2000-12-28 Roland McGrath <roland@frob.com>

	Override the shared code from serverboot for device access with
	a new, simpler implementation with no filesystem format support.
	The new code implements the new default_pager_paging_storage RPC
	to set up paging areas that can be subsets of whole Mach devices,
	with no Linux signature checking.  The compatibility setup entry
	point works as before on whole devices.
	* setup.c, file_io.h: New files.
	* Makefile (SRCS): Remove def_pager_setup.c, file_io.c, strfcns.c,
	ext2_file_io.c, ffs_file_io.c, ffs_compat.c, minix_file_io.c,
	minix_ffs_compat.c; add setup.c to replace them all.

3 files changed, 368 insertions, 8 deletions

diff --git a/mach-defpager/Makefile b/mach-defpager/Makefile
index ef59ed1f..96068fc2 100644
--- a/mach-defpager/Makefile
+++ b/mach-defpager/Makefile
@@ -22,15 +22,12 @@ dir	:= mach-defpager
 makemode:= server
 target	:= mach-defpager
 
-SRCS	:= default_pager.c def_pager_setup.c \
-	   kalloc.c wiring.c main.c \
-	   file_io.c strfcns.c \
-	   ext2_file_io.c ffs_file_io.c ffs_compat.c \
-	   minix_file_io.c minix_ffs_compat.c
-
-OBJS = $(SRCS:.c=.o) \
-       $(addsuffix Server.o,memory_object default_pager memory_object_default \
+SRCS	:= default_pager.c kalloc.c wiring.c main.c setup.c
+OBJS 	:= $(SRCS:.c=.o) \
+	   $(addsuffix Server.o,\
+		       memory_object default_pager memory_object_default \
                             exc bootstrap)
+# None of those pay attention to SERVERPREFIX.
 
 HURDLIBS:= threads
 LDFLAGS += -static
diff --git a/mach-defpager/file_io.h b/mach-defpager/file_io.h
new file mode 100644
index 00000000..d0b03f33
--- /dev/null
+++ b/mach-defpager/file_io.h
@@ -0,0 +1,69 @@
+/* Backing store access callbacks for Hurd version of Mach default pager.
+   Copyright (C) 2000 Free Software Foundation, Inc.
+
+   This file is part of the GNU Hurd.
+
+   The GNU Hurd 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, or (at
+   your option) any later version.
+
+   The GNU Hurd 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., 675 Mass Ave, Cambridge, MA 02139, USA. */
+
+#ifndef _file_io_h
+#define _file_io_h 1
+
+/* The original Mach default pager code used in serverboot can read
+   filesystem meta-data to find the blocks used by paging files.
+   We replace those interfaces with simpler code that only supports
+   subsets of devices represented by a list of runs a la libstore.  */
+
+#include <sys/types.h>
+
+#include <device/device_types.h>
+#include <device/device.h>
+
+/* A run of device records, expressed in the device's record size.  */
+struct storage_run
+{
+  recnum_t start, length;
+};
+
+struct file_direct
+{
+  mach_port_t device;
+
+  int bshift;			/* size of device records (disk blocks) */
+  size_t fd_bsize;		/* log2 of that */
+  recnum_t fd_size;		/* number of blocks total */
+
+  /* The paging area consists of the concatentation of NRUNS contiguous
+     regions of the device, as described by RUNS.  */
+  size_t nruns;
+  struct storage_run runs[0];
+};
+
+/* These are in fact only called to read or write a single page, from
+   default_pager.c::default_read/default_write.  The SIZE argument is
+   always vm_page_size and OFFSET is always page-aligned.  */
+
+int page_read_file_direct (struct file_direct *fdp,
+			   vm_offset_t offset,
+			   vm_size_t size,
+			   vm_offset_t *addr,			/* out */
+			   mach_msg_type_number_t *size_read);	/* out */
+int page_write_file_direct(struct file_direct *fdp,
+			   vm_offset_t offset,
+			   vm_offset_t addr,
+			   vm_size_t size,
+			   vm_offset_t *size_written); /* out */
+
+
+#endif /* file_io.h */
diff --git a/mach-defpager/setup.c b/mach-defpager/setup.c
new file mode 100644
index 00000000..134036ae
--- /dev/null
+++ b/mach-defpager/setup.c
@@ -0,0 +1,294 @@
+/* Backing store access callbacks for Hurd version of Mach default pager.
+   Copyright (C) 2001 Free Software Foundation, Inc.
+
+   This file is part of the GNU Hurd.
+
+   The GNU Hurd 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, or (at
+   your option) any later version.
+
+   The GNU Hurd 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., 675 Mass Ave, Cambridge, MA 02139, USA. */
+
+#include <errno.h>
+#include <stddef.h>
+#include <assert.h>
+#include <mach.h>
+
+#include "file_io.h"
+#include "default_pager_S.h"
+
+/* This should be in some system header...  XXX  */
+static inline int page_aligned (vm_offset_t num)
+{
+  return trunc_page (num) == num;
+}
+
+/* From serverboot/kalloc.c.  */
+extern void *kalloc (vm_size_t);
+
+kern_return_t
+default_pager_paging_storage (mach_port_t pager,
+			      mach_port_t device,
+			      recnum_t *runs, mach_msg_type_number_t nrun,
+			      default_pager_filename_t name,
+			      boolean_t add)
+{
+  struct file_direct *fdp;
+  int sizes[DEV_GET_SIZE_COUNT];
+  natural_t count;
+  mach_msg_type_number_t i;
+  error_t err;
+  recnum_t devsize;
+
+  if (! add)
+    return remove_paging_file (name); /* XXX ? */
+
+  if (nrun < 2 || nrun % 2 != 0)
+    return EINVAL;
+
+  count = DEV_GET_SIZE_COUNT;
+  err = device_get_status (device, DEV_GET_SIZE, sizes, &count);
+  if (err)
+    return err;
+  if (count < DEV_GET_SIZE_COUNT || sizes[DEV_GET_SIZE_RECORD_SIZE] <= 0)
+    return EINVAL;
+  devsize = sizes[DEV_GET_SIZE_DEVICE_SIZE] / sizes[DEV_GET_SIZE_RECORD_SIZE];
+
+  if (vm_page_size % sizes[DEV_GET_SIZE_RECORD_SIZE] != 0)
+    /* We can't write disk blocks larger than pages.  */
+    return EINVAL;
+
+  fdp = kalloc (offsetof (struct file_direct, runs[nrun]));
+  if (fdp == 0)
+    return ENOMEM;
+
+  fdp->device = device;
+  fdp->bshift = ffs (sizes[DEV_GET_SIZE_RECORD_SIZE]) - 1;
+  fdp->fd_bsize = sizes[DEV_GET_SIZE_RECORD_SIZE];
+  fdp->nruns = nrun / 2;
+  fdp->fd_size = 0;
+  for (i = 0; i < nrun; i += 2)
+    {
+      fdp->runs[i].start = runs[i];
+      fdp->runs[i].length = runs[i + 1];
+      if (fdp->runs[i].start + fdp->runs[i].length > devsize)
+	{
+	  kfree (fdp);
+	  return EINVAL;
+	}
+      fdp->fd_size += fdp->runs[i].length;
+    }
+
+  /* Now really do it.  */
+  create_paging_partition (name, fdp, 0, -3);
+  return 0;
+}
+
+
+/* Called to read a page from backing store.  */
+int
+page_read_file_direct (struct file_direct *fdp,
+		       vm_offset_t offset,
+		       vm_size_t size,
+		       vm_offset_t *addr,			/* out */
+		       mach_msg_type_number_t *size_read)	/* out */
+{
+  struct storage_run *r;
+  error_t err;
+  char *readloc;
+  char *page;
+  mach_msg_type_number_t nread;
+
+  assert (page_aligned (offset));
+  assert (size == vm_page_size);
+
+  offset >>= fdp->bshift;
+
+  assert (offset + (size >> fdp->bshift) <= fdp->fd_size);
+
+  /* Find the run containing the beginning of the page.  */
+  for (r = fdp->runs; r->length > offset; ++r)
+    offset -= r->length;
+
+  if (offset + (size >> fdp->bshift) <= r->length)
+    /* The first run contains the whole page.  */
+    return device_read (fdp->device, 0,
+			r->start + offset, size >> fdp->bshift,
+			(char **) addr, size_read);
+
+  /* Read the first part of the run.  */
+  err = device_read (fdp->device, 0, r->start + offset, r->length - offset,
+		     (char **) addr, &nread);
+  if (err)
+    return err;
+
+  size -= nread;
+  readloc = (char *) *addr;
+  do
+    {
+      readloc += nread;
+      offset += nread >> fdp->bshift;
+      if (r->length > offset)
+	offset -= r++->length;
+
+      /* We always get another out-of-line page, so we have to copy
+	 out of that page and deallocate it.  */
+      err = device_read (fdp->device, 0, r->start + offset, r->length - offset,
+			 &page, &nread);
+      if (err)
+	{
+	  vm_deallocate (mach_task_self (),
+			 (vm_address_t) *addr, vm_page_size);
+	  return err;
+	}
+      memcpy (readloc, page, nread);
+      vm_deallocate (mach_task_self (), (vm_address_t) page, vm_page_size);
+      size -= nread;
+    } while (size > 0);
+
+  *size_read = vm_page_size;
+  return 0;
+}
+
+/* Called to write a page to backing store.  */
+int
+page_write_file_direct(struct file_direct *fdp,
+		       vm_offset_t offset,
+		       vm_offset_t addr,
+		       vm_size_t size,
+		       vm_offset_t *size_written) /* out */
+{
+  struct storage_run *r;
+  error_t err;
+  int wrote;
+
+  assert (page_aligned (offset));
+  assert (size == vm_page_size);
+
+  offset >>= fdp->bshift;
+
+  assert (offset + (size >> fdp->bshift) <= fdp->fd_size);
+
+  /* Find the run containing the beginning of the page.  */
+  for (r = fdp->runs; r->length > offset; ++r)
+    offset -= r->length;
+
+  if (offset + (size >> fdp->bshift) <= r->length)
+    /* The first run contains the whole page.  */
+    return device_write (fdp->device, 0,
+			 r->start + offset, (char *) addr, size, &wrote);
+
+  /* Write the first part of the run.  */
+  err = device_write (fdp->device, 0,
+		      r->start + offset, (char *) addr,
+		      (r->length - offset) << fdp->bshift,
+		      &wrote);
+  if (err)
+    return err;
+
+  size -= wrote;
+  do
+    {
+      mach_msg_type_number_t segsize;
+
+      addr += wrote;
+      offset += wrote >> fdp->bshift;
+      if (r->length > offset)
+	offset -= r++->length;
+
+      segsize = (r->length - offset) >> fdp->bshift;
+      if (segsize > size)
+	segsize = size;
+      err = device_write (fdp->device, 0,
+			  r->start + offset, (char *) addr, segsize, &wrote);
+      if (err)
+	{
+	  vm_deallocate (mach_task_self (),
+			 (vm_address_t) addr, vm_page_size);
+	  return err;
+	}
+
+      size -= wrote;
+    } while (size > 0);
+
+  *size_written = vm_page_size;
+  return 0;
+}
+
+
+/* Compatibility entry points used by default_pager_paging_file RPC.  */
+
+kern_return_t
+add_paging_file(master_device_port, file_name, linux_signature)
+	mach_port_t		master_device_port;
+	char			*file_name;
+	int			linux_signature;
+{
+  error_t err;
+  mach_port_t dev;
+  int sizes[DEV_GET_SIZE_COUNT];
+  natural_t count;
+  char *devname = file_name;
+
+  assert (linux_signature == 0);
+
+  if (!strncmp (file_name, "/dev/", 5))
+    devname += 5;
+
+  err = device_open (master_device_port, D_READ|D_WRITE, devname, &dev);
+  if (err)
+    return err;
+
+  count = DEV_GET_SIZE_COUNT;
+  err = device_get_status (dev, DEV_GET_SIZE, sizes, &count);
+  if (!err && count < DEV_GET_SIZE_COUNT)
+    err = EGRATUITOUS;
+  if (err)
+    mach_port_deallocate (mach_task_self (), dev);
+  else
+    {
+      struct file_direct *fdp;
+      fdp = kalloc (offsetof (struct file_direct, runs[1]));
+      if (fdp == 0)
+	return ENOMEM;
+
+      fdp->device = dev;
+      fdp->fd_bsize = sizes[DEV_GET_SIZE_RECORD_SIZE];
+      fdp->bshift = ffs (sizes[DEV_GET_SIZE_RECORD_SIZE]) - 1;
+      fdp->fd_size = sizes[DEV_GET_SIZE_DEVICE_SIZE] >> fdp->bshift;
+      fdp->nruns = 1;
+      fdp->runs[0].start = 0;
+      fdp->runs[0].length = fdp->fd_size;
+
+      /* Now really do it.  */
+      create_paging_partition (file_name, fdp, 0, 0);
+    }
+
+  return err;
+}
+
+/*
+ * Destroy a paging_partition given a file name
+ */
+kern_return_t
+remove_paging_file (char *file_name)
+{
+  struct file_direct *fdp = 0;
+  kern_return_t kr;
+
+  kr = destroy_paging_partition(file_name, &fdp);
+  if (kr == KERN_SUCCESS && fdp != 0)
+    {
+      mach_port_deallocate (mach_task_self (), fdp->device);
+      kfree (fdp, (char *) &fdp->runs[fdp->nruns] - (char *) fdp);
+    }
+  return kr;
+}