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/shmem_fs.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/ptrace.h>
32 #include <linux/seq_file.h>
33 #include <linux/vs_context.h>
34 #include <linux/vs_limit.h>
36 #include <asm/uaccess.h>
40 static struct file_operations shm_file_operations;
41 static struct vm_operations_struct shm_vm_ops;
43 static struct ipc_ids shm_ids;
45 #define shm_lock(id) ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
46 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
47 #define shm_get(id) ((struct shmid_kernel*)ipc_get(&shm_ids,id))
48 #define shm_buildid(id, seq) \
49 ipc_buildid(&shm_ids, id, seq)
51 static int newseg (key_t key, int shmflg, size_t size);
52 static void shm_open (struct vm_area_struct *shmd);
53 static void shm_close (struct vm_area_struct *shmd);
55 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
58 size_t shm_ctlmax = SHMMAX;
59 size_t shm_ctlall = SHMALL;
60 int shm_ctlmni = SHMMNI;
62 static int shm_tot; /* total number of shared memory pages */
64 void __init shm_init (void)
66 ipc_init_ids(&shm_ids, 1);
67 ipc_init_proc_interface("sysvipc/shm",
68 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
70 sysvipc_shm_proc_show);
73 static inline int shm_checkid(struct shmid_kernel *s, int id)
75 if (ipc_checkid(&shm_ids,&s->shm_perm,id))
80 static inline struct shmid_kernel *shm_rmid(int id)
82 return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
85 static inline int shm_addid(struct shmid_kernel *shp)
87 return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
92 static inline void shm_inc (int id) {
93 struct shmid_kernel *shp;
95 if(!(shp = shm_lock(id)))
97 shp->shm_atim = get_seconds();
98 shp->shm_lprid = current->tgid;
103 /* This is called by fork, once for every shm attach. */
104 static void shm_open (struct vm_area_struct *shmd)
106 shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
110 * shm_destroy - free the struct shmid_kernel
112 * @shp: struct to free
114 * It has to be called with shp and shm_ids.sem locked,
115 * but returns with shp unlocked and freed.
117 static void shm_destroy (struct shmid_kernel *shp)
119 struct vx_info *vxi = lookup_vx_info(shp->shm_perm.xid);
120 int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
122 vx_ipcshm_sub(vxi, shp, numpages);
127 if (!is_file_hugepages(shp->shm_file))
128 shmem_lock(shp->shm_file, 0, shp->mlock_user);
130 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
132 fput (shp->shm_file);
133 security_shm_free(shp);
139 * remove the attach descriptor shmd.
140 * free memory for segment if it is marked destroyed.
141 * The descriptor has already been removed from the current->mm->mmap list
142 * and will later be kfree()d.
144 static void shm_close (struct vm_area_struct *shmd)
146 struct file * file = shmd->vm_file;
147 int id = file->f_dentry->d_inode->i_ino;
148 struct shmid_kernel *shp;
151 /* remove from the list of attaches of the shm segment */
152 if(!(shp = shm_lock(id)))
154 shp->shm_lprid = current->tgid;
155 shp->shm_dtim = get_seconds();
157 if(shp->shm_nattch == 0 &&
158 shp->shm_perm.mode & SHM_DEST)
165 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
169 ret = shmem_mmap(file, vma);
171 vma->vm_ops = &shm_vm_ops;
172 if (!(vma->vm_flags & VM_WRITE))
173 vma->vm_flags &= ~VM_MAYWRITE;
174 shm_inc(file->f_dentry->d_inode->i_ino);
180 static struct file_operations shm_file_operations = {
183 .get_unmapped_area = shmem_get_unmapped_area,
187 static struct vm_operations_struct shm_vm_ops = {
188 .open = shm_open, /* callback for a new vm-area open */
189 .close = shm_close, /* callback for when the vm-area is released */
190 .nopage = shmem_nopage,
191 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
192 .set_policy = shmem_set_policy,
193 .get_policy = shmem_get_policy,
197 static int newseg (key_t key, int shmflg, size_t size)
200 struct shmid_kernel *shp;
201 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
206 if (size < SHMMIN || size > shm_ctlmax)
209 if (shm_tot + numpages >= shm_ctlall)
211 if (!vx_ipcshm_avail(current->vx_info, numpages))
214 shp = ipc_rcu_alloc(sizeof(*shp));
218 shp->shm_perm.key = key;
219 shp->shm_perm.xid = vx_current_xid();
220 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
221 shp->mlock_user = NULL;
223 shp->shm_perm.security = NULL;
224 error = security_shm_alloc(shp);
230 if (shmflg & SHM_HUGETLB) {
231 /* hugetlb_zero_setup takes care of mlock user accounting */
232 file = hugetlb_zero_setup(size);
233 shp->mlock_user = current->user;
235 int acctflag = VM_ACCOUNT;
237 * Do not allow no accounting for OVERCOMMIT_NEVER, even
240 if ((shmflg & SHM_NORESERVE) &&
241 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
243 sprintf (name, "SYSV%08x", key);
244 file = shmem_file_setup(name, size, acctflag);
246 error = PTR_ERR(file);
255 shp->shm_cprid = current->tgid;
257 shp->shm_atim = shp->shm_dtim = 0;
258 shp->shm_ctim = get_seconds();
259 shp->shm_segsz = size;
261 shp->id = shm_buildid(id,shp->shm_perm.seq);
262 shp->shm_file = file;
263 file->f_dentry->d_inode->i_ino = shp->id;
265 /* Hugetlb ops would have already been assigned. */
266 if (!(shmflg & SHM_HUGETLB))
267 file->f_op = &shm_file_operations;
270 vx_ipcshm_add(current->vx_info, key, numpages);
277 security_shm_free(shp);
282 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
284 struct shmid_kernel *shp;
288 if (key == IPC_PRIVATE) {
289 err = newseg(key, shmflg, size);
290 } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
291 if (!(shmflg & IPC_CREAT))
294 err = newseg(key, shmflg, size);
295 } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
301 if (shp->shm_segsz < size)
303 else if (ipcperms(&shp->shm_perm, shmflg))
306 int shmid = shm_buildid(id, shp->shm_perm.seq);
307 err = security_shm_associate(shp, shmflg);
318 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
322 return copy_to_user(buf, in, sizeof(*in));
327 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
328 out.shm_segsz = in->shm_segsz;
329 out.shm_atime = in->shm_atime;
330 out.shm_dtime = in->shm_dtime;
331 out.shm_ctime = in->shm_ctime;
332 out.shm_cpid = in->shm_cpid;
333 out.shm_lpid = in->shm_lpid;
334 out.shm_nattch = in->shm_nattch;
336 return copy_to_user(buf, &out, sizeof(out));
349 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
354 struct shmid64_ds tbuf;
356 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
359 out->uid = tbuf.shm_perm.uid;
360 out->gid = tbuf.shm_perm.gid;
361 out->mode = tbuf.shm_perm.mode;
367 struct shmid_ds tbuf_old;
369 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
372 out->uid = tbuf_old.shm_perm.uid;
373 out->gid = tbuf_old.shm_perm.gid;
374 out->mode = tbuf_old.shm_perm.mode;
383 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
387 return copy_to_user(buf, in, sizeof(*in));
392 if(in->shmmax > INT_MAX)
393 out.shmmax = INT_MAX;
395 out.shmmax = (int)in->shmmax;
397 out.shmmin = in->shmmin;
398 out.shmmni = in->shmmni;
399 out.shmseg = in->shmseg;
400 out.shmall = in->shmall;
402 return copy_to_user(buf, &out, sizeof(out));
409 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
416 for (i = 0; i <= shm_ids.max_id; i++) {
417 struct shmid_kernel *shp;
424 inode = shp->shm_file->f_dentry->d_inode;
426 if (is_file_hugepages(shp->shm_file)) {
427 struct address_space *mapping = inode->i_mapping;
428 *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
430 struct shmem_inode_info *info = SHMEM_I(inode);
431 spin_lock(&info->lock);
432 *rss += inode->i_mapping->nrpages;
433 *swp += info->swapped;
434 spin_unlock(&info->lock);
439 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
441 struct shm_setbuf setbuf;
442 struct shmid_kernel *shp;
445 if (cmd < 0 || shmid < 0) {
450 version = ipc_parse_version(&cmd);
452 switch (cmd) { /* replace with proc interface ? */
455 struct shminfo64 shminfo;
457 err = security_shm_shmctl(NULL, cmd);
461 memset(&shminfo,0,sizeof(shminfo));
462 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
463 shminfo.shmmax = shm_ctlmax;
464 shminfo.shmall = shm_ctlall;
466 shminfo.shmmin = SHMMIN;
467 if(copy_shminfo_to_user (buf, &shminfo, version))
469 /* reading a integer is always atomic */
477 struct shm_info shm_info;
479 err = security_shm_shmctl(NULL, cmd);
483 memset(&shm_info,0,sizeof(shm_info));
485 shm_info.used_ids = shm_ids.in_use;
486 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
487 shm_info.shm_tot = shm_tot;
488 shm_info.swap_attempts = 0;
489 shm_info.swap_successes = 0;
490 err = shm_ids.max_id;
492 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
497 err = err < 0 ? 0 : err;
503 struct shmid64_ds tbuf;
505 memset(&tbuf, 0, sizeof(tbuf));
506 shp = shm_lock(shmid);
510 } else if(cmd==SHM_STAT) {
512 if (shmid > shm_ids.max_id)
514 result = shm_buildid(shmid, shp->shm_perm.seq);
516 err = shm_checkid(shp,shmid);
522 if (ipcperms (&shp->shm_perm, S_IRUGO))
524 err = security_shm_shmctl(shp, cmd);
527 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
528 tbuf.shm_segsz = shp->shm_segsz;
529 tbuf.shm_atime = shp->shm_atim;
530 tbuf.shm_dtime = shp->shm_dtim;
531 tbuf.shm_ctime = shp->shm_ctim;
532 tbuf.shm_cpid = shp->shm_cprid;
533 tbuf.shm_lpid = shp->shm_lprid;
534 if (!is_file_hugepages(shp->shm_file))
535 tbuf.shm_nattch = shp->shm_nattch;
537 tbuf.shm_nattch = file_count(shp->shm_file) - 1;
539 if(copy_shmid_to_user (buf, &tbuf, version))
548 shp = shm_lock(shmid);
553 err = shm_checkid(shp,shmid);
557 if (!capable(CAP_IPC_LOCK)) {
559 if (current->euid != shp->shm_perm.uid &&
560 current->euid != shp->shm_perm.cuid)
562 if (cmd == SHM_LOCK &&
563 !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
567 err = security_shm_shmctl(shp, cmd);
572 struct user_struct * user = current->user;
573 if (!is_file_hugepages(shp->shm_file)) {
574 err = shmem_lock(shp->shm_file, 1, user);
576 shp->shm_perm.mode |= SHM_LOCKED;
577 shp->mlock_user = user;
580 } else if (!is_file_hugepages(shp->shm_file)) {
581 shmem_lock(shp->shm_file, 0, shp->mlock_user);
582 shp->shm_perm.mode &= ~SHM_LOCKED;
583 shp->mlock_user = NULL;
591 * We cannot simply remove the file. The SVID states
592 * that the block remains until the last person
593 * detaches from it, then is deleted. A shmat() on
594 * an RMID segment is legal in older Linux and if
595 * we change it apps break...
597 * Instead we set a destroyed flag, and then blow
598 * the name away when the usage hits zero.
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)) {
616 err = security_shm_shmctl(shp, cmd);
620 if (shp->shm_nattch){
621 shp->shm_perm.mode |= SHM_DEST;
622 /* Do not find it any more */
623 shp->shm_perm.key = IPC_PRIVATE;
633 if (copy_shmid_from_user (&setbuf, buf, version)) {
637 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
640 shp = shm_lock(shmid);
644 err = shm_checkid(shp,shmid);
648 if (current->euid != shp->shm_perm.uid &&
649 current->euid != shp->shm_perm.cuid &&
650 !capable(CAP_SYS_ADMIN)) {
654 err = security_shm_shmctl(shp, cmd);
658 shp->shm_perm.uid = setbuf.uid;
659 shp->shm_perm.gid = setbuf.gid;
660 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
661 | (setbuf.mode & S_IRWXUGO);
662 shp->shm_ctim = get_seconds();
684 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
686 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
687 * "raddr" thing points to kernel space, and there has to be a wrapper around
690 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
692 struct shmid_kernel *shp;
699 unsigned long o_flags;
706 } else if ((addr = (ulong)shmaddr)) {
707 if (addr & (SHMLBA-1)) {
708 if (shmflg & SHM_RND)
709 addr &= ~(SHMLBA-1); /* round down */
711 #ifndef __ARCH_FORCE_SHMLBA
712 if (addr & ~PAGE_MASK)
716 flags = MAP_SHARED | MAP_FIXED;
718 if ((shmflg & SHM_REMAP))
724 if (shmflg & SHM_RDONLY) {
729 prot = PROT_READ | PROT_WRITE;
731 acc_mode = S_IRUGO | S_IWUGO;
733 if (shmflg & SHM_EXEC) {
739 * We cannot rely on the fs check since SYSV IPC does have an
740 * additional creator id...
742 shp = shm_lock(shmid);
747 err = shm_checkid(shp,shmid);
752 if (ipcperms(&shp->shm_perm, acc_mode)) {
758 err = security_shm_shmat(shp, shmaddr, shmflg);
764 file = shp->shm_file;
765 size = i_size_read(file->f_dentry->d_inode);
769 down_write(¤t->mm->mmap_sem);
770 if (addr && !(shmflg & SHM_REMAP)) {
771 user_addr = ERR_PTR(-EINVAL);
772 if (find_vma_intersection(current->mm, addr, addr + size))
775 * If shm segment goes below stack, make sure there is some
776 * space left for the stack to grow (at least 4 pages).
778 if (addr < current->mm->start_stack &&
779 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
783 user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
786 up_write(¤t->mm->mmap_sem);
789 if(!(shp = shm_lock(shmid)))
792 if(shp->shm_nattch == 0 &&
793 shp->shm_perm.mode & SHM_DEST)
799 *raddr = (unsigned long) user_addr;
801 if (IS_ERR(user_addr))
802 err = PTR_ERR(user_addr);
807 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
812 err = do_shmat(shmid, shmaddr, shmflg, &ret);
815 force_successful_syscall_return();
820 * detach and kill segment if marked destroyed.
821 * The work is done in shm_close.
823 asmlinkage long sys_shmdt(char __user *shmaddr)
825 struct mm_struct *mm = current->mm;
826 struct vm_area_struct *vma, *next;
827 unsigned long addr = (unsigned long)shmaddr;
829 int retval = -EINVAL;
831 down_write(&mm->mmap_sem);
834 * This function tries to be smart and unmap shm segments that
835 * were modified by partial mlock or munmap calls:
836 * - It first determines the size of the shm segment that should be
837 * unmapped: It searches for a vma that is backed by shm and that
838 * started at address shmaddr. It records it's size and then unmaps
840 * - Then it unmaps all shm vmas that started at shmaddr and that
841 * are within the initially determined size.
842 * Errors from do_munmap are ignored: the function only fails if
843 * it's called with invalid parameters or if it's called to unmap
844 * a part of a vma. Both calls in this function are for full vmas,
845 * the parameters are directly copied from the vma itself and always
846 * valid - therefore do_munmap cannot fail. (famous last words?)
849 * If it had been mremap()'d, the starting address would not
850 * match the usual checks anyway. So assume all vma's are
851 * above the starting address given.
853 vma = find_vma(mm, addr);
859 * Check if the starting address would match, i.e. it's
860 * a fragment created by mprotect() and/or munmap(), or it
861 * otherwise it starts at this address with no hassles.
863 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
864 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
867 size = vma->vm_file->f_dentry->d_inode->i_size;
868 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
870 * We discovered the size of the shm segment, so
871 * break out of here and fall through to the next
872 * loop that uses the size information to stop
873 * searching for matching vma's.
883 * We need look no further than the maximum address a fragment
884 * could possibly have landed at. Also cast things to loff_t to
885 * prevent overflows and make comparisions vs. equal-width types.
887 size = PAGE_ALIGN(size);
888 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
891 /* finding a matching vma now does not alter retval */
892 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
893 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
895 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
899 up_write(&mm->mmap_sem);
903 #ifdef CONFIG_PROC_FS
904 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
906 struct shmid_kernel *shp = it;
909 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
910 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
912 if (!vx_check(shp->shm_perm.xid, VX_IDENT))
915 if (sizeof(size_t) <= sizeof(int))
916 format = SMALL_STRING;
919 return seq_printf(s, format,
926 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,