fs/proc/kcore.c: Add bounce buffer for ktext data
[cascardo/linux.git] / fs / proc / kcore.c
index a939f5e..5c89a07 100644 (file)
@@ -430,6 +430,7 @@ static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
 static ssize_t
 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
 {
+       char *buf = file->private_data;
        ssize_t acc = 0;
        size_t size, tsz;
        size_t elf_buflen;
@@ -500,23 +501,20 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
                        if (clear_user(buffer, tsz))
                                return -EFAULT;
                } else if (is_vmalloc_or_module_addr((void *)start)) {
-                       char * elf_buf;
-
-                       elf_buf = kzalloc(tsz, GFP_KERNEL);
-                       if (!elf_buf)
-                               return -ENOMEM;
-                       vread(elf_buf, (char *)start, tsz);
+                       vread(buf, (char *)start, tsz);
                        /* we have to zero-fill user buffer even if no read */
-                       if (copy_to_user(buffer, elf_buf, tsz)) {
-                               kfree(elf_buf);
+                       if (copy_to_user(buffer, buf, tsz))
                                return -EFAULT;
-                       }
-                       kfree(elf_buf);
                } else {
                        if (kern_addr_valid(start)) {
                                unsigned long n;
 
-                               n = copy_to_user(buffer, (char *)start, tsz);
+                               /*
+                                * Using bounce buffer to bypass the
+                                * hardened user copy kernel text checks.
+                                */
+                               memcpy(buf, (char *) start, tsz);
+                               n = copy_to_user(buffer, buf, tsz);
                                /*
                                 * We cannot distinguish between fault on source
                                 * and fault on destination. When this happens
@@ -549,6 +547,11 @@ static int open_kcore(struct inode *inode, struct file *filp)
 {
        if (!capable(CAP_SYS_RAWIO))
                return -EPERM;
+
+       filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
+       if (!filp->private_data)
+               return -ENOMEM;
+
        if (kcore_need_update)
                kcore_update_ram();
        if (i_size_read(inode) != proc_root_kcore->size) {
@@ -559,10 +562,16 @@ static int open_kcore(struct inode *inode, struct file *filp)
        return 0;
 }
 
+static int release_kcore(struct inode *inode, struct file *file)
+{
+       kfree(file->private_data);
+       return 0;
+}
 
 static const struct file_operations proc_kcore_operations = {
        .read           = read_kcore,
        .open           = open_kcore,
+       .release        = release_kcore,
        .llseek         = default_llseek,
 };