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>
20 #include <linux/slab.h>
22 #include <linux/hugetlb.h>
23 #include <linux/shm.h>
24 #include <linux/init.h>
25 #include <linux/file.h>
26 #include <linux/mman.h>
27 #include <linux/shmem_fs.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/ptrace.h>
33 #include <linux/seq_file.h>
34 #include <linux/mutex.h>
35 #include <linux/vs_context.h>
36 #include <linux/vs_limit.h>
38 #include <asm/uaccess.h>
42 static struct file_operations shm_file_operations;
43 static struct vm_operations_struct shm_vm_ops;
45 static struct ipc_ids shm_ids;
47 #define shm_lock(id) ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
48 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
49 #define shm_get(id) ((struct shmid_kernel*)ipc_get(&shm_ids,id))
50 #define shm_buildid(id, seq) \
51 ipc_buildid(&shm_ids, id, seq)
53 static int newseg (key_t key, int shmflg, size_t size);
54 static void shm_open (struct vm_area_struct *shmd);
55 static void shm_close (struct vm_area_struct *shmd);
57 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
60 size_t shm_ctlmax = SHMMAX;
61 size_t shm_ctlall = SHMALL;
62 int shm_ctlmni = SHMMNI;
64 static int shm_tot; /* total number of shared memory pages */
66 void __init shm_init (void)
68 ipc_init_ids(&shm_ids, 1);
69 ipc_init_proc_interface("sysvipc/shm",
70 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
72 sysvipc_shm_proc_show);
75 static inline int shm_checkid(struct shmid_kernel *s, int id)
77 if (ipc_checkid(&shm_ids,&s->shm_perm,id))
82 static inline struct shmid_kernel *shm_rmid(int id)
84 return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
87 static inline int shm_addid(struct shmid_kernel *shp)
89 return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
94 static inline void shm_inc (int id) {
95 struct shmid_kernel *shp;
99 shp->shm_atim = get_seconds();
100 shp->shm_lprid = current->tgid;
105 /* This is called by fork, once for every shm attach. */
106 static void shm_open (struct vm_area_struct *shmd)
108 shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
112 * shm_destroy - free the struct shmid_kernel
114 * @shp: struct to free
116 * It has to be called with shp and shm_ids.mutex locked,
117 * but returns with shp unlocked and freed.
119 static void shm_destroy (struct shmid_kernel *shp)
121 struct vx_info *vxi = lookup_vx_info(shp->shm_perm.xid);
122 int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
124 vx_ipcshm_sub(vxi, shp, numpages);
129 if (!is_file_hugepages(shp->shm_file))
130 shmem_lock(shp->shm_file, 0, shp->mlock_user);
132 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
134 fput (shp->shm_file);
135 security_shm_free(shp);
141 * remove the attach descriptor shmd.
142 * free memory for segment if it is marked destroyed.
143 * The descriptor has already been removed from the current->mm->mmap list
144 * and will later be kfree()d.
146 static void shm_close (struct vm_area_struct *shmd)
148 struct file * file = shmd->vm_file;
149 int id = file->f_dentry->d_inode->i_ino;
150 struct shmid_kernel *shp;
152 mutex_lock(&shm_ids.mutex);
153 /* remove from the list of attaches of the shm segment */
156 shp->shm_lprid = current->tgid;
157 shp->shm_dtim = get_seconds();
159 if(shp->shm_nattch == 0 &&
160 shp->shm_perm.mode & SHM_DEST)
164 mutex_unlock(&shm_ids.mutex);
167 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
171 ret = shmem_mmap(file, vma);
173 vma->vm_ops = &shm_vm_ops;
174 if (!(vma->vm_flags & VM_WRITE))
175 vma->vm_flags &= ~VM_MAYWRITE;
176 shm_inc(file->f_dentry->d_inode->i_ino);
182 static struct file_operations shm_file_operations = {
185 .get_unmapped_area = shmem_get_unmapped_area,
189 static struct vm_operations_struct shm_vm_ops = {
190 .open = shm_open, /* callback for a new vm-area open */
191 .close = shm_close, /* callback for when the vm-area is released */
192 .nopage = shmem_nopage,
193 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
194 .set_policy = shmem_set_policy,
195 .get_policy = shmem_get_policy,
199 static int newseg (key_t key, int shmflg, size_t size)
202 struct shmid_kernel *shp;
203 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
208 if (size < SHMMIN || size > shm_ctlmax)
211 if (shm_tot + numpages >= shm_ctlall)
213 if (!vx_ipcshm_avail(current->vx_info, numpages))
216 shp = ipc_rcu_alloc(sizeof(*shp));
220 shp->shm_perm.key = key;
221 shp->shm_perm.xid = vx_current_xid();
222 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
223 shp->mlock_user = NULL;
225 shp->shm_perm.security = NULL;
226 error = security_shm_alloc(shp);
232 if (shmflg & SHM_HUGETLB) {
233 /* hugetlb_zero_setup takes care of mlock user accounting */
234 file = hugetlb_zero_setup(size);
235 shp->mlock_user = current->user;
237 int acctflag = VM_ACCOUNT;
239 * Do not allow no accounting for OVERCOMMIT_NEVER, even
242 if ((shmflg & SHM_NORESERVE) &&
243 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
245 sprintf (name, "SYSV%08x", key);
246 file = shmem_file_setup(name, size, acctflag);
248 error = PTR_ERR(file);
257 shp->shm_cprid = current->tgid;
259 shp->shm_atim = shp->shm_dtim = 0;
260 shp->shm_ctim = get_seconds();
261 shp->shm_segsz = size;
263 shp->id = shm_buildid(id,shp->shm_perm.seq);
264 shp->shm_file = file;
265 file->f_dentry->d_inode->i_ino = shp->id;
267 /* Hugetlb ops would have already been assigned. */
268 if (!(shmflg & SHM_HUGETLB))
269 file->f_op = &shm_file_operations;
272 vx_ipcshm_add(current->vx_info, key, numpages);
279 security_shm_free(shp);
284 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
286 struct shmid_kernel *shp;
289 mutex_lock(&shm_ids.mutex);
290 if (key == IPC_PRIVATE) {
291 err = newseg(key, shmflg, size);
292 } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
293 if (!(shmflg & IPC_CREAT))
296 err = newseg(key, shmflg, size);
297 } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
302 if (shp->shm_segsz < size)
304 else if (ipcperms(&shp->shm_perm, shmflg))
307 int shmid = shm_buildid(id, shp->shm_perm.seq);
308 err = security_shm_associate(shp, shmflg);
314 mutex_unlock(&shm_ids.mutex);
319 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
323 return copy_to_user(buf, in, sizeof(*in));
328 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
329 out.shm_segsz = in->shm_segsz;
330 out.shm_atime = in->shm_atime;
331 out.shm_dtime = in->shm_dtime;
332 out.shm_ctime = in->shm_ctime;
333 out.shm_cpid = in->shm_cpid;
334 out.shm_lpid = in->shm_lpid;
335 out.shm_nattch = in->shm_nattch;
337 return copy_to_user(buf, &out, sizeof(out));
350 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
355 struct shmid64_ds tbuf;
357 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
360 out->uid = tbuf.shm_perm.uid;
361 out->gid = tbuf.shm_perm.gid;
362 out->mode = tbuf.shm_perm.mode;
368 struct shmid_ds tbuf_old;
370 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
373 out->uid = tbuf_old.shm_perm.uid;
374 out->gid = tbuf_old.shm_perm.gid;
375 out->mode = tbuf_old.shm_perm.mode;
384 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
388 return copy_to_user(buf, in, sizeof(*in));
393 if(in->shmmax > INT_MAX)
394 out.shmmax = INT_MAX;
396 out.shmmax = (int)in->shmmax;
398 out.shmmin = in->shmmin;
399 out.shmmni = in->shmmni;
400 out.shmseg = in->shmseg;
401 out.shmall = in->shmall;
403 return copy_to_user(buf, &out, sizeof(out));
410 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
417 for (i = 0; i <= shm_ids.max_id; i++) {
418 struct shmid_kernel *shp;
425 inode = shp->shm_file->f_dentry->d_inode;
427 if (is_file_hugepages(shp->shm_file)) {
428 struct address_space *mapping = inode->i_mapping;
429 *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
431 struct shmem_inode_info *info = SHMEM_I(inode);
432 spin_lock(&info->lock);
433 *rss += inode->i_mapping->nrpages;
434 *swp += info->swapped;
435 spin_unlock(&info->lock);
440 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
442 struct shm_setbuf setbuf;
443 struct shmid_kernel *shp;
446 if (cmd < 0 || shmid < 0) {
451 version = ipc_parse_version(&cmd);
453 switch (cmd) { /* replace with proc interface ? */
456 struct shminfo64 shminfo;
458 err = security_shm_shmctl(NULL, cmd);
462 memset(&shminfo,0,sizeof(shminfo));
463 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
464 shminfo.shmmax = shm_ctlmax;
465 shminfo.shmall = shm_ctlall;
467 shminfo.shmmin = SHMMIN;
468 if(copy_shminfo_to_user (buf, &shminfo, version))
470 /* reading a integer is always atomic */
478 struct shm_info shm_info;
480 err = security_shm_shmctl(NULL, cmd);
484 memset(&shm_info,0,sizeof(shm_info));
485 mutex_lock(&shm_ids.mutex);
486 shm_info.used_ids = shm_ids.in_use;
487 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
488 shm_info.shm_tot = shm_tot;
489 shm_info.swap_attempts = 0;
490 shm_info.swap_successes = 0;
491 err = shm_ids.max_id;
492 mutex_unlock(&shm_ids.mutex);
493 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
498 err = err < 0 ? 0 : err;
504 struct shmid64_ds tbuf;
506 memset(&tbuf, 0, sizeof(tbuf));
507 shp = shm_lock(shmid);
511 } else if(cmd==SHM_STAT) {
513 if (shmid > shm_ids.max_id)
515 result = shm_buildid(shmid, shp->shm_perm.seq);
517 err = shm_checkid(shp,shmid);
523 if (ipcperms (&shp->shm_perm, S_IRUGO))
525 err = security_shm_shmctl(shp, cmd);
528 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
529 tbuf.shm_segsz = shp->shm_segsz;
530 tbuf.shm_atime = shp->shm_atim;
531 tbuf.shm_dtime = shp->shm_dtim;
532 tbuf.shm_ctime = shp->shm_ctim;
533 tbuf.shm_cpid = shp->shm_cprid;
534 tbuf.shm_lpid = shp->shm_lprid;
535 if (!is_file_hugepages(shp->shm_file))
536 tbuf.shm_nattch = shp->shm_nattch;
538 tbuf.shm_nattch = file_count(shp->shm_file) - 1;
540 if(copy_shmid_to_user (buf, &tbuf, version))
549 shp = shm_lock(shmid);
554 err = shm_checkid(shp,shmid);
558 err = audit_ipc_obj(&(shp->shm_perm));
562 if (!capable(CAP_IPC_LOCK)) {
564 if (current->euid != shp->shm_perm.uid &&
565 current->euid != shp->shm_perm.cuid)
567 if (cmd == SHM_LOCK &&
568 !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
572 err = security_shm_shmctl(shp, cmd);
577 struct user_struct * user = current->user;
578 if (!is_file_hugepages(shp->shm_file)) {
579 err = shmem_lock(shp->shm_file, 1, user);
581 shp->shm_perm.mode |= SHM_LOCKED;
582 shp->mlock_user = user;
585 } else if (!is_file_hugepages(shp->shm_file)) {
586 shmem_lock(shp->shm_file, 0, shp->mlock_user);
587 shp->shm_perm.mode &= ~SHM_LOCKED;
588 shp->mlock_user = NULL;
596 * We cannot simply remove the file. The SVID states
597 * that the block remains until the last person
598 * detaches from it, then is deleted. A shmat() on
599 * an RMID segment is legal in older Linux and if
600 * we change it apps break...
602 * Instead we set a destroyed flag, and then blow
603 * the name away when the usage hits zero.
605 mutex_lock(&shm_ids.mutex);
606 shp = shm_lock(shmid);
610 err = shm_checkid(shp, shmid);
614 err = audit_ipc_obj(&(shp->shm_perm));
618 if (current->euid != shp->shm_perm.uid &&
619 current->euid != shp->shm_perm.cuid &&
620 !capable(CAP_SYS_ADMIN)) {
625 err = security_shm_shmctl(shp, cmd);
629 if (shp->shm_nattch){
630 shp->shm_perm.mode |= SHM_DEST;
631 /* Do not find it any more */
632 shp->shm_perm.key = IPC_PRIVATE;
636 mutex_unlock(&shm_ids.mutex);
642 if (copy_shmid_from_user (&setbuf, buf, version)) {
646 mutex_lock(&shm_ids.mutex);
647 shp = shm_lock(shmid);
651 err = shm_checkid(shp,shmid);
654 err = audit_ipc_obj(&(shp->shm_perm));
657 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
661 if (current->euid != shp->shm_perm.uid &&
662 current->euid != shp->shm_perm.cuid &&
663 !capable(CAP_SYS_ADMIN)) {
667 err = security_shm_shmctl(shp, cmd);
671 shp->shm_perm.uid = setbuf.uid;
672 shp->shm_perm.gid = setbuf.gid;
673 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
674 | (setbuf.mode & S_IRWXUGO);
675 shp->shm_ctim = get_seconds();
688 mutex_unlock(&shm_ids.mutex);
697 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
699 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
700 * "raddr" thing points to kernel space, and there has to be a wrapper around
703 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
705 struct shmid_kernel *shp;
718 } else if ((addr = (ulong)shmaddr)) {
719 if (addr & (SHMLBA-1)) {
720 if (shmflg & SHM_RND)
721 addr &= ~(SHMLBA-1); /* round down */
723 #ifndef __ARCH_FORCE_SHMLBA
724 if (addr & ~PAGE_MASK)
728 flags = MAP_SHARED | MAP_FIXED;
730 if ((shmflg & SHM_REMAP))
736 if (shmflg & SHM_RDONLY) {
740 prot = PROT_READ | PROT_WRITE;
741 acc_mode = S_IRUGO | S_IWUGO;
743 if (shmflg & SHM_EXEC) {
749 * We cannot rely on the fs check since SYSV IPC does have an
750 * additional creator id...
752 shp = shm_lock(shmid);
757 err = shm_checkid(shp,shmid);
762 if (ipcperms(&shp->shm_perm, acc_mode)) {
768 err = security_shm_shmat(shp, shmaddr, shmflg);
774 file = shp->shm_file;
775 size = i_size_read(file->f_dentry->d_inode);
779 down_write(¤t->mm->mmap_sem);
780 if (addr && !(shmflg & SHM_REMAP)) {
781 user_addr = ERR_PTR(-EINVAL);
782 if (find_vma_intersection(current->mm, addr, addr + size))
785 * If shm segment goes below stack, make sure there is some
786 * space left for the stack to grow (at least 4 pages).
788 if (addr < current->mm->start_stack &&
789 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
793 user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
796 up_write(¤t->mm->mmap_sem);
798 mutex_lock(&shm_ids.mutex);
799 shp = shm_lock(shmid);
802 if(shp->shm_nattch == 0 &&
803 shp->shm_perm.mode & SHM_DEST)
807 mutex_unlock(&shm_ids.mutex);
809 *raddr = (unsigned long) user_addr;
811 if (IS_ERR(user_addr))
812 err = PTR_ERR(user_addr);
817 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
822 err = do_shmat(shmid, shmaddr, shmflg, &ret);
825 force_successful_syscall_return();
830 * detach and kill segment if marked destroyed.
831 * The work is done in shm_close.
833 asmlinkage long sys_shmdt(char __user *shmaddr)
835 struct mm_struct *mm = current->mm;
836 struct vm_area_struct *vma, *next;
837 unsigned long addr = (unsigned long)shmaddr;
839 int retval = -EINVAL;
841 if (addr & ~PAGE_MASK)
844 down_write(&mm->mmap_sem);
847 * This function tries to be smart and unmap shm segments that
848 * were modified by partial mlock or munmap calls:
849 * - It first determines the size of the shm segment that should be
850 * unmapped: It searches for a vma that is backed by shm and that
851 * started at address shmaddr. It records it's size and then unmaps
853 * - Then it unmaps all shm vmas that started at shmaddr and that
854 * are within the initially determined size.
855 * Errors from do_munmap are ignored: the function only fails if
856 * it's called with invalid parameters or if it's called to unmap
857 * a part of a vma. Both calls in this function are for full vmas,
858 * the parameters are directly copied from the vma itself and always
859 * valid - therefore do_munmap cannot fail. (famous last words?)
862 * If it had been mremap()'d, the starting address would not
863 * match the usual checks anyway. So assume all vma's are
864 * above the starting address given.
866 vma = find_vma(mm, addr);
872 * Check if the starting address would match, i.e. it's
873 * a fragment created by mprotect() and/or munmap(), or it
874 * otherwise it starts at this address with no hassles.
876 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
877 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
880 size = vma->vm_file->f_dentry->d_inode->i_size;
881 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
883 * We discovered the size of the shm segment, so
884 * break out of here and fall through to the next
885 * loop that uses the size information to stop
886 * searching for matching vma's.
896 * We need look no further than the maximum address a fragment
897 * could possibly have landed at. Also cast things to loff_t to
898 * prevent overflows and make comparisions vs. equal-width types.
900 size = PAGE_ALIGN(size);
901 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
904 /* finding a matching vma now does not alter retval */
905 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
906 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
908 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
912 up_write(&mm->mmap_sem);
916 #ifdef CONFIG_PROC_FS
917 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
919 struct shmid_kernel *shp = it;
922 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
923 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
925 if (!vx_check(shp->shm_perm.xid, VX_IDENT))
928 if (sizeof(size_t) <= sizeof(int))
929 format = SMALL_STRING;
932 return seq_printf(s, format,
939 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,