sos-code-article10/drivers/zero.c

/* Copyright (C) 2005 David Decotigny

   This program is free software; you can redistribute it and/or
   modify it under the terms of the GNU General Public License
   as published by the Free Software Foundation; either version 2
   of the License, or (at your option) any later version.
   
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
   USA. 
*/

#include <sos/assert.h>
#include <sos/kmalloc.h>
#include <sos/physmem.h>
#include <hwcore/paging.h>
#include <sos/kmem_slab.h>
#include <sos/list.h>
#include <hwcore/paging.h>
#include <sos/uaccess.h>
#include <sos/chardev.h>
#include <drivers/devices.h>

#include "zero.h"


/**
 * A mapped page for a shared mapping of /dev/zero
 */
struct zero_mapped_page
{
  sos_uoffset_t page_id;
  sos_paddr_t   ppage_paddr;

  struct zero_mapped_page *prev, *next;
};
/** The Slab cache of shared mapped pages */
struct sos_kslab_cache * cache_of_zero_mapped_pages;


/**
 * A mapped /dev/zero resource
 */
struct zero_mapped_resource
{
  int ref_cnt;

  /**
   * For shared mappings: the list of shared pages mapped inside one
   * or multiple VRs
   */
  struct zero_mapped_page *list_mapped_pages;

  struct sos_umem_vmm_mapped_resource mr;
};


/** Forward declaration: the FS operation for the /dev/zero character
    device */
static struct sos_chardev_ops dev_zero_fs_ops;


/** Helper function to insert the given physical page in the list of
    physical pages used for shared anonymous mappings */
static sos_ret_t insert_anonymous_physpage(struct zero_mapped_resource *mr,
					   sos_paddr_t ppage_paddr,
					   sos_uoffset_t page_id);


/** Helper function to insert the given physical page in the list of
    physical pages used for shared anonymous mappings */
static sos_paddr_t lookup_anonymous_physpage(struct zero_mapped_resource *mr,
					     sos_uoffset_t page_id);


sos_ret_t sos_dev_zero_subsystem_setup()
{
  sos_ret_t retval;

  cache_of_zero_mapped_pages =
    sos_kmem_cache_create("shared anonymous mappings",
			  sizeof(struct zero_mapped_page),
			  1, 0,
			  SOS_KSLAB_CREATE_MAP | SOS_KSLAB_CREATE_ZERO);
  if (! cache_of_zero_mapped_pages)
    return -SOS_ENOMEM;

  retval = sos_chardev_register_class(SOS_CHARDEV_ZERO_MAJOR,
				      & dev_zero_fs_ops,
				      NULL);
  if (SOS_OK != retval)
    {
      sos_kmem_cache_destroy(cache_of_zero_mapped_pages);
      return retval;
    }

  return SOS_OK;
}


/** Called after the virtual region has been inserted inside its
    address space */
static void zero_ref(struct sos_umem_vmm_vr * vr)
{
  /* Retrieve the 'zero' structure associated with the mapped resource */
  struct zero_mapped_resource * zero_resource;
  zero_resource
    = (struct zero_mapped_resource*)
    sos_umem_vmm_get_mapped_resource_of_vr(vr)->custom_data;
  
  /* Increment ref counter */
  zero_resource->ref_cnt ++;
}


/** Called when the virtual region is removed from its address
    space */
static void zero_unref(struct sos_umem_vmm_vr * vr)
{
  /* Retrieve the 'zero' structure associated with the mapped resource */
  struct zero_mapped_resource * zero_resource;
  zero_resource
    = (struct zero_mapped_resource*)
    sos_umem_vmm_get_mapped_resource_of_vr(vr)->custom_data;

  /* Decrement ref coutner */
  SOS_ASSERT_FATAL(zero_resource->ref_cnt > 0);
  zero_resource->ref_cnt --;

  /* Free the resource if it becomes unused */
  if (zero_resource->ref_cnt == 0)
    {
      /* Delete the list of anonymous shared mappings */
      struct zero_mapped_page *zmp;
      list_collapse(zero_resource->list_mapped_pages, zmp)
	{
	  /* Unreference the underlying physical page */
	  sos_physmem_unref_physpage(zmp->ppage_paddr);
	  sos_kfree((sos_vaddr_t)zmp);
	}

      sos_kfree((sos_vaddr_t)zero_resource);
    }
}


/** MOST IMPORTANT callback ! Called when a thread page faults on the
    resource's mapping */
static sos_ret_t zero_page_in(struct sos_umem_vmm_vr * vr,
			      sos_uaddr_t uaddr,
			      sos_bool_t write_access)
{
  sos_ret_t retval = SOS_OK;
  sos_paddr_t ppage_paddr;
  sos_uoffset_t required_page_id;
  struct zero_mapped_resource * zero_resource;
  sos_ui32_t vr_prot, vr_flags;

  /* Retrieve the 'zero' structure associated with the mapped resource */
  zero_resource
    = (struct zero_mapped_resource*)
    sos_umem_vmm_get_mapped_resource_of_vr(vr)->custom_data;

  /* Retrieve access rights/flags of the VR */
  vr_prot  = sos_umem_vmm_get_prot_of_vr(vr);
  vr_flags = sos_umem_vmm_get_flags_of_vr(vr);

  /* Identifies the page in the mapping that's being paged-in */
  required_page_id = SOS_PAGE_ALIGN_INF(uaddr)
    - sos_umem_vmm_get_start_of_vr(vr)
    + sos_umem_vmm_get_offset_in_resource(vr);

  /* For shared mappings, check if there is a page already mapping the
     required address */
  if (vr_flags & SOS_VR_MAP_SHARED)
    {
      ppage_paddr = lookup_anonymous_physpage(zero_resource, required_page_id);
      if (NULL != (void*)ppage_paddr)
	{
	  retval = sos_paging_map(ppage_paddr,
				  SOS_PAGE_ALIGN_INF(uaddr),
				  TRUE,
				  vr_prot);

	  return retval;
	}
    }

  /* For write accesses, directly maps a new page. For read accesses,
     simply map in the zero_page (and wait for COW to handle the next
     write accesses) */
  if (write_access)
    {
      /* Allocate a new page for the virtual address */
      ppage_paddr = sos_physmem_ref_physpage_new(FALSE);
      if (! ppage_paddr)
	return -SOS_ENOMEM;

      retval = sos_paging_map(ppage_paddr,
			      SOS_PAGE_ALIGN_INF(uaddr),
			      TRUE,
			      vr_prot);
      if (SOS_OK != retval)
	{
	  sos_physmem_unref_physpage(ppage_paddr);
	  return retval;
	}
      
      memset((void*)SOS_PAGE_ALIGN_INF(uaddr), 0x0, SOS_PAGE_SIZE);

      /* For shared mappings, add the page in the list of shared
	 mapped pages */
      if (vr_flags & SOS_VR_MAP_SHARED)
	insert_anonymous_physpage(zero_resource, ppage_paddr,
				  required_page_id);

      sos_physmem_unref_physpage(ppage_paddr);
    }
  else
    {
      /* Map-in the zero page in READ ONLY whatever the access_rights
	 or the type (shared/private) of the VR to activate COW */
      retval = sos_paging_map(sos_zero_physpage,
			      SOS_PAGE_ALIGN_INF(uaddr),
			      TRUE,
			      SOS_VM_MAP_PROT_READ);
    }

  return retval;
}


/** The callbacks for a mapped /dev/zero resource */
static struct sos_umem_vmm_vr_ops zero_ops = (struct sos_umem_vmm_vr_ops)
{
  .ref     = zero_ref,
  .unref   = zero_unref,
  .page_in = zero_page_in,
  .unmap   = NULL
};


/** The callback that gets called when the resource gets mapped */
static sos_ret_t zero_mmap(struct sos_umem_vmm_vr *vr)
{
  return sos_umem_vmm_set_ops_of_vr(vr, &zero_ops);
}


/** The function responsible for mapping the /dev/zero resource in
    user space */
sos_ret_t sos_dev_zero_map(struct sos_umem_vmm_as * dest_as,
			   sos_uaddr_t *uaddr,
			   sos_size_t size,
			   sos_ui32_t access_rights,
			   sos_ui32_t flags)
{
  sos_ret_t retval;
  struct zero_mapped_resource * zero_resource;

  zero_resource
    = (struct zero_mapped_resource*) sos_kmalloc(sizeof(*zero_resource), 0);
  if (! zero_resource)
    return -SOS_ENOMEM;

  memset(zero_resource, 0x0, sizeof(*zero_resource));
  zero_resource->mr.allowed_access_rights 
    = SOS_VM_MAP_PROT_READ
    | SOS_VM_MAP_PROT_WRITE
    | SOS_VM_MAP_PROT_EXEC;
  zero_resource->mr.flags         |= SOS_MAPPED_RESOURCE_ANONYMOUS;
  zero_resource->mr.custom_data    = zero_resource;
  zero_resource->mr.mmap           = zero_mmap;

  retval = sos_umem_vmm_map(dest_as, uaddr, size,
			    access_rights, flags,
			    &zero_resource->mr, 0);
  if (SOS_OK != retval)
    {
      sos_kfree((sos_vaddr_t)zero_resource);
      return retval;
    }

  return SOS_OK;
}


static sos_ret_t insert_anonymous_physpage(struct zero_mapped_resource *mr,
					   sos_paddr_t ppage_paddr,
					   sos_uoffset_t page_id)
{
  struct zero_mapped_page * zmp
    = (struct zero_mapped_page*)sos_kmem_cache_alloc(cache_of_zero_mapped_pages,
						     0);
  if (! zmp)
    return -SOS_ENOMEM;

  zmp->page_id     = page_id;
  zmp->ppage_paddr = ppage_paddr;

  list_add_head(mr->list_mapped_pages, zmp);
  sos_physmem_ref_physpage_at(ppage_paddr);
  return SOS_OK;
}


static sos_paddr_t lookup_anonymous_physpage(struct zero_mapped_resource *mr,
					     sos_uoffset_t page_id)
{
  struct zero_mapped_page * zmp;
  int nb_elts;

  list_foreach_forward(mr->list_mapped_pages, zmp, nb_elts)
    {
      if (zmp->page_id == page_id)
	return zmp->ppage_paddr;
    }
  
  return (sos_paddr_t)NULL;
}

/*
 * /dev/zero character device FS operations
 */

static sos_ret_t dev_zero_fs_open(struct sos_fs_node        * fsnode,
				  struct sos_fs_opened_file * of,
				  void * chardev_class_custom_data)
{
  /* Make sure the device instance is known to the driver */
  if ( (SOS_CHARDEV_NULL_MINOR != fsnode->dev_id.device_instance)
       && (SOS_CHARDEV_ZERO_MINOR != fsnode->dev_id.device_instance) )
    return -SOS_ENODEV;

  return SOS_OK;
}


static sos_ret_t dev_zero_fs_seek(struct sos_fs_opened_file *this,
				  sos_lsoffset_t offset,
				  sos_seek_whence_t whence,
				  /* out */ sos_lsoffset_t * result_position)
{
  /* Artificiallly update the position in the "file" */
  sos_lsoffset_t ref_offs;

  *result_position = this->position;
  switch (whence)
    {
    case SOS_SEEK_SET:
      ref_offs = 0;
      break;

    case SOS_SEEK_CUR:
      ref_offs = this->position;
      break;

    case SOS_SEEK_END:
      return -SOS_ENOSUP;
      break;

    default:
      return -SOS_EINVAL;
    }

  if (offset < -ref_offs)
    return -SOS_EINVAL;
  
  this->position = ref_offs + offset;
  *result_position = this->position;
  return SOS_OK;
}


static sos_ret_t dev_zero_fs_read(struct sos_fs_opened_file *this,
				  sos_uaddr_t dest_buf,
				  sos_size_t * /* in/out */len)
{
  struct sos_fs_node * fsnode = sos_fs_nscache_get_fs_node(this->direntry);
  sos_size_t offs, rdlen;

  /* Reading /dev/null returns immediately */
  if (SOS_CHARDEV_NULL_MINOR == fsnode->dev_id.device_instance)
    {
      *len = 0;
      return SOS_OK;
    }

  /* ZERO the destination buffer using the zero page (by page_size
     increments) */
  for (rdlen = offs = 0 ; offs < *len ; offs += SOS_PAGE_SIZE)
    {
      sos_ret_t retval;
      sos_size_t memcpy_len = SOS_PAGE_SIZE;
      if (offs + memcpy_len > *len)
	memcpy_len = *len - offs;
      
      retval = sos_memcpy_to_user(dest_buf + offs, sos_zero_kernelpage,
				  memcpy_len);
      if (retval < 0)
	break;

      rdlen += retval;
      if (retval != (sos_ret_t)memcpy_len)
	break;
    }

  /* Artificiallly update the position in the "file" */
  *len = rdlen;
  this->position += rdlen;
  return SOS_OK;
}


static sos_ret_t dev_zero_fs_write(struct sos_fs_opened_file *this,
				   sos_uaddr_t src_buf,
				   sos_size_t * /* in/out */len)
{
  /* Artificiallly update the position in the "file" */
  this->position += *len;
  return SOS_OK;
}


static sos_ret_t dev_zero_fs_mmap(struct sos_fs_opened_file *this,
				  sos_uaddr_t *uaddr, sos_size_t size,
				  sos_ui32_t access_rights,
				  sos_ui32_t flags,
				  sos_luoffset_t offset)
{
  return sos_dev_zero_map(sos_process_get_address_space(this->owner),
			  uaddr, size, access_rights, flags);
}


static struct sos_chardev_ops dev_zero_fs_ops
  = (struct sos_chardev_ops) {
    .open  = dev_zero_fs_open,
    .close = NULL,
    .seek  = dev_zero_fs_seek,
    .read  = dev_zero_fs_read,
    .write = dev_zero_fs_write,
    .mmap  = dev_zero_fs_mmap,
    .fcntl = NULL,
    .ioctl = NULL
  };