current->nsproxy->ipc_ns;
771 switch (cmd) { /* replace with proc interface ? */
774 struct shminfo64 shminfo;
776 err = security_shm_shmctl(NULL, cmd);
780 memset(&shminfo, 0, sizeof(shminfo));
781 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
782 shminfo.shmmax = ns->shm_ctlmax;
783 shminfo.shmall = ns->shm_ctlall;
785 shminfo.shmmin = SHMMIN;
786 if(copy_shminfo_to_user (buf, &shminfo, version))
789 down_read(&shm_ids(ns).rw_mutex);
790 err = ipc_get_maxid(&shm_ids(ns));
791 up_read(&shm_ids(ns).rw_mutex);
799 struct shm_info shm_info;
801 err = security_shm_shmctl(NULL, cmd);
805 memset(&shm_info, 0, sizeof(shm_info));
806 down_read(&shm_ids(ns).rw_mutex);
807 shm_info.used_ids = shm_ids(ns).in_use;
808 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
809 shm_info.shm_tot = ns->shm_tot;
810 shm_info.swap_attempts = 0;
811 shm_info.swap_successes = 0;
812 err = ipc_get_maxid(&shm_ids(ns));
813 up_read(&shm_ids(ns).rw_mutex);
814 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
819 err = err < 0 ? 0 : err;
825 struct shmid64_ds tbuf;
828 if (cmd == SHM_STAT) {
829 shp = shm_lock(ns, shmid);
834 result = shp->shm_perm.id;
836 shp = shm_lock_check(ns, shmid);
844 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
846 err = security_shm_shmctl(shp, cmd);
849 memset(&tbuf, 0, sizeof(tbuf));
850 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
851 tbuf.shm_segsz = shp->shm_segsz;
852 tbuf.shm_atime = shp->shm_atim;
853 tbuf.shm_dtime = shp->shm_dtim;
854 tbuf.shm_ctime = shp->shm_ctim;
855 tbuf.shm_cpid = shp->shm_cprid;
856 tbuf.shm_lpid = shp->shm_lprid;
857 tbuf.shm_nattch = shp->shm_nattch;
859 if(copy_shmid_to_user (buf, &tbuf, version))
868 struct file *uninitialized_var(shm_file);
870 lru_add_drain_all(); /* drain pagevecs to lru lists */
872 shp = shm_lock_check(ns, shmid);
878 audit_ipc_obj(&(shp->shm_perm));
880 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
881 uid_t euid = current_euid();
883 if (euid != shp->shm_perm.uid &&
884 euid != shp->shm_perm.cuid)
886 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
890 err = security_shm_shmctl(shp, cmd);
895 struct user_struct *user = current_user();
896 if (!is_file_hugepages(shp->shm_file)) {
897 err = shmem_lock(shp->shm_file, 1, user);
898 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
899 shp->shm_perm.mode |= SHM_LOCKED;
900 shp->mlock_user = user;
903 } else if (!is_file_hugepages(shp->shm_file)) {
904 shmem_lock(shp->shm_file, 0, shp->mlock_user);
905 shp->shm_perm.mode &= ~SHM_LOCKED;
906 shp->mlock_user = NULL;
913 err = shmctl_down(ns, shmid, cmd, buf, version);
926 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
928 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
929 * "raddr" thing points to kernel space, and there has to be a wrapper around
932 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
934 struct shmid_kernel *shp;
942 unsigned long user_addr;
943 struct ipc_namespace *ns;
944 struct shm_file_data *sfd;
951 else if ((addr = (ulong)shmaddr)) {
952 if (addr & (SHMLBA-1)) {
953 if (shmflg & SHM_RND)
954 addr &= ~(SHMLBA-1); /* round down */
956 #ifndef __ARCH_FORCE_SHMLBA
957 if (addr & ~PAGE_MASK)
961 flags = MAP_SHARED | MAP_FIXED;
963 if ((shmflg & SHM_REMAP))
969 if (shmflg & SHM_RDONLY) {
974 prot = PROT_READ | PROT_WRITE;
975 acc_mode = S_IRUGO | S_IWUGO;
976 f_mode = FMODE_READ | FMODE_WRITE;
978 if (shmflg & SHM_EXEC) {
984 * We cannot rely on the fs check since SYSV IPC does have an
985 * additional creator id...
987 ns = current->nsproxy->ipc_ns;
988 shp = shm_lock_check(ns, shmid);
995 if (ipcperms(ns, &shp->shm_perm, acc_mode))
998 err = security_shm_shmat(shp, shmaddr, shmflg);
1002 path = shp->shm_file->f_path;
1005 size = i_size_read(path.dentry->d_inode);
1009 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1011 goto out_put_dentry;
1013 file = alloc_file(&path, f_mode,
1014 is_file_hugepages(shp->shm_file) ?
1015 &shm_file_operations_huge :
1016 &shm_file_operations);
1020 file->private_data = sfd;
1021 file->f_mapping = shp->shm_file->f_mapping;
1022 sfd->id = shp->shm_perm.id;
1023 sfd->ns = get_ipc_ns(ns);
1024 sfd->file = shp->shm_file;
1027 down_write(¤t->mm->mmap_sem);
1028 if (addr && !(shmflg & SHM_REMAP)) {
1030 if (find_vma_intersection(current->mm, addr, addr + size))
1033 * If shm segment goes below stack, make sure there is some
1034 * space left for the stack to grow (at least 4 pages).
1036 if (addr < current->mm->start_stack &&
1037 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1041 user_addr = do_mmap (file, addr, size, prot, flags, 0);
1044 if (IS_ERR_VALUE(user_addr))
1045 err = (long)user_addr;
1047 up_write(¤t->mm->mmap_sem);
1052 down_write(&shm_ids(ns).rw_mutex);
1053 shp = shm_lock(ns, shmid);
1054 BUG_ON(IS_ERR(shp));
1056 if (shm_may_destroy(ns, shp))
1057 shm_destroy(ns, shp);
1060 up_write(&shm_ids(ns).rw_mutex);
1076 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1081 err = do_shmat(shmid, shmaddr, shmflg, &ret);
1084 force_successful_syscall_return();
1089 * detach and kill segment if marked destroyed.
1090 * The work is done in shm_close.
1092 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1094 struct mm_struct *mm = current->mm;
1095 struct vm_area_struct *vma;
1096 unsigned long addr = (unsigned long)shmaddr;
1097 int retval = -EINVAL;
1100 struct vm_area_struct *next;
1103 if (addr & ~PAGE_MASK)
1106 down_write(&mm->mmap_sem);
1109 * This function tries to be smart and unmap shm segments that
1110 * were modified by partial mlock or munmap calls:
1111 * - It first determines the size of the shm segment that should be
1112 * unmapped: It searches for a vma that is backed by shm and that
1113 * started at address shmaddr. It records it's size and then unmaps
1115 * - Then it unmaps all shm vmas that started at shmaddr and that
1116 * are within the initially determined size.
1117 * Errors from do_munmap are ignored: the function only fails if
1118 * it's called with invalid parameters or if it's called to unmap
1119 * a part of a vma. Both calls in this function are for full vmas,
1120 * the parameters are directly copied from the vma itself and always
1121 * valid - therefore do_munmap cannot fail. (famous last words?)
1124 * If it had been mremap()'d, the starting address would not
1125 * match the usual checks anyway. So assume all vma's are
1126 * above the starting address given.
1128 vma = find_vma(mm, addr);
1132 next = vma->vm_next;
1135 * Check if the starting address would match, i.e. it's
1136 * a fragment created by mprotect() and/or munmap(), or it
1137 * otherwise it starts at this address with no hassles.
1139 if ((vma->vm_ops == &shm_vm_ops) &&
1140 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1143 size = vma->vm_file->f_path.dentry->d_inode->i_size;
1144 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1146 * We discovered the size of the shm segment, so
1147 * break out of here and fall through to the next
1148 * loop that uses the size information to stop
1149 * searching for matching vma's.
1159 * We need look no further than the maximum address a fragment
1160 * could possibly have landed at. Also cast things to loff_t to
1161 * prevent overflows and make comparisons vs. equal-width types.
1163 size = PAGE_ALIGN(size);
1164 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1165 next = vma->vm_next;
1167 /* finding a matching vma now does not alter retval */
1168 if ((vma->vm_ops == &shm_vm_ops) &&
1169 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1171 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1175 #else /* CONFIG_MMU */
1176 /* under NOMMU conditions, the exact address to be destroyed must be
1179 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1180 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1186 up_write(&mm->mmap_sem);
1190 #ifdef CONFIG_PROC_FS
1191 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1193 struct shmid_kernel *shp = it;
1194 unsigned long rss = 0, swp = 0;
1196 shm_add_rss_swap(shp, &rss, &swp);
1198 #if BITS_PER_LONG <= 32
1199 #define SIZE_SPEC "%10lu"
1201 #define SIZE_SPEC "%21lu"
1204 return seq_printf(s,
1205 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1206 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1207 SIZE_SPEC " " SIZE_SPEC "\n",