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>
18 #include <linux/config.h>
19 #include <linux/slab.h>
21 #include <linux/hugetlb.h>
22 #include <linux/shm.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
25 #include <linux/mman.h>
26 #include <linux/proc_fs.h>
27 #include <linux/shmem_fs.h>
28 #include <linux/security.h>
29 #include <asm/uaccess.h>
33 #define shm_flags shm_perm.mode
35 static struct file_operations shm_file_operations;
36 static struct vm_operations_struct shm_vm_ops;
38 static struct ipc_ids shm_ids;
40 #define shm_lock(id) ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
41 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
42 #define shm_get(id) ((struct shmid_kernel*)ipc_get(&shm_ids,id))
43 #define shm_buildid(id, seq) \
44 ipc_buildid(&shm_ids, id, seq)
46 static int newseg (key_t key, int shmflg, size_t size);
47 static void shm_open (struct vm_area_struct *shmd);
48 static void shm_close (struct vm_area_struct *shmd);
50 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
53 size_t shm_ctlmax = SHMMAX;
54 size_t shm_ctlall = SHMALL;
55 int shm_ctlmni = SHMMNI;
57 static int shm_tot; /* total number of shared memory pages */
59 void __init shm_init (void)
61 ipc_init_ids(&shm_ids, 1);
63 create_proc_read_entry("sysvipc/shm", 0, 0, sysvipc_shm_read_proc, NULL);
67 static inline int shm_checkid(struct shmid_kernel *s, int id)
69 if (ipc_checkid(&shm_ids,&s->shm_perm,id))
74 static inline struct shmid_kernel *shm_rmid(int id)
76 return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
79 static inline int shm_addid(struct shmid_kernel *shp)
81 return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni+1);
86 static inline void shm_inc (int id) {
87 struct shmid_kernel *shp;
89 if(!(shp = shm_lock(id)))
91 shp->shm_atim = get_seconds();
92 shp->shm_lprid = current->tgid;
97 /* This is called by fork, once for every shm attach. */
98 static void shm_open (struct vm_area_struct *shmd)
100 shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
104 * shm_destroy - free the struct shmid_kernel
106 * @shp: struct to free
108 * It has to be called with shp and shm_ids.sem locked,
109 * but returns with shp unlocked and freed.
111 static void shm_destroy (struct shmid_kernel *shp)
113 shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
116 if (!is_file_hugepages(shp->shm_file))
117 shmem_lock(shp->shm_file, 0);
118 fput (shp->shm_file);
119 security_shm_free(shp);
120 ipc_rcu_free(shp, sizeof(struct shmid_kernel));
124 * remove the attach descriptor shmd.
125 * free memory for segment if it is marked destroyed.
126 * The descriptor has already been removed from the current->mm->mmap list
127 * and will later be kfree()d.
129 static void shm_close (struct vm_area_struct *shmd)
131 struct file * file = shmd->vm_file;
132 int id = file->f_dentry->d_inode->i_ino;
133 struct shmid_kernel *shp;
136 /* remove from the list of attaches of the shm segment */
137 if(!(shp = shm_lock(id)))
139 shp->shm_lprid = current->tgid;
140 shp->shm_dtim = get_seconds();
142 if(shp->shm_nattch == 0 &&
143 shp->shm_flags & SHM_DEST)
150 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
153 vma->vm_ops = &shm_vm_ops;
154 shm_inc(file->f_dentry->d_inode->i_ino);
158 static struct file_operations shm_file_operations = {
162 static struct vm_operations_struct shm_vm_ops = {
163 .open = shm_open, /* callback for a new vm-area open */
164 .close = shm_close, /* callback for when the vm-area is released */
165 .nopage = shmem_nopage,
168 static int newseg (key_t key, int shmflg, size_t size)
171 struct shmid_kernel *shp;
172 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
177 if (size < SHMMIN || size > shm_ctlmax)
180 if (shm_tot + numpages >= shm_ctlall)
183 shp = ipc_rcu_alloc(sizeof(*shp));
187 shp->shm_perm.key = key;
188 shp->shm_perm.xid = current->xid;
189 shp->shm_flags = (shmflg & S_IRWXUGO);
191 shp->shm_perm.security = NULL;
192 error = security_shm_alloc(shp);
194 ipc_rcu_free(shp, sizeof(*shp));
198 if (shmflg & SHM_HUGETLB)
199 file = hugetlb_zero_setup(size);
201 sprintf (name, "SYSV%08x", key);
202 file = shmem_file_setup(name, size, VM_ACCOUNT);
204 error = PTR_ERR(file);
213 shp->shm_cprid = current->tgid;
215 shp->shm_atim = shp->shm_dtim = 0;
216 shp->shm_ctim = get_seconds();
217 shp->shm_segsz = size;
219 shp->id = shm_buildid(id,shp->shm_perm.seq);
220 shp->shm_file = file;
221 file->f_dentry->d_inode->i_ino = shp->id;
222 if (shmflg & SHM_HUGETLB)
223 set_file_hugepages(file);
225 file->f_op = &shm_file_operations;
233 security_shm_free(shp);
234 ipc_rcu_free(shp, sizeof(*shp));
238 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
240 struct shmid_kernel *shp;
244 if (key == IPC_PRIVATE) {
245 err = newseg(key, shmflg, size);
246 } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
247 if (!(shmflg & IPC_CREAT))
250 err = newseg(key, shmflg, size);
251 } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
257 if (shp->shm_segsz < size)
259 else if (ipcperms(&shp->shm_perm, shmflg))
262 int shmid = shm_buildid(id, shp->shm_perm.seq);
263 err = security_shm_associate(shp, shmflg);
274 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
278 return copy_to_user(buf, in, sizeof(*in));
283 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
284 out.shm_segsz = in->shm_segsz;
285 out.shm_atime = in->shm_atime;
286 out.shm_dtime = in->shm_dtime;
287 out.shm_ctime = in->shm_ctime;
288 out.shm_cpid = in->shm_cpid;
289 out.shm_lpid = in->shm_lpid;
290 out.shm_nattch = in->shm_nattch;
292 return copy_to_user(buf, &out, sizeof(out));
305 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
310 struct shmid64_ds tbuf;
312 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
315 out->uid = tbuf.shm_perm.uid;
316 out->gid = tbuf.shm_perm.gid;
317 out->mode = tbuf.shm_flags;
323 struct shmid_ds tbuf_old;
325 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
328 out->uid = tbuf_old.shm_perm.uid;
329 out->gid = tbuf_old.shm_perm.gid;
330 out->mode = tbuf_old.shm_flags;
339 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
343 return copy_to_user(buf, in, sizeof(*in));
348 if(in->shmmax > INT_MAX)
349 out.shmmax = INT_MAX;
351 out.shmmax = (int)in->shmmax;
353 out.shmmin = in->shmmin;
354 out.shmmni = in->shmmni;
355 out.shmseg = in->shmseg;
356 out.shmall = in->shmall;
358 return copy_to_user(buf, &out, sizeof(out));
365 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
372 for (i = 0; i <= shm_ids.max_id; i++) {
373 struct shmid_kernel *shp;
380 inode = shp->shm_file->f_dentry->d_inode;
382 if (is_file_hugepages(shp->shm_file)) {
383 struct address_space *mapping = inode->i_mapping;
384 *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
386 struct shmem_inode_info *info = SHMEM_I(inode);
387 spin_lock(&info->lock);
388 *rss += inode->i_mapping->nrpages;
389 *swp += info->swapped;
390 spin_unlock(&info->lock);
395 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
397 struct shm_setbuf setbuf;
398 struct shmid_kernel *shp;
401 if (cmd < 0 || shmid < 0) {
406 version = ipc_parse_version(&cmd);
408 switch (cmd) { /* replace with proc interface ? */
411 struct shminfo64 shminfo;
413 err = security_shm_shmctl(NULL, cmd);
417 memset(&shminfo,0,sizeof(shminfo));
418 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
419 shminfo.shmmax = shm_ctlmax;
420 shminfo.shmall = shm_ctlall;
422 shminfo.shmmin = SHMMIN;
423 if(copy_shminfo_to_user (buf, &shminfo, version))
425 /* reading a integer is always atomic */
433 struct shm_info shm_info;
435 err = security_shm_shmctl(NULL, cmd);
439 memset(&shm_info,0,sizeof(shm_info));
441 shm_info.used_ids = shm_ids.in_use;
442 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
443 shm_info.shm_tot = shm_tot;
444 shm_info.swap_attempts = 0;
445 shm_info.swap_successes = 0;
446 err = shm_ids.max_id;
448 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
453 err = err < 0 ? 0 : err;
459 struct shmid64_ds tbuf;
461 memset(&tbuf, 0, sizeof(tbuf));
462 shp = shm_lock(shmid);
466 } else if(cmd==SHM_STAT) {
468 if (shmid > shm_ids.max_id)
470 result = shm_buildid(shmid, shp->shm_perm.seq);
472 err = shm_checkid(shp,shmid);
478 if (ipcperms (&shp->shm_perm, S_IRUGO))
480 err = security_shm_shmctl(shp, cmd);
483 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
484 tbuf.shm_segsz = shp->shm_segsz;
485 tbuf.shm_atime = shp->shm_atim;
486 tbuf.shm_dtime = shp->shm_dtim;
487 tbuf.shm_ctime = shp->shm_ctim;
488 tbuf.shm_cpid = shp->shm_cprid;
489 tbuf.shm_lpid = shp->shm_lprid;
490 if (!is_file_hugepages(shp->shm_file))
491 tbuf.shm_nattch = shp->shm_nattch;
493 tbuf.shm_nattch = file_count(shp->shm_file) - 1;
495 if(copy_shmid_to_user (buf, &tbuf, version))
504 /* Allow superuser to lock segment in memory */
505 /* Should the pages be faulted in here or leave it to user? */
506 /* need to determine interaction with current->swappable */
507 if (!capable(CAP_IPC_LOCK)) {
512 shp = shm_lock(shmid);
517 err = shm_checkid(shp,shmid);
521 err = security_shm_shmctl(shp, cmd);
526 if (!is_file_hugepages(shp->shm_file))
527 shmem_lock(shp->shm_file, 1);
528 shp->shm_flags |= SHM_LOCKED;
530 if (!is_file_hugepages(shp->shm_file))
531 shmem_lock(shp->shm_file, 0);
532 shp->shm_flags &= ~SHM_LOCKED;
540 * We cannot simply remove the file. The SVID states
541 * that the block remains until the last person
542 * detaches from it, then is deleted. A shmat() on
543 * an RMID segment is legal in older Linux and if
544 * we change it apps break...
546 * Instead we set a destroyed flag, and then blow
547 * the name away when the usage hits zero.
550 shp = shm_lock(shmid);
554 err = shm_checkid(shp, shmid);
558 if (current->euid != shp->shm_perm.uid &&
559 current->euid != shp->shm_perm.cuid &&
560 !capable(CAP_SYS_ADMIN)) {
565 err = security_shm_shmctl(shp, cmd);
569 if (shp->shm_nattch){
570 shp->shm_flags |= SHM_DEST;
571 /* Do not find it any more */
572 shp->shm_perm.key = IPC_PRIVATE;
582 if (copy_shmid_from_user (&setbuf, buf, version)) {
587 shp = shm_lock(shmid);
591 err = shm_checkid(shp,shmid);
595 if (current->euid != shp->shm_perm.uid &&
596 current->euid != shp->shm_perm.cuid &&
597 !capable(CAP_SYS_ADMIN)) {
601 err = security_shm_shmctl(shp, cmd);
605 shp->shm_perm.uid = setbuf.uid;
606 shp->shm_perm.gid = setbuf.gid;
607 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
608 | (setbuf.mode & S_IRWXUGO);
609 shp->shm_ctim = get_seconds();
631 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
633 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
634 * "raddr" thing points to kernel space, and there has to be a wrapper around
637 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
639 struct shmid_kernel *shp;
646 unsigned long o_flags;
653 } else if ((addr = (ulong)shmaddr)) {
654 if (addr & (SHMLBA-1)) {
655 if (shmflg & SHM_RND)
656 addr &= ~(SHMLBA-1); /* round down */
658 #ifndef __ARCH_FORCE_SHMLBA
659 if (addr & ~PAGE_MASK)
663 flags = MAP_SHARED | MAP_FIXED;
665 if ((shmflg & SHM_REMAP))
671 if (shmflg & SHM_RDONLY) {
676 prot = PROT_READ | PROT_WRITE;
678 acc_mode = S_IRUGO | S_IWUGO;
682 * We cannot rely on the fs check since SYSV IPC does have an
683 * additional creator id...
685 shp = shm_lock(shmid);
690 err = shm_checkid(shp,shmid);
695 if (ipcperms(&shp->shm_perm, acc_mode)) {
701 err = security_shm_shmat(shp, shmaddr, shmflg);
707 file = shp->shm_file;
708 size = i_size_read(file->f_dentry->d_inode);
712 down_write(¤t->mm->mmap_sem);
713 if (addr && !(shmflg & SHM_REMAP)) {
714 user_addr = ERR_PTR(-EINVAL);
715 if (find_vma_intersection(current->mm, addr, addr + size))
718 * If shm segment goes below stack, make sure there is some
719 * space left for the stack to grow (at least 4 pages).
721 if (addr < current->mm->start_stack &&
722 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
726 user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
729 up_write(¤t->mm->mmap_sem);
732 if(!(shp = shm_lock(shmid)))
735 if(shp->shm_nattch == 0 &&
736 shp->shm_flags & SHM_DEST)
742 *raddr = (unsigned long) user_addr;
744 if (IS_ERR(user_addr))
745 err = PTR_ERR(user_addr);
751 * detach and kill segment if marked destroyed.
752 * The work is done in shm_close.
754 asmlinkage long sys_shmdt(char __user *shmaddr)
756 struct mm_struct *mm = current->mm;
757 struct vm_area_struct *vma, *next;
758 unsigned long addr = (unsigned long)shmaddr;
760 int retval = -EINVAL;
762 down_write(&mm->mmap_sem);
765 * This function tries to be smart and unmap shm segments that
766 * were modified by partial mlock or munmap calls:
767 * - It first determines the size of the shm segment that should be
768 * unmapped: It searches for a vma that is backed by shm and that
769 * started at address shmaddr. It records it's size and then unmaps
771 * - Then it unmaps all shm vmas that started at shmaddr and that
772 * are within the initially determined size.
773 * Errors from do_munmap are ignored: the function only fails if
774 * it's called with invalid parameters or if it's called to unmap
775 * a part of a vma. Both calls in this function are for full vmas,
776 * the parameters are directly copied from the vma itself and always
777 * valid - therefore do_munmap cannot fail. (famous last words?)
780 * If it had been mremap()'d, the starting address would not
781 * match the usual checks anyway. So assume all vma's are
782 * above the starting address given.
784 vma = find_vma(mm, addr);
790 * Check if the starting address would match, i.e. it's
791 * a fragment created by mprotect() and/or munmap(), or it
792 * otherwise it starts at this address with no hassles.
794 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
795 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
798 size = vma->vm_file->f_dentry->d_inode->i_size;
799 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
801 * We discovered the size of the shm segment, so
802 * break out of here and fall through to the next
803 * loop that uses the size information to stop
804 * searching for matching vma's.
814 * We need look no further than the maximum address a fragment
815 * could possibly have landed at. Also cast things to loff_t to
816 * prevent overflows and make comparisions vs. equal-width types.
818 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
821 /* finding a matching vma now does not alter retval */
822 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
823 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
825 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
829 up_write(&mm->mmap_sem);
833 #ifdef CONFIG_PROC_FS
834 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
841 len += sprintf(buffer, " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n");
843 for(i = 0; i <= shm_ids.max_id; i++) {
844 struct shmid_kernel* shp;
848 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
849 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
852 if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
856 if (sizeof(size_t) <= sizeof(int))
857 format = SMALL_STRING;
860 len += sprintf(buffer + len, format,
862 shm_buildid(i, shp->shm_perm.seq),
867 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
882 if(pos > offset + length)
889 *start = buffer + (offset - begin);
890 len -= (offset - begin);