Merge branch 'no_kernel_modifications'

Author: JohnMalliotakis

README.md

@@ -4,7 +4,7 @@ This library provides an optimized memory-mapped I/O path inside the Linux kerne
 
 ## Prerequisites
 
-FastMap runs on a modified version of Linux 4.14. To build and run this version of FastMap you can clone this kernel with:
+The main branch of FastMap runs on a modified version of Linux 4.14. To build and run this version of FastMap you can clone this kernel with:
 ```bash
 git clone https://github.com/tpapagian/linux-4.14.72-spf.git -b fastmap
 ```
@@ -13,6 +13,9 @@ The modifications are minimal as they contain only some exported symbols and the
 
 After that, you should configure, build, and install this kernel using specific instructions based on your setup.
 
+The no kernel modifications branch consists of a patchset so that FastMap can run on any Linux 4.14 kernel without any modifications being necessary. It should work properly,
+however it has not been tested as thoroughly as the main branch.
+
 ## Building FastMap
 
 In order to build FastMap simply run ```make``` on the top-level directory. This will build the driver (located in the driver directory) and the associated userspace tools (located in the ioctl directory). After that, it will also install the FastMap module and run ```depmod```.

driver/Kbuild

@@ -21,4 +21,5 @@ dmap-objs := 	main.o \
 							dmapfs_mmap.o \
 							dmapfs_super.o \
 							lpfifo_non_atomic.o \
-							dfifo_non_atomic.o
+							dfifo_non_atomic.o \
+							kernel_syms.o 

driver/dmap.h

@@ -161,6 +161,17 @@ struct pr_vma_data
 	unsigned long magic2; // -----
 };
 
+/*
+ * Struct used to encapsulate
+ * fastmap structs required in vm_area_struct,
+ * stored in vm_private_data member
+ */
+struct fastmap_info {
+	bool is_fastmap;
+	struct pr_vma_data *pvd;
+	struct pr_vma_entry *pve;
+};
+
 struct raw_device_data
 {
 	int inuse;

driver/dmapfs_file.c

@@ -237,7 +237,7 @@ static ssize_t wrapfs_read(struct file *file, char __user *buf, size_t count, lo
 
 	lower_file = wrapfs_lower_file(file);
 
-	pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 
 	if(
 		(pvd != NULL) 
@@ -313,7 +313,7 @@ static ssize_t wrapfs_write(struct file *file, const char __user *buf, size_t co
 	unsigned int radix_tree_id;
 #endif
 
-	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 
 	if(
 		(pvd != NULL) 
@@ -376,7 +376,7 @@ static long dmap_priority_ioctl(struct file *file, unsigned int cmd, unsigned lo
 {
 #if 0
 	int retval = 0;
-	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 	struct dmap_page_prio __user *__dpp_p;
 	struct dmap_page_prio __dpp_d;
 
@@ -487,6 +487,12 @@ static int wrapfs_open(struct inode *inode, struct file *file)
 		goto out_err;
 	}
 
+	((struct wrapfs_file_info *)file->private_data)->fmap_info = kzalloc(sizeof(struct fastmap_info), GFP_KERNEL);
+	if(!((struct wrapfs_file_info *)file->private_data)->fmap_info){
+		err = -ENOMEM;
+		goto out_err;
+	}
+
 	/* open lower object and link wrapfs's file struct to lower's */
 	wrapfs_get_lower_path(file->f_path.dentry, &lower_path);
 	lower_file = dentry_open(&lower_path, file->f_flags, current_cred());
@@ -527,7 +533,7 @@ static int wrapfs_open(struct inode *inode, struct file *file)
 
 		// set it to the private data of the file
 		DMAP_BGON( (pvd->magic1 != PVD_MAGIC_1) || (pvd->magic2 != PVD_MAGIC_2) );
-		((struct wrapfs_file_info*)file->private_data)->pvd = pvd;
+		((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd = pvd;
 	}
 
 	if (IS_ERR(lower_file)) {
@@ -588,11 +594,11 @@ static int wrapfs_file_release(struct inode *inode, struct file *file)
 
 	lower_file = wrapfs_lower_file(file);
 
-	pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 	if(pvd != NULL && pvd->is_mmap == true && pvd->is_valid == true && atomic64_read(&pvd->mmaped) == 1){
 
 		if(lower_file)
-			((struct wrapfs_file_info*)file->private_data)->pvd->bk.filp = lower_file;
+			((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd->bk.filp = lower_file;
 		else
 			DMAP_BGON(1);
 
@@ -613,12 +619,12 @@ static int wrapfs_file_release(struct inode *inode, struct file *file)
 
 		for(i = 0; i < MAX_OPEN_FDS; ++i){
 			if(pvd->open_fds[i] != NULL && pvd->open_fds[i]->f_inode != NULL){
-				((struct wrapfs_file_info*)file->private_data)->pvd->bk.filp = pvd->open_fds[i];
+				((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd->bk.filp = pvd->open_fds[i];
 				break;
 			}
 		}
 
-		((struct wrapfs_file_info*)file->private_data)->pvd = NULL;
+		((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd = NULL;
 	}else if(pvd != NULL){
 		for(i = 0; i < MAX_OPEN_FDS; ++i)
 			if(pvd->open_fds[i] == lower_file){
@@ -644,7 +650,7 @@ static int wrapfs_fsync(struct file *file, loff_t start, loff_t end, int datasyn
 	struct dentry *dentry = file->f_path.dentry;
 	struct pr_vma_data *pvd;
 
-	pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 	if(pvd == NULL)
 		goto mmap_fsync;
 
@@ -712,7 +718,7 @@ wrapfs_read_iter(struct kiocb *iocb, struct iov_iter *iter)
 	struct file *file = iocb->ki_filp, *lower_file;
 
 #ifdef DEBUG_DIRECT_RW
-	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 	if(pvd != NULL){
 		DMAP_BGON( (pvd->magic1 != PVD_MAGIC_1) || (pvd->magic2 != PVD_MAGIC_2) );
 		WARN(atomic64_read(&pvd->cnt) > 0, "[%s:%s:%d][%ld]\n", __FILE__, __func__, __LINE__, atomic64_read(&pvd->cnt));
@@ -750,7 +756,7 @@ wrapfs_write_iter(struct kiocb *iocb, struct iov_iter *iter)
 	struct file *file = iocb->ki_filp, *lower_file;
 
 #ifdef DEBUG_DIRECT_RW
-	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->pvd;
+	struct pr_vma_data *pvd = ((struct wrapfs_file_info*)file->private_data)->fmap_info->pvd;
 	if(pvd != NULL){
 		DMAP_BGON( (pvd->magic1 != PVD_MAGIC_1) || (pvd->magic2 != PVD_MAGIC_2) );
 		WARN(atomic64_read(&pvd->cnt) > 0, "[%s:%s:%d][%ld]\n", __FILE__, __func__, __LINE__, atomic64_read(&pvd->cnt));

driver/helpers.c

@@ -39,6 +39,9 @@ void zap_page_from_address_space(struct vm_area_struct *vma, unsigned long addre
 
 	DMAP_BGON(vma->vm_mm == NULL);
 
+	pvmw.pte = NULL;
+	pvmw.ptl = NULL;
+
 	/* If the pte returned is valid, the lock will also be held. */
 	while(page_vma_mapped_walk(&pvmw)){
 		if(pvmw.pte){

driver/kernel_syms.c

@@ -0,0 +1,83 @@
+#include <linux/kallsyms.h>
+#include <linux/fs.h>
+#include <linux/mm_types.h>
+#include <linux/types.h>
+#include <linux/compiler_types.h>
+#include <linux/rmap.h>
+
+/* Auxiliary function pointers here */
+
+void (*flush_tlb_mm_range_ksym)(struct mm_struct *mm, unsigned long start,
+ 	unsigned long end, unsigned long vmflag);
+
+ssize_t (*vfs_read_ksym)(struct file *file, char __user *buf, 
+ 	size_t count, loff_t *pos);
+
+ssize_t (*vfs_write_ksym)(struct file *file, const char __user *buf, 
+ 	size_t count, loff_t *pos);
+
+void (*wake_up_page_bit_ksym)(struct page *page, int bit_nr);
+
+bool (*page_vma_mapped_walk_ksym)(struct page_vma_mapped_walk *pvmw);
+
+int acquire_ksyms(void)
+{
+	/* 
+	 * Try to find all necessary symbols,
+	 * return -1 if any lookup fails
+	 */
+	flush_tlb_mm_range_ksym = (void *)kallsyms_lookup_name("flush_tlb_mm_range");
+	if(!flush_tlb_mm_range_ksym)
+		return -1;
+
+	vfs_read_ksym = (void *)kallsyms_lookup_name("vfs_read");
+	if(!vfs_read_ksym)
+		return -1;
+
+	vfs_write_ksym = (void *)kallsyms_lookup_name("vfs_write");
+	if(!vfs_write_ksym)
+		return -1;
+
+	wake_up_page_bit_ksym = (void *)kallsyms_lookup_name("wake_up_page_bit");
+	if(!wake_up_page_bit_ksym)
+		return -1;
+
+	page_vma_mapped_walk_ksym = (void *)kallsyms_lookup_name("page_vma_mapped_walk");
+	if(!page_vma_mapped_walk_ksym)
+		return -1;
+
+	return 0;
+}
+
+/* 
+ * Provide the unexported kernel symbols
+ * by calling the appropriate initialized
+ * function pointer
+ */
+void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
+ 	unsigned long end, unsigned long vmflag)
+{
+	flush_tlb_mm_range_ksym(mm, start, end, vmflag);
+}
+
+ssize_t vfs_read(struct file *file, char __user *buf, 
+ 	size_t count, loff_t *pos)
+{
+	return vfs_read_ksym(file, buf, count, pos);
+}
+
+ssize_t vfs_write(struct file *file, const char __user *buf, 
+ 	size_t count, loff_t *pos)
+{
+	return vfs_write_ksym(file, buf, count, pos);
+}
+
+void wake_up_page_bit(struct page *page, int bit_nr)
+{
+	wake_up_page_bit_ksym(page, bit_nr);
+}
+
+bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
+{
+	return page_vma_mapped_walk_ksym(pvmw);
+}

driver/main.c

@@ -102,6 +102,7 @@ extern void write_bio_tagged_page(struct tagged_page *evicted_page);
 extern void write_dirty_pages(struct pr_vma_data *pvd, void (*__page_cleanup)(struct tagged_page *));
 
 extern void ino_cache_init(void);
+extern int acquire_ksyms(void);
 
 #ifdef USE_HUGEPAGES
 struct page *hugepages[16];
@@ -183,6 +184,7 @@ static int raw_open(struct inode *inode, struct file *filp)
 	const int minor = iminor(inode);
 	struct block_device *bdev;
 	struct pr_vma_data *pvd;
+	struct fastmap_info *fmap_info;
 	int cpu, err;
 
 	if (minor == 0)
@@ -216,6 +218,15 @@ static int raw_open(struct inode *inode, struct file *filp)
 		file_inode(filp)->i_mapping = bdev->bd_inode->i_mapping;
 
 	//filp->private_data = &raw_devices[minor];
+	
+	/* Allocate fastmap_info struct first */
+	fmap_info = kzalloc(sizeof(struct fastmap_info), GFP_KERNEL);
+	DMAP_BGON(fmap_info == NULL);
+	/*
+	 * Since kzalloc is used all fmap_info members are initialized
+	 * to 0, therefore there is no need to set is_fastmap and pve
+	 * to false and NULL respectively
+	 */
 	pvd = kzalloc(sizeof(struct pr_vma_data), GFP_KERNEL);
 	DMAP_BGON(pvd == NULL);
 
@@ -250,7 +261,9 @@ static int raw_open(struct inode *inode, struct file *filp)
 
 	pvd->magic2 = PVD_MAGIC_2;
 
-	filp->private_data = pvd;
+	fmap_info->pvd = pvd;
+
+	filp->private_data = fmap_info;
 
 	mutex_unlock(&raw_mutex);
 
@@ -282,7 +295,7 @@ static int raw_release(struct inode *inode, struct file *filp)
 
 	printk(KERN_ERR "[%s:%s:%d] Calling release\n", __FILE__, __func__, __LINE__);
 
-	pvd = (struct pr_vma_data *)filp->private_data;
+	pvd = ((struct fastmap_info *)filp->private_data)->pvd;
   if(pvd != NULL){
 
 
@@ -858,7 +871,7 @@ int dmap_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 {
 #if 1
 	struct mm_struct *mm = current->mm;
-	struct pr_vma_data *pvd = (struct pr_vma_data *)file->private_data;
+	struct pr_vma_data *pvd = ((struct fastmap_info *)file->private_data)->pvd;
 
 	printk(KERN_ERR "Calling msync() .....\n");
 	down_write(&mm->mmap_sem);
@@ -1214,6 +1227,12 @@ static int __init raw_init(void)
 
 	memset(&dimmap_buffer, 0, sizeof(dimmap_buffer_t));
 
+	/* Try to find the unexported kernel symbols which we need */
+	if(acquire_ksyms() == -1){
+		printk(KERN_ERR "Could not translate all necessary kernel symbols\n");
+		return -EINVAL;
+	}
+
 	/* Loadtime buffer parameters */
 	init_ld_params(&dimmap_buffer.ld_params);
 	dimmap_buffer.ld_params.mmap_buf_size = perma_mmap_buf_size >> 0;

driver/mmap_banked_buffer.c

@@ -81,7 +81,6 @@ void cleanup_mmap_buffer_data_banked_buffer_t(banked_buffer_t *buf)
 
 	printk(KERN_INFO "Cleaning up data structures for the banked buffer\n");
 	cleanup_mmap_buffer_data(buf->banks);
-	kfree(buf->banks);
 	kfree(buf);
 }