On Wednesday, August 16, 2017, Nicolas Pitre wrote:
> When cramfs_physmem is used then we have the opportunity to map files
> directly from ROM, directly into user space, saving on RAM usage.
> This gives us Execute-In-Place (XIP) support.
>
> For a file to be mmap()-able, the map area has to correspond to a range
> of uncompressed and contiguous blocks, and in the MMU case it also has
> to be page aligned. A version of mkcramfs with appropriate support is
> necessary to create such a filesystem image.
>
> In the MMU case it may happen for a vma structure to extend beyond the
> actual file size. This is notably the case in binfmt_elf.c:elf_map().
> Or the file's last block is shared with other files and cannot be mapped
> as is. Rather than refusing to mmap it, we do a partial map and set up a
> special vm_ops fault handler that splits the vma in two: the direct
> mapping
> vma and the memory-backed vma populated by the readpage method.
>
> In the non-MMU case it is the get_unmapped_area method that is responsible
> for providing the address where the actual data can be found. No mapping
> is necessary of course.
>
> Signed-off-by: Nicolas Pitre <nico@xxxxxxxxxx>
> ---
> fs/cramfs/inode.c | 270
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 270 insertions(+)
>
> diff --git a/fs/cramfs/inode.c b/fs/cramfs/inode.c
> index b825ae162c..e3884c607b 100644
> --- a/fs/cramfs/inode.c
> +++ b/fs/cramfs/inode.c
> @@ -16,6 +16,7 @@
> #include <linux/module.h>
> #include <linux/fs.h>
> #include <linux/pagemap.h>
> +#include <linux/ramfs.h>
> #include <linux/init.h>
> #include <linux/string.h>
> #include <linux/blkdev.h>
> @@ -49,6 +50,7 @@ static inline struct cramfs_sb_info *CRAMFS_SB(struct
> super_block *sb)
> static const struct super_operations cramfs_ops;
> static const struct inode_operations cramfs_dir_inode_operations;
> static const struct file_operations cramfs_directory_operations;
> +static const struct file_operations cramfs_physmem_fops;
> static const struct address_space_operations cramfs_aops;
>
> static DEFINE_MUTEX(read_mutex);
> @@ -96,6 +98,10 @@ static struct inode *get_cramfs_inode(struct
> super_block *sb,
> case S_IFREG:
> inode->i_fop = &generic_ro_fops;
> inode->i_data.a_ops = &cramfs_aops;
> + if (IS_ENABLED(CONFIG_CRAMFS_PHYSMEM) &&
> + CRAMFS_SB(sb)->flags & CRAMFS_FLAG_EXT_BLOCK_POINTERS &&
> + CRAMFS_SB(sb)->linear_phys_addr)
> + inode->i_fop = &cramfs_physmem_fops;
> break;
> case S_IFDIR:
> inode->i_op = &cramfs_dir_inode_operations;
> @@ -277,6 +283,270 @@ static void *cramfs_read(struct super_block *sb,
> unsigned int offset,
> return NULL;
> }
>
> +/*
> + * For a mapping to be possible, we need a range of uncompressed and
> + * contiguous blocks. Return the offset for the first block and number of
> + * valid blocks for which that is true, or zero otherwise.
> + */
> +static u32 cramfs_get_block_range(struct inode *inode, u32 pgoff, u32
> *pages)
> +{
> + struct super_block *sb = inode->i_sb;
> + struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
> + int i;
> + u32 *blockptrs, blockaddr;
> +
> + /*
> + * We can dereference memory directly here as this code may be
> + * reached only when there is a direct filesystem image mapping
> + * available in memory.
> + */
> + blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode) +
> pgoff*4);
> + blockaddr = blockptrs[0] & ~CRAMFS_BLK_FLAGS;
> + i = 0;
> + do {
> + u32 expect = blockaddr + i * (PAGE_SIZE >> 2);
> + expect |=
> CRAMFS_BLK_FLAG_DIRECT_PTR|CRAMFS_BLK_FLAG_UNCOMPRESSED;
> + if (blockptrs[i] != expect) {
> + pr_debug("range: block %d/%d got %#x expects %#x\n",
> + pgoff+i, pgoff+*pages-1, blockptrs[i], expect);
> + if (i == 0)
> + return 0;
> + break;
> + }
> + } while (++i < *pages);
> +
> + *pages = i;
> +
> + /* stored "direct" block ptrs are shifted down by 2 bits */
> + return blockaddr << 2;
> +}
> +
> +/*
> + * It is possible for cramfs_physmem_mmap() to partially populate the
> mapping
> + * causing page faults in the unmapped area. When that happens, we need
> to
> + * split the vma so that the unmapped area gets its own vma that can be
> backed
> + * with actual memory pages and loaded normally. This is necessary
> because
> + * remap_pfn_range() overwrites vma->vm_pgoff with the pfn and
> filemap_fault()
> + * no longer works with it. Furthermore this makes /proc/x/maps right.
> + * Q: is there a way to do split vma at mmap() time?
> + */
> +static const struct vm_operations_struct cramfs_vmasplit_ops;
> +static int cramfs_vmasplit_fault(struct vm_fault *vmf)
> +{
> + struct mm_struct *mm = vmf->vma->vm_mm;
> + struct vm_area_struct *vma, *new_vma;
> + unsigned long split_val, split_addr;
> + unsigned int split_pgoff, split_page;
> + int ret;
> +
> + /* Retrieve the vma split address and validate it */
> + vma = vmf->vma;
> + split_val = (unsigned long)vma->vm_private_data;
> + split_pgoff = split_val & 0xffff;
> + split_page = split_val >> 16;
> + split_addr = vma->vm_start + split_page * PAGE_SIZE;
> + pr_debug("fault: addr=%#lx vma=%#lx-%#lx split=%#lx\n",
> + vmf->address, vma->vm_start, vma->vm_end, split_addr);
> + if (!split_val || split_addr >= vma->vm_end || vmf->address <
> split_addr)
> + return VM_FAULT_SIGSEGV;
> +
> + /* We have some vma surgery to do and need the write lock. */
> + up_read(&mm->mmap_sem);
> + if (down_write_killable(&mm->mmap_sem))
> + return VM_FAULT_RETRY;
> +
> + /* Make sure the vma didn't change between the locks */
> + vma = find_vma(mm, vmf->address);
> + if (vma->vm_ops != &cramfs_vmasplit_ops) {
> + /*
> + * Someone else raced with us and could have handled the fault.
> + * Let it go back to user space and fault again if necessary.
> + */
> + downgrade_write(&mm->mmap_sem);
> + return VM_FAULT_NOPAGE;
> + }
> +
> + /* Split the vma between the directly mapped area and the rest */
> + ret = split_vma(mm, vma, split_addr, 0);
> + if (ret) {
> + downgrade_write(&mm->mmap_sem);
> + return VM_FAULT_OOM;
> + }
> +
> + /* The direct vma should no longer ever fault */
> + vma->vm_ops = NULL;
> +
> + /* Retrieve the new vma covering the unmapped area */
> + new_vma = find_vma(mm, split_addr);
> + BUG_ON(new_vma == vma);
> + if (!new_vma) {
> + downgrade_write(&mm->mmap_sem);
> + return VM_FAULT_SIGSEGV;
> + }
> +
> + /*
> + * Readjust the new vma with the actual file based pgoff and
> + * process the fault normally on it.
> + */
> + new_vma->vm_pgoff = split_pgoff;
> + new_vma->vm_ops = &generic_file_vm_ops;
> + vmf->vma = new_vma;
> + vmf->pgoff = split_pgoff;
> + vmf->pgoff += (vmf->address - new_vma->vm_start) >> PAGE_SHIFT;
> + downgrade_write(&mm->mmap_sem);
> + return filemap_fault(vmf);
> +}
> +
> +static const struct vm_operations_struct cramfs_vmasplit_ops = {
> + .fault = cramfs_vmasplit_fault,
> +};
> +
> +static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct
> *vma)
> +{
> + struct inode *inode = file_inode(file);
> + struct super_block *sb = inode->i_sb;
> + struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
> + unsigned int pages, vma_pages, max_pages, offset;
> + unsigned long address;
> + char *fail_reason;
> + int ret;
> +
> + if (!IS_ENABLED(CONFIG_MMU))
> + return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -
> ENOSYS;
> +
> + if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
> + return -EINVAL;
> +
> + /* Could COW work here? */
> + fail_reason = "vma is writable";
> + if (vma->vm_flags & VM_WRITE)
> + goto fail;
> +
> + vma_pages = (vma->vm_end - vma->vm_start + PAGE_SIZE - 1) >>
> PAGE_SHIFT;
> + max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
> + fail_reason = "beyond file limit";
> + if (vma->vm_pgoff >= max_pages)
> + goto fail;
> + pages = vma_pages;
> + if (pages > max_pages - vma->vm_pgoff)
> + pages = max_pages - vma->vm_pgoff;
> +
> + offset = cramfs_get_block_range(inode, vma->vm_pgoff, &pages);
> + fail_reason = "unsuitable block layout";
> + if (!offset)
> + goto fail;
> + address = sbi->linear_phys_addr + offset;
> + fail_reason = "data is not page aligned";
> + if (!PAGE_ALIGNED(address))
> + goto fail;
> +
> + /* Don't map the last page if it contains some other data */
> + if (unlikely(vma->vm_pgoff + pages == max_pages)) {
> + unsigned int partial = offset_in_page(inode->i_size);
> + if (partial) {
> + char *data = sbi->linear_virt_addr + offset;
> + data += (max_pages - 1) * PAGE_SIZE + partial;
> + while ((unsigned long)data & 7)
> + if (*data++ != 0)
> + goto nonzero;
> + while (offset_in_page(data)) {
> + if (*(u64 *)data != 0) {
> + nonzero:
> + pr_debug("mmap: %s: last page is shared\n",
> + file_dentry(file)->d_name.name);
> + pages--;
> + break;
> + }
> + data += 8;
> + }
> + }
> + }
> +
> + if (pages) {
> + /*
> + * If we can't map it all, page faults will occur if the
> + * unmapped area is accessed. Let's handle them to split the
> + * vma and let the normal paging machinery take care of the
> + * rest through cramfs_readpage(). Because remap_pfn_range()
> + * repurposes vma->vm_pgoff, we have to save it somewhere.
> + * Let's use vma->vm_private_data to hold both the pgoff and
> the actual address split point.
> + * Maximum file size is 16MB so we can pack both together.
> + */
> + if (pages != vma_pages) {
> + unsigned int split_pgoff = vma->vm_pgoff + pages;
> + unsigned long split_val = split_pgoff + (pages << 16);
> + vma->vm_private_data = (void *)split_val;
> + vma->vm_ops = &cramfs_vmasplit_ops;
> + /* to keep remap_pfn_range() happy */
> + vma->vm_end = vma->vm_start + pages * PAGE_SIZE;
> + }
> +
> + ret = remap_pfn_range(vma, vma->vm_start, address >>
> PAGE_SHIFT,
> + pages * PAGE_SIZE, vma->vm_page_prot);
space before tab in indent
-Chris
