1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright IBM Corporation, 2021
5 * Author: Mike Rapoport <rppt@linux.ibm.com>
10 #include <linux/swap.h>
11 #include <linux/mount.h>
12 #include <linux/memfd.h>
13 #include <linux/bitops.h>
14 #include <linux/printk.h>
15 #include <linux/pagemap.h>
16 #include <linux/syscalls.h>
17 #include <linux/pseudo_fs.h>
18 #include <linux/secretmem.h>
19 #include <linux/set_memory.h>
20 #include <linux/sched/signal.h>
22 #include <uapi/linux/magic.h>
24 #include <asm/tlbflush.h>
29 #define pr_fmt(fmt) "secretmem: " fmt
32 * Define mode and flag masks to allow validation of the system call
35 #define SECRETMEM_MODE_MASK (0x0)
36 #define SECRETMEM_FLAGS_MASK SECRETMEM_MODE_MASK
38 static bool secretmem_enable __ro_after_init;
39 module_param_named(enable, secretmem_enable, bool, 0400);
40 MODULE_PARM_DESC(secretmem_enable,
41 "Enable secretmem and memfd_secret(2) system call");
43 static atomic_t secretmem_users;
45 bool secretmem_active(void)
47 return !!atomic_read(&secretmem_users);
50 static vm_fault_t secretmem_fault(struct vm_fault *vmf)
52 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
53 struct inode *inode = file_inode(vmf->vma->vm_file);
54 pgoff_t offset = vmf->pgoff;
55 gfp_t gfp = vmf->gfp_mask;
60 if (((loff_t)vmf->pgoff << PAGE_SHIFT) >= i_size_read(inode))
61 return vmf_error(-EINVAL);
64 page = find_lock_page(mapping, offset);
66 page = alloc_page(gfp | __GFP_ZERO);
70 err = set_direct_map_invalid_noflush(page);
73 return vmf_error(err);
76 __SetPageUptodate(page);
77 err = add_to_page_cache_lru(page, mapping, offset, gfp);
81 * If a split of large page was required, it
82 * already happened when we marked the page invalid
83 * which guarantees that this call won't fail
85 set_direct_map_default_noflush(page);
89 return vmf_error(err);
92 addr = (unsigned long)page_address(page);
93 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
97 return VM_FAULT_LOCKED;
100 static const struct vm_operations_struct secretmem_vm_ops = {
101 .fault = secretmem_fault,
104 static int secretmem_release(struct inode *inode, struct file *file)
106 atomic_dec(&secretmem_users);
110 static int secretmem_mmap(struct file *file, struct vm_area_struct *vma)
112 unsigned long len = vma->vm_end - vma->vm_start;
114 if ((vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) == 0)
117 if (mlock_future_check(vma->vm_mm, vma->vm_flags | VM_LOCKED, len))
120 vma->vm_flags |= VM_LOCKED | VM_DONTDUMP;
121 vma->vm_ops = &secretmem_vm_ops;
126 bool vma_is_secretmem(struct vm_area_struct *vma)
128 return vma->vm_ops == &secretmem_vm_ops;
131 static const struct file_operations secretmem_fops = {
132 .release = secretmem_release,
133 .mmap = secretmem_mmap,
136 static bool secretmem_isolate_page(struct page *page, isolate_mode_t mode)
141 static int secretmem_migratepage(struct address_space *mapping,
142 struct page *newpage, struct page *page,
143 enum migrate_mode mode)
148 static void secretmem_free_folio(struct folio *folio)
150 set_direct_map_default_noflush(&folio->page);
151 folio_zero_segment(folio, 0, folio_size(folio));
154 const struct address_space_operations secretmem_aops = {
155 .dirty_folio = noop_dirty_folio,
156 .free_folio = secretmem_free_folio,
157 .migratepage = secretmem_migratepage,
158 .isolate_page = secretmem_isolate_page,
161 static int secretmem_setattr(struct user_namespace *mnt_userns,
162 struct dentry *dentry, struct iattr *iattr)
164 struct inode *inode = d_inode(dentry);
165 unsigned int ia_valid = iattr->ia_valid;
167 if ((ia_valid & ATTR_SIZE) && inode->i_size)
170 return simple_setattr(mnt_userns, dentry, iattr);
173 static const struct inode_operations secretmem_iops = {
174 .setattr = secretmem_setattr,
177 static struct vfsmount *secretmem_mnt;
179 static struct file *secretmem_file_create(unsigned long flags)
181 struct file *file = ERR_PTR(-ENOMEM);
183 const char *anon_name = "[secretmem]";
184 const struct qstr qname = QSTR_INIT(anon_name, strlen(anon_name));
187 inode = alloc_anon_inode(secretmem_mnt->mnt_sb);
189 return ERR_CAST(inode);
191 err = security_inode_init_security_anon(inode, &qname, NULL);
197 file = alloc_file_pseudo(inode, secretmem_mnt, "secretmem",
198 O_RDWR, &secretmem_fops);
202 mapping_set_gfp_mask(inode->i_mapping, GFP_HIGHUSER);
203 mapping_set_unevictable(inode->i_mapping);
205 inode->i_op = &secretmem_iops;
206 inode->i_mapping->a_ops = &secretmem_aops;
208 /* pretend we are a normal file with zero size */
209 inode->i_mode |= S_IFREG;
219 SYSCALL_DEFINE1(memfd_secret, unsigned int, flags)
224 /* make sure local flags do not confict with global fcntl.h */
225 BUILD_BUG_ON(SECRETMEM_FLAGS_MASK & O_CLOEXEC);
227 if (!secretmem_enable)
230 if (flags & ~(SECRETMEM_FLAGS_MASK | O_CLOEXEC))
232 if (atomic_read(&secretmem_users) < 0)
235 fd = get_unused_fd_flags(flags & O_CLOEXEC);
239 file = secretmem_file_create(flags);
245 file->f_flags |= O_LARGEFILE;
247 atomic_inc(&secretmem_users);
248 fd_install(fd, file);
256 static int secretmem_init_fs_context(struct fs_context *fc)
258 return init_pseudo(fc, SECRETMEM_MAGIC) ? 0 : -ENOMEM;
261 static struct file_system_type secretmem_fs = {
263 .init_fs_context = secretmem_init_fs_context,
264 .kill_sb = kill_anon_super,
267 static int secretmem_init(void)
271 if (!secretmem_enable)
274 secretmem_mnt = kern_mount(&secretmem_fs);
275 if (IS_ERR(secretmem_mnt))
276 ret = PTR_ERR(secretmem_mnt);
278 /* prevent secretmem mappings from ever getting PROT_EXEC */
279 secretmem_mnt->mnt_flags |= MNT_NOEXEC;
283 fs_initcall(secretmem_init);