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/mutex.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
42 #include <asm/uaccess.h>
46 struct shm_file_data {
48 struct ipc_namespace *ns;
50 const struct vm_operations_struct *vm_ops;
53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55 static const struct file_operations shm_file_operations;
56 static struct vm_operations_struct shm_vm_ops;
58 static struct ipc_ids init_shm_ids;
60 #define shm_ids(ns) (*((ns)->ids[IPC_SHM_IDS]))
62 #define shm_lock(ns, id) \
63 ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
64 #define shm_unlock(shp) \
65 ipc_unlock(&(shp)->shm_perm)
66 #define shm_get(ns, id) \
67 ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
68 #define shm_buildid(ns, id, seq) \
69 ipc_buildid(&shm_ids(ns), id, seq)
71 static int newseg (struct ipc_namespace *ns, key_t key,
72 int shmflg, size_t size);
73 static void shm_open(struct vm_area_struct *vma);
74 static void shm_close(struct vm_area_struct *vma);
75 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
77 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
80 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
82 ns->ids[IPC_SHM_IDS] = ids;
83 ns->shm_ctlmax = SHMMAX;
84 ns->shm_ctlall = SHMALL;
85 ns->shm_ctlmni = SHMMNI;
90 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
93 shp->shm_perm.mode |= SHM_DEST;
94 /* Do not find it any more */
95 shp->shm_perm.key = IPC_PRIVATE;
101 int shm_init_ns(struct ipc_namespace *ns)
105 ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
109 __shm_init_ns(ns, ids);
113 void shm_exit_ns(struct ipc_namespace *ns)
116 struct shmid_kernel *shp;
118 mutex_lock(&shm_ids(ns).mutex);
119 for (i = 0; i <= shm_ids(ns).max_id; i++) {
120 shp = shm_lock(ns, i);
124 do_shm_rmid(ns, shp);
126 mutex_unlock(&shm_ids(ns).mutex);
128 ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
129 kfree(ns->ids[IPC_SHM_IDS]);
130 ns->ids[IPC_SHM_IDS] = NULL;
133 void __init shm_init (void)
135 __shm_init_ns(&init_ipc_ns, &init_shm_ids);
136 ipc_init_proc_interface("sysvipc/shm",
137 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
138 IPC_SHM_IDS, sysvipc_shm_proc_show);
141 static inline int shm_checkid(struct ipc_namespace *ns,
142 struct shmid_kernel *s, int id)
144 if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
149 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
151 return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
154 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
156 return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
161 /* This is called by fork, once for every shm attach. */
162 static void shm_open(struct vm_area_struct *vma)
164 struct file *file = vma->vm_file;
165 struct shm_file_data *sfd = shm_file_data(file);
166 struct shmid_kernel *shp;
168 shp = shm_lock(sfd->ns, sfd->id);
170 shp->shm_atim = get_seconds();
171 shp->shm_lprid = current->tgid;
177 * shm_destroy - free the struct shmid_kernel
179 * @shp: struct to free
181 * It has to be called with shp and shm_ids.mutex locked,
182 * but returns with shp unlocked and freed.
184 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
186 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
187 shm_rmid(ns, shp->id);
189 if (!is_file_hugepages(shp->shm_file))
190 shmem_lock(shp->shm_file, 0, shp->mlock_user);
192 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
194 fput (shp->shm_file);
195 security_shm_free(shp);
200 * remove the attach descriptor vma.
201 * free memory for segment if it is marked destroyed.
202 * The descriptor has already been removed from the current->mm->mmap list
203 * and will later be kfree()d.
205 static void shm_close(struct vm_area_struct *vma)
207 struct file * file = vma->vm_file;
208 struct shm_file_data *sfd = shm_file_data(file);
209 struct shmid_kernel *shp;
210 struct ipc_namespace *ns = sfd->ns;
212 mutex_lock(&shm_ids(ns).mutex);
213 /* remove from the list of attaches of the shm segment */
214 shp = shm_lock(ns, sfd->id);
216 shp->shm_lprid = current->tgid;
217 shp->shm_dtim = get_seconds();
219 if(shp->shm_nattch == 0 &&
220 shp->shm_perm.mode & SHM_DEST)
221 shm_destroy(ns, shp);
224 mutex_unlock(&shm_ids(ns).mutex);
227 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
229 struct file *file = vma->vm_file;
230 struct shm_file_data *sfd = shm_file_data(file);
232 return sfd->vm_ops->fault(vma, vmf);
236 int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
238 struct file *file = vma->vm_file;
239 struct shm_file_data *sfd = shm_file_data(file);
241 if (sfd->vm_ops->set_policy)
242 err = sfd->vm_ops->set_policy(vma, new);
246 struct mempolicy *shm_get_policy(struct vm_area_struct *vma, unsigned long addr)
248 struct file *file = vma->vm_file;
249 struct shm_file_data *sfd = shm_file_data(file);
250 struct mempolicy *pol = NULL;
252 if (sfd->vm_ops->get_policy)
253 pol = sfd->vm_ops->get_policy(vma, addr);
254 else if (vma->vm_policy)
255 pol = vma->vm_policy;
257 pol = current->mempolicy;
262 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
264 struct shm_file_data *sfd = shm_file_data(file);
267 ret = sfd->file->f_op->mmap(sfd->file, vma);
270 sfd->vm_ops = vma->vm_ops;
272 BUG_ON(!sfd->vm_ops->fault);
274 vma->vm_ops = &shm_vm_ops;
280 static int shm_release(struct inode *ino, struct file *file)
282 struct shm_file_data *sfd = shm_file_data(file);
285 shm_file_data(file) = NULL;
290 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
292 int (*fsync) (struct file *, struct dentry *, int datasync);
293 struct shm_file_data *sfd = shm_file_data(file);
296 fsync = sfd->file->f_op->fsync;
298 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
302 static unsigned long shm_get_unmapped_area(struct file *file,
303 unsigned long addr, unsigned long len, unsigned long pgoff,
306 struct shm_file_data *sfd = shm_file_data(file);
307 return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
310 int is_file_shm_hugepages(struct file *file)
314 if (file->f_op == &shm_file_operations) {
315 struct shm_file_data *sfd;
316 sfd = shm_file_data(file);
317 ret = is_file_hugepages(sfd->file);
322 static const struct file_operations shm_file_operations = {
325 .release = shm_release,
326 .get_unmapped_area = shm_get_unmapped_area,
329 static struct vm_operations_struct shm_vm_ops = {
330 .open = shm_open, /* callback for a new vm-area open */
331 .close = shm_close, /* callback for when the vm-area is released */
333 #if defined(CONFIG_NUMA)
334 .set_policy = shm_set_policy,
335 .get_policy = shm_get_policy,
339 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
342 struct shmid_kernel *shp;
343 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
348 if (size < SHMMIN || size > ns->shm_ctlmax)
351 if (ns->shm_tot + numpages > ns->shm_ctlall)
354 shp = ipc_rcu_alloc(sizeof(*shp));
358 shp->shm_perm.key = key;
359 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
360 shp->mlock_user = NULL;
362 shp->shm_perm.security = NULL;
363 error = security_shm_alloc(shp);
369 sprintf (name, "SYSV%08x", key);
370 if (shmflg & SHM_HUGETLB) {
371 /* hugetlb_file_setup takes care of mlock user accounting */
372 file = hugetlb_file_setup(name, size);
373 shp->mlock_user = current->user;
375 int acctflag = VM_ACCOUNT;
377 * Do not allow no accounting for OVERCOMMIT_NEVER, even
380 if ((shmflg & SHM_NORESERVE) &&
381 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
383 file = shmem_file_setup(name, size, acctflag);
385 error = PTR_ERR(file);
390 id = shm_addid(ns, shp);
394 shp->shm_cprid = current->tgid;
396 shp->shm_atim = shp->shm_dtim = 0;
397 shp->shm_ctim = get_seconds();
398 shp->shm_segsz = size;
400 shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
401 shp->shm_file = file;
403 * shmid gets reported as "inode#" in /proc/pid/maps.
404 * proc-ps tools use this. Changing this will break them.
406 file->f_dentry->d_inode->i_ino = shp->id;
408 ns->shm_tot += numpages;
415 security_shm_free(shp);
420 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
422 struct shmid_kernel *shp;
424 struct ipc_namespace *ns;
426 ns = current->nsproxy->ipc_ns;
428 mutex_lock(&shm_ids(ns).mutex);
429 if (key == IPC_PRIVATE) {
430 err = newseg(ns, key, shmflg, size);
431 } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
432 if (!(shmflg & IPC_CREAT))
435 err = newseg(ns, key, shmflg, size);
436 } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
439 shp = shm_lock(ns, id);
441 if (shp->shm_segsz < size)
443 else if (ipcperms(&shp->shm_perm, shmflg))
446 int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
447 err = security_shm_associate(shp, shmflg);
453 mutex_unlock(&shm_ids(ns).mutex);
458 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
462 return copy_to_user(buf, in, sizeof(*in));
467 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
468 out.shm_segsz = in->shm_segsz;
469 out.shm_atime = in->shm_atime;
470 out.shm_dtime = in->shm_dtime;
471 out.shm_ctime = in->shm_ctime;
472 out.shm_cpid = in->shm_cpid;
473 out.shm_lpid = in->shm_lpid;
474 out.shm_nattch = in->shm_nattch;
476 return copy_to_user(buf, &out, sizeof(out));
489 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
494 struct shmid64_ds tbuf;
496 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
499 out->uid = tbuf.shm_perm.uid;
500 out->gid = tbuf.shm_perm.gid;
501 out->mode = tbuf.shm_perm.mode;
507 struct shmid_ds tbuf_old;
509 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
512 out->uid = tbuf_old.shm_perm.uid;
513 out->gid = tbuf_old.shm_perm.gid;
514 out->mode = tbuf_old.shm_perm.mode;
523 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
527 return copy_to_user(buf, in, sizeof(*in));
532 if(in->shmmax > INT_MAX)
533 out.shmmax = INT_MAX;
535 out.shmmax = (int)in->shmmax;
537 out.shmmin = in->shmmin;
538 out.shmmni = in->shmmni;
539 out.shmseg = in->shmseg;
540 out.shmall = in->shmall;
542 return copy_to_user(buf, &out, sizeof(out));
549 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
557 for (i = 0; i <= shm_ids(ns).max_id; i++) {
558 struct shmid_kernel *shp;
561 shp = shm_get(ns, i);
565 inode = shp->shm_file->f_path.dentry->d_inode;
567 if (is_file_hugepages(shp->shm_file)) {
568 struct address_space *mapping = inode->i_mapping;
569 *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
571 struct shmem_inode_info *info = SHMEM_I(inode);
572 spin_lock(&info->lock);
573 *rss += inode->i_mapping->nrpages;
574 *swp += info->swapped;
575 spin_unlock(&info->lock);
580 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
582 struct shm_setbuf setbuf;
583 struct shmid_kernel *shp;
585 struct ipc_namespace *ns;
587 if (cmd < 0 || shmid < 0) {
592 version = ipc_parse_version(&cmd);
593 ns = current->nsproxy->ipc_ns;
595 switch (cmd) { /* replace with proc interface ? */
598 struct shminfo64 shminfo;
600 err = security_shm_shmctl(NULL, cmd);
604 memset(&shminfo,0,sizeof(shminfo));
605 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
606 shminfo.shmmax = ns->shm_ctlmax;
607 shminfo.shmall = ns->shm_ctlall;
609 shminfo.shmmin = SHMMIN;
610 if(copy_shminfo_to_user (buf, &shminfo, version))
612 /* reading a integer is always atomic */
613 err= shm_ids(ns).max_id;
620 struct shm_info shm_info;
622 err = security_shm_shmctl(NULL, cmd);
626 memset(&shm_info,0,sizeof(shm_info));
627 mutex_lock(&shm_ids(ns).mutex);
628 shm_info.used_ids = shm_ids(ns).in_use;
629 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
630 shm_info.shm_tot = ns->shm_tot;
631 shm_info.swap_attempts = 0;
632 shm_info.swap_successes = 0;
633 err = shm_ids(ns).max_id;
634 mutex_unlock(&shm_ids(ns).mutex);
635 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
640 err = err < 0 ? 0 : err;
646 struct shmid64_ds tbuf;
648 memset(&tbuf, 0, sizeof(tbuf));
649 shp = shm_lock(ns, shmid);
653 } else if(cmd==SHM_STAT) {
655 if (shmid > shm_ids(ns).max_id)
657 result = shm_buildid(ns, shmid, shp->shm_perm.seq);
659 err = shm_checkid(ns, shp,shmid);
665 if (ipcperms (&shp->shm_perm, S_IRUGO))
667 err = security_shm_shmctl(shp, cmd);
670 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
671 tbuf.shm_segsz = shp->shm_segsz;
672 tbuf.shm_atime = shp->shm_atim;
673 tbuf.shm_dtime = shp->shm_dtim;
674 tbuf.shm_ctime = shp->shm_ctim;
675 tbuf.shm_cpid = shp->shm_cprid;
676 tbuf.shm_lpid = shp->shm_lprid;
677 tbuf.shm_nattch = shp->shm_nattch;
679 if(copy_shmid_to_user (buf, &tbuf, version))
688 shp = shm_lock(ns, shmid);
693 err = shm_checkid(ns, shp,shmid);
697 err = audit_ipc_obj(&(shp->shm_perm));
701 if (!capable(CAP_IPC_LOCK)) {
703 if (current->euid != shp->shm_perm.uid &&
704 current->euid != shp->shm_perm.cuid)
706 if (cmd == SHM_LOCK &&
707 !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
711 err = security_shm_shmctl(shp, cmd);
716 struct user_struct * user = current->user;
717 if (!is_file_hugepages(shp->shm_file)) {
718 err = shmem_lock(shp->shm_file, 1, user);
719 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
720 shp->shm_perm.mode |= SHM_LOCKED;
721 shp->mlock_user = user;
724 } else if (!is_file_hugepages(shp->shm_file)) {
725 shmem_lock(shp->shm_file, 0, shp->mlock_user);
726 shp->shm_perm.mode &= ~SHM_LOCKED;
727 shp->mlock_user = NULL;
735 * We cannot simply remove the file. The SVID states
736 * that the block remains until the last person
737 * detaches from it, then is deleted. A shmat() on
738 * an RMID segment is legal in older Linux and if
739 * we change it apps break...
741 * Instead we set a destroyed flag, and then blow
742 * the name away when the usage hits zero.
744 mutex_lock(&shm_ids(ns).mutex);
745 shp = shm_lock(ns, shmid);
749 err = shm_checkid(ns, shp, shmid);
753 err = audit_ipc_obj(&(shp->shm_perm));
757 if (current->euid != shp->shm_perm.uid &&
758 current->euid != shp->shm_perm.cuid &&
759 !capable(CAP_SYS_ADMIN)) {
764 err = security_shm_shmctl(shp, cmd);
768 do_shm_rmid(ns, shp);
769 mutex_unlock(&shm_ids(ns).mutex);
775 if (copy_shmid_from_user (&setbuf, buf, version)) {
779 mutex_lock(&shm_ids(ns).mutex);
780 shp = shm_lock(ns, shmid);
784 err = shm_checkid(ns, shp,shmid);
787 err = audit_ipc_obj(&(shp->shm_perm));
790 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
794 if (current->euid != shp->shm_perm.uid &&
795 current->euid != shp->shm_perm.cuid &&
796 !capable(CAP_SYS_ADMIN)) {
800 err = security_shm_shmctl(shp, cmd);
804 shp->shm_perm.uid = setbuf.uid;
805 shp->shm_perm.gid = setbuf.gid;
806 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
807 | (setbuf.mode & S_IRWXUGO);
808 shp->shm_ctim = get_seconds();
821 mutex_unlock(&shm_ids(ns).mutex);
830 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
832 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
833 * "raddr" thing points to kernel space, and there has to be a wrapper around
836 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
838 struct shmid_kernel *shp;
846 unsigned long user_addr;
847 struct ipc_namespace *ns;
848 struct shm_file_data *sfd;
855 else if ((addr = (ulong)shmaddr)) {
856 if (addr & (SHMLBA-1)) {
857 if (shmflg & SHM_RND)
858 addr &= ~(SHMLBA-1); /* round down */
860 #ifndef __ARCH_FORCE_SHMLBA
861 if (addr & ~PAGE_MASK)
865 flags = MAP_SHARED | MAP_FIXED;
867 if ((shmflg & SHM_REMAP))
873 if (shmflg & SHM_RDONLY) {
878 prot = PROT_READ | PROT_WRITE;
879 acc_mode = S_IRUGO | S_IWUGO;
880 f_mode = FMODE_READ | FMODE_WRITE;
882 if (shmflg & SHM_EXEC) {
888 * We cannot rely on the fs check since SYSV IPC does have an
889 * additional creator id...
891 ns = current->nsproxy->ipc_ns;
892 shp = shm_lock(ns, shmid);
896 err = shm_checkid(ns, shp,shmid);
901 if (ipcperms(&shp->shm_perm, acc_mode))
904 err = security_shm_shmat(shp, shmaddr, shmflg);
908 path.dentry = dget(shp->shm_file->f_path.dentry);
909 path.mnt = mntget(shp->shm_file->f_path.mnt);
911 size = i_size_read(path.dentry->d_inode);
915 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
920 file = get_empty_filp();
924 file->f_op = &shm_file_operations;
925 file->private_data = sfd;
927 file->f_mapping = shp->shm_file->f_mapping;
928 file->f_mode = f_mode;
930 sfd->ns = get_ipc_ns(ns);
931 sfd->file = shp->shm_file;
934 down_write(¤t->mm->mmap_sem);
935 if (addr && !(shmflg & SHM_REMAP)) {
937 if (find_vma_intersection(current->mm, addr, addr + size))
940 * If shm segment goes below stack, make sure there is some
941 * space left for the stack to grow (at least 4 pages).
943 if (addr < current->mm->start_stack &&
944 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
948 user_addr = do_mmap (file, addr, size, prot, flags, 0);
951 if (IS_ERR_VALUE(user_addr))
952 err = (long)user_addr;
954 up_write(¤t->mm->mmap_sem);
959 mutex_lock(&shm_ids(ns).mutex);
960 shp = shm_lock(ns, shmid);
963 if(shp->shm_nattch == 0 &&
964 shp->shm_perm.mode & SHM_DEST)
965 shm_destroy(ns, shp);
968 mutex_unlock(&shm_ids(ns).mutex);
985 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
990 err = do_shmat(shmid, shmaddr, shmflg, &ret);
993 force_successful_syscall_return();
998 * detach and kill segment if marked destroyed.
999 * The work is done in shm_close.
1001 asmlinkage long sys_shmdt(char __user *shmaddr)
1003 struct mm_struct *mm = current->mm;
1004 struct vm_area_struct *vma, *next;
1005 unsigned long addr = (unsigned long)shmaddr;
1007 int retval = -EINVAL;
1009 if (addr & ~PAGE_MASK)
1012 down_write(&mm->mmap_sem);
1015 * This function tries to be smart and unmap shm segments that
1016 * were modified by partial mlock or munmap calls:
1017 * - It first determines the size of the shm segment that should be
1018 * unmapped: It searches for a vma that is backed by shm and that
1019 * started at address shmaddr. It records it's size and then unmaps
1021 * - Then it unmaps all shm vmas that started at shmaddr and that
1022 * are within the initially determined size.
1023 * Errors from do_munmap are ignored: the function only fails if
1024 * it's called with invalid parameters or if it's called to unmap
1025 * a part of a vma. Both calls in this function are for full vmas,
1026 * the parameters are directly copied from the vma itself and always
1027 * valid - therefore do_munmap cannot fail. (famous last words?)
1030 * If it had been mremap()'d, the starting address would not
1031 * match the usual checks anyway. So assume all vma's are
1032 * above the starting address given.
1034 vma = find_vma(mm, addr);
1037 next = vma->vm_next;
1040 * Check if the starting address would match, i.e. it's
1041 * a fragment created by mprotect() and/or munmap(), or it
1042 * otherwise it starts at this address with no hassles.
1044 if ((vma->vm_ops == &shm_vm_ops) &&
1045 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1048 size = vma->vm_file->f_path.dentry->d_inode->i_size;
1049 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051 * We discovered the size of the shm segment, so
1052 * break out of here and fall through to the next
1053 * loop that uses the size information to stop
1054 * searching for matching vma's.
1064 * We need look no further than the maximum address a fragment
1065 * could possibly have landed at. Also cast things to loff_t to
1066 * prevent overflows and make comparisions vs. equal-width types.
1068 size = PAGE_ALIGN(size);
1069 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1070 next = vma->vm_next;
1072 /* finding a matching vma now does not alter retval */
1073 if ((vma->vm_ops == &shm_vm_ops) &&
1074 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1076 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1080 up_write(&mm->mmap_sem);
1084 #ifdef CONFIG_PROC_FS
1085 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1087 struct shmid_kernel *shp = it;
1090 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1091 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1093 if (sizeof(size_t) <= sizeof(int))
1094 format = SMALL_STRING;
1096 format = BIG_STRING;
1097 return seq_printf(s, format,