Introduce guest mem offset, static link example launcher

In order to avoid problematic special linking of the Launcher, we give
the Host an offset: this means we can use any memory region in the
Launcher as Guest memory rather than insisting on mmap() at 0.

The result is quite pleasing: a number of casts are replaced with
simple additions.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

6 files changed, 50 insertions, 38 deletions

diff --git a/drivers/lguest/core.c b/drivers/lguest/core.c
index a0788c1..eb95860 100644
--- a/drivers/lguest/core.c
+++ b/drivers/lguest/core.c
@@ -325,8 +325,8 @@ static int emulate_insn(struct lguest *lg)
  * Dealing With Guest Memory.
  *
  * When the Guest gives us (what it thinks is) a physical address, we can use
- * the normal copy_from_user() & copy_to_user() on that address: remember,
- * Guest physical == Launcher virtual.
+ * the normal copy_from_user() & copy_to_user() on the corresponding place in
+ * the memory region allocated by the Launcher.
  *
  * But we can't trust the Guest: it might be trying to access the Launcher
  * code.  We have to check that the range is below the pfn_limit the Launcher
@@ -348,8 +348,8 @@ u32 lgread_u32(struct lguest *lg, unsigned long addr)
 
 	/* Don't let them access lguest binary. */
 	if (!lguest_address_ok(lg, addr, sizeof(val))
-	    || get_user(val, (u32 __user *)addr) != 0)
-		kill_guest(lg, "bad read address %#lx", addr);
+	    || get_user(val, (u32 *)(lg->mem_base + addr)) != 0)
+		kill_guest(lg, "bad read address %#lx: pfn_limit=%u membase=%p", addr, lg->pfn_limit, lg->mem_base);
 	return val;
 }
 
@@ -357,7 +357,7 @@ u32 lgread_u32(struct lguest *lg, unsigned long addr)
 void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
 {
 	if (!lguest_address_ok(lg, addr, sizeof(val))
-	    || put_user(val, (u32 __user *)addr) != 0)
+	    || put_user(val, (u32 *)(lg->mem_base + addr)) != 0)
 		kill_guest(lg, "bad write address %#lx", addr);
 }
 
@@ -367,7 +367,7 @@ void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
 void lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || copy_from_user(b, (void __user *)addr, bytes) != 0) {
+	    || copy_from_user(b, lg->mem_base + addr, bytes) != 0) {
 		/* copy_from_user should do this, but as we rely on it... */
 		memset(b, 0, bytes);
 		kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
@@ -379,7 +379,7 @@ void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
 	     unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || copy_to_user((void __user *)addr, b, bytes) != 0)
+	    || copy_to_user(lg->mem_base + addr, b, bytes) != 0)
 		kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
 }
 /* (end of memory access helper routines) :*/
@@ -616,11 +616,9 @@ int run_guest(struct lguest *lg, unsigned long __user *user)
 			 *
 			 * Note that if the Guest were really messed up, this
 			 * could happen before it's done the INITIALIZE
-			 * hypercall, so lg->lguest_data will be NULL, so
-			 * &lg->lguest_data->cr2 will be address 8.  Writing
-			 * into that address won't hurt the Host at all,
-			 * though. */
-			if (put_user(cr2, &lg->lguest_data->cr2))
+			 * hypercall, so lg->lguest_data will be NULL */
+			if (lg->lguest_data
+			    && put_user(cr2, &lg->lguest_data->cr2))
 				kill_guest(lg, "Writing cr2");
 			break;
 		case 7: /* We've intercepted a Device Not Available fault. */
diff --git a/drivers/lguest/hypercalls.c b/drivers/lguest/hypercalls.c
index 5ecd60b..02e67b4 100644
--- a/drivers/lguest/hypercalls.c
+++ b/drivers/lguest/hypercalls.c
@@ -205,16 +205,19 @@ static void initialize(struct lguest *lg)
 		tsc_speed = 0;
 
 	/* The pointer to the Guest's "struct lguest_data" is the only
-	 * argument. */
-	lg->lguest_data = (struct lguest_data __user *)lg->regs->edx;
-	/* If we check the address they gave is OK now, we can simply
-	 * copy_to_user/from_user from now on rather than using lgread/lgwrite.
-	 * I put this in to show that I'm not immune to writing stupid
-	 * optimizations. */
+	 * argument.  We check that address now. */
 	if (!lguest_address_ok(lg, lg->regs->edx, sizeof(*lg->lguest_data))) {
 		kill_guest(lg, "bad guest page %p", lg->lguest_data);
 		return;
 	}
+
+	/* Having checked it, we simply set lg->lguest_data to point straight
+	 * into the Launcher's memory at the right place and then use
+	 * copy_to_user/from_user from now on, instead of lgread/write.  I put
+	 * this in to show that I'm not immune to writing stupid
+	 * optimizations. */
+	lg->lguest_data = lg->mem_base + lg->regs->edx;
+
 	/* The Guest tells us where we're not to deliver interrupts by putting
 	 * the range of addresses into "struct lguest_data". */
 	if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
diff --git a/drivers/lguest/io.c b/drivers/lguest/io.c
index ea68613..3a84533 100644
--- a/drivers/lguest/io.c
+++ b/drivers/lguest/io.c
@@ -186,7 +186,7 @@ int bind_dma(struct lguest *lg,
 	 * we're doing this. */
 	mutex_lock(&lguest_lock);
 	down_read(fshared);
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
+	if (get_futex_key(lg->mem_base + ukey, fshared, &key) != 0) {
 		kill_guest(lg, "bad dma key %#lx", ukey);
 		goto unlock;
 	}
@@ -247,7 +247,8 @@ static int lgread_other(struct lguest *lg,
 			void *buf, u32 addr, unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || access_process_vm(lg->tsk, addr, buf, bytes, 0) != bytes) {
+	    || access_process_vm(lg->tsk, (unsigned long)lg->mem_base + addr,
+				 buf, bytes, 0) != bytes) {
 		memset(buf, 0, bytes);
 		kill_guest(lg, "bad address in registered DMA struct");
 		return 0;
@@ -261,8 +262,8 @@ static int lgwrite_other(struct lguest *lg, u32 addr,
 			 const void *buf, unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || (access_process_vm(lg->tsk, addr, (void *)buf, bytes, 1)
-		!= bytes)) {
+	    || access_process_vm(lg->tsk, (unsigned long)lg->mem_base + addr,
+				 (void *)buf, bytes, 1) != bytes) {
 		kill_guest(lg, "bad address writing to registered DMA");
 		return 0;
 	}
@@ -318,7 +319,7 @@ static u32 copy_data(struct lguest *srclg,
 		 * copy_to_user_page(), and some arch's seem to need special
 		 * flushes.  x86 is fine. */
 		if (copy_from_user(maddr + (dst->addr[di] + dstoff)%PAGE_SIZE,
-				   (void __user *)src->addr[si], len) != 0) {
+				   srclg->mem_base+src->addr[si], len) != 0) {
 			/* If a copy failed, it's the source's fault. */
 			kill_guest(srclg, "bad address in sending DMA");
 			totlen = 0;
@@ -377,7 +378,8 @@ static u32 do_dma(struct lguest *srclg, const struct lguest_dma *src,
 		 * number of pages.  Note that we're holding the destination's
 		 * mmap_sem, as get_user_pages() requires. */
 		if (get_user_pages(dstlg->tsk, dstlg->mm,
-				   dst->addr[i], 1, 1, 1, pages+i, NULL)
+				   (unsigned long)dstlg->mem_base+dst->addr[i],
+				   1, 1, 1, pages+i, NULL)
 		    != 1) {
 			/* This means the destination gave us a bogus buffer */
 			kill_guest(dstlg, "Error mapping DMA pages");
@@ -493,7 +495,7 @@ again:
 	mutex_lock(&lguest_lock);
 	down_read(fshared);
 	/* Get the futex key for the key the Guest gave us */
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
+	if (get_futex_key(lg->mem_base + ukey, fshared, &key) != 0) {
 		kill_guest(lg, "bad sending DMA key");
 		goto unlock;
 	}
@@ -584,7 +586,7 @@ unsigned long get_dma_buffer(struct lguest *lg,
 
 	/* This can fail if it's not a valid address, or if the address is not
 	 * divisible by 4 (the futex code needs that, we don't really). */
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
+	if (get_futex_key(lg->mem_base + ukey, fshared, &key) != 0) {
 		kill_guest(lg, "bad registered DMA buffer");
 		goto unlock;
 	}
diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h
index 399eab8..54f2c24 100644
--- a/drivers/lguest/lg.h
+++ b/drivers/lguest/lg.h
@@ -142,6 +142,9 @@ struct lguest
 	struct mm_struct *mm; 	/* == tsk->mm, but that becomes NULL on exit */
 	u16 guestid;
 	u32 pfn_limit;
+	/* This provides the offset to the base of guest-physical
+	 * memory in the Launcher. */
+	void __user *mem_base;
 	u32 page_offset;
 	u32 cr2;
 	int halted;
diff --git a/drivers/lguest/lguest_user.c b/drivers/lguest/lguest_user.c
index 80d1b58..816d4d1 100644
--- a/drivers/lguest/lguest_user.c
+++ b/drivers/lguest/lguest_user.c
@@ -1,9 +1,9 @@
 /*P:200 This contains all the /dev/lguest code, whereby the userspace launcher
  * controls and communicates with the Guest.  For example, the first write will
- * tell us the memory size, pagetable, entry point and kernel address offset.
- * A read will run the Guest until a signal is pending (-EINTR), or the Guest
- * does a DMA out to the Launcher.  Writes are also used to get a DMA buffer
- * registered by the Guest and to send the Guest an interrupt. :*/
+ * tell us the Guest's memory layout, pagetable, entry point and kernel address
+ * offset.  A read will run the Guest until something happens, such as a signal
+ * or the Guest doing a DMA out to the Launcher.  Writes are also used to get a
+ * DMA buffer registered by the Guest and to send the Guest an interrupt. :*/
 #include <linux/uaccess.h>
 #include <linux/miscdevice.h>
 #include <linux/fs.h>
@@ -142,9 +142,11 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
 	return run_guest(lg, (unsigned long __user *)user);
 }
 
-/*L:020 The initialization write supplies 4 32-bit values (in addition to the
+/*L:020 The initialization write supplies 5 32-bit values (in addition to the
  * 32-bit LHREQ_INITIALIZE value).  These are:
  *
+ * base: The start of the Guest-physical memory inside the Launcher memory.
+ *
  * pfnlimit: The highest (Guest-physical) page number the Guest should be
  * allowed to access.  The Launcher has to live in Guest memory, so it sets
  * this to ensure the Guest can't reach it.
@@ -166,7 +168,7 @@ static int initialize(struct file *file, const u32 __user *input)
 	 * Guest. */
 	struct lguest *lg;
 	int err, i;
-	u32 args[4];
+	u32 args[5];
 
 	/* We grab the Big Lguest lock, which protects the global array
 	 * "lguests" and multiple simultaneous initializations. */
@@ -194,8 +196,9 @@ static int initialize(struct file *file, const u32 __user *input)
 
 	/* Populate the easy fields of our "struct lguest" */
 	lg->guestid = i;
-	lg->pfn_limit = args[0];
-	lg->page_offset = args[3];
+	lg->mem_base = (void __user *)(long)args[0];
+	lg->pfn_limit = args[1];
+	lg->page_offset = args[4];
 
 	/* We need a complete page for the Guest registers: they are accessible
 	 * to the Guest and we can only grant it access to whole pages. */
@@ -210,13 +213,13 @@ static int initialize(struct file *file, const u32 __user *input)
 	/* Initialize the Guest's shadow page tables, using the toplevel
 	 * address the Launcher gave us.  This allocates memory, so can
 	 * fail. */
-	err = init_guest_pagetable(lg, args[1]);
+	err = init_guest_pagetable(lg, args[2]);
 	if (err)
 		goto free_regs;
 
 	/* Now we initialize the Guest's registers, handing it the start
 	 * address. */
-	setup_regs(lg->regs, args[2]);
+	setup_regs(lg->regs, args[3]);
 
 	/* There are a couple of GDT entries the Guest expects when first
 	 * booting. */
diff --git a/drivers/lguest/page_tables.c b/drivers/lguest/page_tables.c
index b7a924a..9cd2fac 100644
--- a/drivers/lguest/page_tables.c
+++ b/drivers/lguest/page_tables.c
@@ -152,7 +152,7 @@ static unsigned long get_pfn(unsigned long virtpfn, int write)
 static spte_t gpte_to_spte(struct lguest *lg, gpte_t gpte, int write)
 {
 	spte_t spte;
-	unsigned long pfn;
+	unsigned long pfn, base;
 
 	/* The Guest sets the global flag, because it thinks that it is using
 	 * PGE.  We only told it to use PGE so it would tell us whether it was
@@ -160,11 +160,14 @@ static spte_t gpte_to_spte(struct lguest *lg, gpte_t gpte, int write)
 	 * use the global bit, so throw it away. */
 	spte.flags = (gpte.flags & ~_PAGE_GLOBAL);
 
+	/* The Guest's pages are offset inside the Launcher. */
+	base = (unsigned long)lg->mem_base / PAGE_SIZE;
+
 	/* We need a temporary "unsigned long" variable to hold the answer from
 	 * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't
 	 * fit in spte.pfn.  get_pfn() finds the real physical number of the
 	 * page, given the virtual number. */
-	pfn = get_pfn(gpte.pfn, write);
+	pfn = get_pfn(base + gpte.pfn, write);
 	if (pfn == -1UL) {
 		kill_guest(lg, "failed to get page %u", gpte.pfn);
 		/* When we destroy the Guest, we'll go through the shadow page