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 <linux/vs_base.h>
31 #include <asm/uaccess.h>
35 #define shm_flags shm_perm.mode
37 static struct file_operations shm_file_operations;
38 static struct vm_operations_struct shm_vm_ops;
40 static struct ipc_ids shm_ids;
42 #define shm_lock(id) ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
43 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
44 #define shm_get(id) ((struct shmid_kernel*)ipc_get(&shm_ids,id))
45 #define shm_buildid(id, seq) \
46 ipc_buildid(&shm_ids, id, seq)
48 static int newseg (key_t key, int shmflg, size_t size);
49 static void shm_open (struct vm_area_struct *shmd);
50 static void shm_close (struct vm_area_struct *shmd);
52 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
55 size_t shm_ctlmax = SHMMAX;
56 size_t shm_ctlall = SHMALL;
57 int shm_ctlmni = SHMMNI;
59 static int shm_tot; /* total number of shared memory pages */
61 void __init shm_init (void)
63 ipc_init_ids(&shm_ids, 1);
65 create_proc_read_entry("sysvipc/shm", 0, NULL, sysvipc_shm_read_proc, NULL);
69 static inline int shm_checkid(struct shmid_kernel *s, int id)
71 if (ipc_checkid(&shm_ids,&s->shm_perm,id))
76 static inline struct shmid_kernel *shm_rmid(int id)
78 return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
81 static inline int shm_addid(struct shmid_kernel *shp)
83 return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
88 static inline void shm_inc (int id) {
89 struct shmid_kernel *shp;
91 if(!(shp = shm_lock(id)))
93 shp->shm_atim = get_seconds();
94 shp->shm_lprid = current->tgid;
99 /* This is called by fork, once for every shm attach. */
100 static void shm_open (struct vm_area_struct *shmd)
102 shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
106 * shm_destroy - free the struct shmid_kernel
108 * @shp: struct to free
110 * It has to be called with shp and shm_ids.sem locked,
111 * but returns with shp unlocked and freed.
113 static void shm_destroy (struct shmid_kernel *shp)
115 shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
118 if (!is_file_hugepages(shp->shm_file))
119 shmem_lock(shp->shm_file, 0, shp->mlock_user);
121 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
123 fput (shp->shm_file);
124 security_shm_free(shp);
129 * remove the attach descriptor shmd.
130 * free memory for segment if it is marked destroyed.
131 * The descriptor has already been removed from the current->mm->mmap list
132 * and will later be kfree()d.
134 static void shm_close (struct vm_area_struct *shmd)
136 struct file * file = shmd->vm_file;
137 int id = file->f_dentry->d_inode->i_ino;
138 struct shmid_kernel *shp;
141 /* remove from the list of attaches of the shm segment */
142 if(!(shp = shm_lock(id)))
144 shp->shm_lprid = current->tgid;
145 shp->shm_dtim = get_seconds();
147 if(shp->shm_nattch == 0 &&
148 shp->shm_flags & SHM_DEST)
155 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
158 vma->vm_ops = &shm_vm_ops;
159 shm_inc(file->f_dentry->d_inode->i_ino);
163 static struct file_operations shm_file_operations = {
167 static struct vm_operations_struct shm_vm_ops = {
168 .open = shm_open, /* callback for a new vm-area open */
169 .close = shm_close, /* callback for when the vm-area is released */
170 .nopage = shmem_nopage,
172 .set_policy = shmem_set_policy,
173 .get_policy = shmem_get_policy,
177 static int newseg (key_t key, int shmflg, size_t size)
180 struct shmid_kernel *shp;
181 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
186 if (size < SHMMIN || size > shm_ctlmax)
189 if (shm_tot + numpages >= shm_ctlall)
192 shp = ipc_rcu_alloc(sizeof(*shp));
196 shp->shm_perm.key = key;
197 shp->shm_perm.xid = current->xid;
198 shp->shm_flags = (shmflg & S_IRWXUGO);
199 shp->mlock_user = NULL;
201 shp->shm_perm.security = NULL;
202 error = security_shm_alloc(shp);
208 if (shmflg & SHM_HUGETLB) {
209 /* hugetlb_zero_setup takes care of mlock user accounting */
210 file = hugetlb_zero_setup(size);
211 shp->mlock_user = current->user;
213 sprintf (name, "SYSV%08x", key);
214 file = shmem_file_setup(name, size, VM_ACCOUNT);
216 error = PTR_ERR(file);
225 shp->shm_cprid = current->tgid;
227 shp->shm_atim = shp->shm_dtim = 0;
228 shp->shm_ctim = get_seconds();
229 shp->shm_segsz = size;
231 shp->id = shm_buildid(id,shp->shm_perm.seq);
232 shp->shm_file = file;
233 file->f_dentry->d_inode->i_ino = shp->id;
234 if (shmflg & SHM_HUGETLB)
235 set_file_hugepages(file);
237 file->f_op = &shm_file_operations;
245 security_shm_free(shp);
250 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
252 struct shmid_kernel *shp;
256 if (key == IPC_PRIVATE) {
257 err = newseg(key, shmflg, size);
258 } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
259 if (!(shmflg & IPC_CREAT))
262 err = newseg(key, shmflg, size);
263 } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
269 if (shp->shm_segsz < size)
271 else if (ipcperms(&shp->shm_perm, shmflg))
274 int shmid = shm_buildid(id, shp->shm_perm.seq);
275 err = security_shm_associate(shp, shmflg);
286 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
290 return copy_to_user(buf, in, sizeof(*in));
295 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
296 out.shm_segsz = in->shm_segsz;
297 out.shm_atime = in->shm_atime;
298 out.shm_dtime = in->shm_dtime;
299 out.shm_ctime = in->shm_ctime;
300 out.shm_cpid = in->shm_cpid;
301 out.shm_lpid = in->shm_lpid;
302 out.shm_nattch = in->shm_nattch;
304 return copy_to_user(buf, &out, sizeof(out));
317 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
322 struct shmid64_ds tbuf;
324 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
327 out->uid = tbuf.shm_perm.uid;
328 out->gid = tbuf.shm_perm.gid;
329 out->mode = tbuf.shm_flags;
335 struct shmid_ds tbuf_old;
337 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
340 out->uid = tbuf_old.shm_perm.uid;
341 out->gid = tbuf_old.shm_perm.gid;
342 out->mode = tbuf_old.shm_flags;
351 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
355 return copy_to_user(buf, in, sizeof(*in));
360 if(in->shmmax > INT_MAX)
361 out.shmmax = INT_MAX;
363 out.shmmax = (int)in->shmmax;
365 out.shmmin = in->shmmin;
366 out.shmmni = in->shmmni;
367 out.shmseg = in->shmseg;
368 out.shmall = in->shmall;
370 return copy_to_user(buf, &out, sizeof(out));
377 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
384 for (i = 0; i <= shm_ids.max_id; i++) {
385 struct shmid_kernel *shp;
392 inode = shp->shm_file->f_dentry->d_inode;
394 if (is_file_hugepages(shp->shm_file)) {
395 struct address_space *mapping = inode->i_mapping;
396 *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
398 struct shmem_inode_info *info = SHMEM_I(inode);
399 spin_lock(&info->lock);
400 *rss += inode->i_mapping->nrpages;
401 *swp += info->swapped;
402 spin_unlock(&info->lock);
407 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
409 struct shm_setbuf setbuf;
410 struct shmid_kernel *shp;
413 if (cmd < 0 || shmid < 0) {
418 version = ipc_parse_version(&cmd);
420 switch (cmd) { /* replace with proc interface ? */
423 struct shminfo64 shminfo;
425 err = security_shm_shmctl(NULL, cmd);
429 memset(&shminfo,0,sizeof(shminfo));
430 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
431 shminfo.shmmax = shm_ctlmax;
432 shminfo.shmall = shm_ctlall;
434 shminfo.shmmin = SHMMIN;
435 if(copy_shminfo_to_user (buf, &shminfo, version))
437 /* reading a integer is always atomic */
445 struct shm_info shm_info;
447 err = security_shm_shmctl(NULL, cmd);
451 memset(&shm_info,0,sizeof(shm_info));
453 shm_info.used_ids = shm_ids.in_use;
454 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
455 shm_info.shm_tot = shm_tot;
456 shm_info.swap_attempts = 0;
457 shm_info.swap_successes = 0;
458 err = shm_ids.max_id;
460 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
465 err = err < 0 ? 0 : err;
471 struct shmid64_ds tbuf;
473 memset(&tbuf, 0, sizeof(tbuf));
474 shp = shm_lock(shmid);
478 } else if(cmd==SHM_STAT) {
480 if (shmid > shm_ids.max_id)
482 result = shm_buildid(shmid, shp->shm_perm.seq);
484 err = shm_checkid(shp,shmid);
490 if (ipcperms (&shp->shm_perm, S_IRUGO))
492 err = security_shm_shmctl(shp, cmd);
495 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
496 tbuf.shm_segsz = shp->shm_segsz;
497 tbuf.shm_atime = shp->shm_atim;
498 tbuf.shm_dtime = shp->shm_dtim;
499 tbuf.shm_ctime = shp->shm_ctim;
500 tbuf.shm_cpid = shp->shm_cprid;
501 tbuf.shm_lpid = shp->shm_lprid;
502 if (!is_file_hugepages(shp->shm_file))
503 tbuf.shm_nattch = shp->shm_nattch;
505 tbuf.shm_nattch = file_count(shp->shm_file) - 1;
507 if(copy_shmid_to_user (buf, &tbuf, version))
516 /* Allow superuser to lock segment in memory */
517 if (!can_do_mlock() && cmd == SHM_LOCK) {
521 shp = shm_lock(shmid);
526 err = shm_checkid(shp,shmid);
530 err = security_shm_shmctl(shp, cmd);
535 struct user_struct * user = current->user;
536 if (!is_file_hugepages(shp->shm_file)) {
537 err = shmem_lock(shp->shm_file, 1, user);
539 shp->shm_flags |= SHM_LOCKED;
540 shp->mlock_user = user;
543 } else if (!is_file_hugepages(shp->shm_file)) {
544 shmem_lock(shp->shm_file, 0, shp->mlock_user);
545 shp->shm_flags &= ~SHM_LOCKED;
546 shp->mlock_user = NULL;
554 * We cannot simply remove the file. The SVID states
555 * that the block remains until the last person
556 * detaches from it, then is deleted. A shmat() on
557 * an RMID segment is legal in older Linux and if
558 * we change it apps break...
560 * Instead we set a destroyed flag, and then blow
561 * the name away when the usage hits zero.
564 shp = shm_lock(shmid);
568 err = shm_checkid(shp, shmid);
572 if (current->euid != shp->shm_perm.uid &&
573 current->euid != shp->shm_perm.cuid &&
574 !capable(CAP_SYS_ADMIN)) {
579 err = security_shm_shmctl(shp, cmd);
583 if (shp->shm_nattch){
584 shp->shm_flags |= SHM_DEST;
585 /* Do not find it any more */
586 shp->shm_perm.key = IPC_PRIVATE;
596 if (copy_shmid_from_user (&setbuf, buf, version)) {
601 shp = shm_lock(shmid);
605 err = shm_checkid(shp,shmid);
609 if (current->euid != shp->shm_perm.uid &&
610 current->euid != shp->shm_perm.cuid &&
611 !capable(CAP_SYS_ADMIN)) {
615 err = security_shm_shmctl(shp, cmd);
619 shp->shm_perm.uid = setbuf.uid;
620 shp->shm_perm.gid = setbuf.gid;
621 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
622 | (setbuf.mode & S_IRWXUGO);
623 shp->shm_ctim = get_seconds();
645 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
647 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
648 * "raddr" thing points to kernel space, and there has to be a wrapper around
651 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
653 struct shmid_kernel *shp;
660 unsigned long o_flags;
667 } else if ((addr = (ulong)shmaddr)) {
668 if (addr & (SHMLBA-1)) {
669 if (shmflg & SHM_RND)
670 addr &= ~(SHMLBA-1); /* round down */
672 #ifndef __ARCH_FORCE_SHMLBA
673 if (addr & ~PAGE_MASK)
677 flags = MAP_SHARED | MAP_FIXED;
679 if ((shmflg & SHM_REMAP))
685 if (shmflg & SHM_RDONLY) {
690 prot = PROT_READ | PROT_WRITE;
692 acc_mode = S_IRUGO | S_IWUGO;
694 if (shmflg & SHM_EXEC) {
700 * We cannot rely on the fs check since SYSV IPC does have an
701 * additional creator id...
703 shp = shm_lock(shmid);
708 err = shm_checkid(shp,shmid);
713 if (ipcperms(&shp->shm_perm, acc_mode)) {
719 err = security_shm_shmat(shp, shmaddr, shmflg);
725 file = shp->shm_file;
726 size = i_size_read(file->f_dentry->d_inode);
730 down_write(¤t->mm->mmap_sem);
731 if (addr && !(shmflg & SHM_REMAP)) {
732 user_addr = ERR_PTR(-EINVAL);
733 if (find_vma_intersection(current->mm, addr, addr + size))
736 * If shm segment goes below stack, make sure there is some
737 * space left for the stack to grow (at least 4 pages).
739 if (addr < current->mm->start_stack &&
740 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
744 user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
747 up_write(¤t->mm->mmap_sem);
750 if(!(shp = shm_lock(shmid)))
753 if(shp->shm_nattch == 0 &&
754 shp->shm_flags & SHM_DEST)
760 *raddr = (unsigned long) user_addr;
762 if (IS_ERR(user_addr))
763 err = PTR_ERR(user_addr);
769 * detach and kill segment if marked destroyed.
770 * The work is done in shm_close.
772 asmlinkage long sys_shmdt(char __user *shmaddr)
774 struct mm_struct *mm = current->mm;
775 struct vm_area_struct *vma, *next;
776 unsigned long addr = (unsigned long)shmaddr;
778 int retval = -EINVAL;
780 down_write(&mm->mmap_sem);
783 * This function tries to be smart and unmap shm segments that
784 * were modified by partial mlock or munmap calls:
785 * - It first determines the size of the shm segment that should be
786 * unmapped: It searches for a vma that is backed by shm and that
787 * started at address shmaddr. It records it's size and then unmaps
789 * - Then it unmaps all shm vmas that started at shmaddr and that
790 * are within the initially determined size.
791 * Errors from do_munmap are ignored: the function only fails if
792 * it's called with invalid parameters or if it's called to unmap
793 * a part of a vma. Both calls in this function are for full vmas,
794 * the parameters are directly copied from the vma itself and always
795 * valid - therefore do_munmap cannot fail. (famous last words?)
798 * If it had been mremap()'d, the starting address would not
799 * match the usual checks anyway. So assume all vma's are
800 * above the starting address given.
802 vma = find_vma(mm, addr);
808 * Check if the starting address would match, i.e. it's
809 * a fragment created by mprotect() and/or munmap(), or it
810 * otherwise it starts at this address with no hassles.
812 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
813 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
816 size = vma->vm_file->f_dentry->d_inode->i_size;
817 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
819 * We discovered the size of the shm segment, so
820 * break out of here and fall through to the next
821 * loop that uses the size information to stop
822 * searching for matching vma's.
832 * We need look no further than the maximum address a fragment
833 * could possibly have landed at. Also cast things to loff_t to
834 * prevent overflows and make comparisions vs. equal-width types.
836 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
839 /* finding a matching vma now does not alter retval */
840 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
841 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
843 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
847 up_write(&mm->mmap_sem);
851 #ifdef CONFIG_PROC_FS
852 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
859 len += sprintf(buffer, " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n");
861 for(i = 0; i <= shm_ids.max_id; i++) {
862 struct shmid_kernel* shp;
866 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
867 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
870 if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
874 if (sizeof(size_t) <= sizeof(int))
875 format = SMALL_STRING;
878 len += sprintf(buffer + len, format,
880 shm_buildid(i, shp->shm_perm.seq),
885 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
900 if(pos > offset + length)
907 *start = buffer + (offset - begin);
908 len -= (offset - begin);