don't panic if a program frees all its memory with sbrk().

if a program sbrk()'s to a non-page-boundary, don't free that page.
corresponding usertests.

kernel/proc.c

@@ -232,9 +232,7 @@ growproc(int n)
       return -1;
     }
   } else if(n < 0){
-    if((sz = uvmdealloc(p->pagetable, sz, sz + n)) == 0) {
-      return -1;
-    }
+    sz = uvmdealloc(p->pagetable, sz, sz + n);
   }
   p->sz = sz;
   return 0;

kernel/vm.c

@@ -268,7 +268,10 @@ uvmdealloc(pagetable_t pagetable, uint64 oldsz, uint64 newsz)
 {
   if(newsz >= oldsz)
     return oldsz;
-  uvmunmap(pagetable, newsz, oldsz - newsz, 1);
+
+  uint64 newup = PGROUNDUP(newsz);
+  uvmunmap(pagetable, newup, oldsz - newup, 1);
+
   return newsz;
 }
 

user/usertests.c

@@ -1923,6 +1923,45 @@ pgbug(char *s)
   exit(0);
 }
 
+// does the kernel panic if a process sbrk()s its size to be less than
+// a page, or zero?
+void
+zerosize(char *s)
+{
+  int pid = fork();
+  if(pid < 0){
+    printf("fork failed\n");
+    exit(1);
+  }
+  if(pid == 0){
+    int sz = (uint64) sbrk(0);
+    // free all user memory; there used to be a bug that
+    // would not adjust p->sz correctly in this case,
+    // causing exit() to panic.
+    sbrk(-sz);
+    // user page fault here.
+    exit(0);
+  }
+  wait(0);
+
+  pid = fork();
+  if(pid < 0){
+    printf("fork failed\n");
+    exit(1);
+  }
+  if(pid == 0){
+    int sz = (uint64) sbrk(0);
+    // set the break to somewhere in the very first
+    // page; there used to be a bug that would incorrectly
+    // free the first page.
+    sbrk(-(sz - 3500));
+    exit(0);
+  }
+  wait(0);
+
+  exit(0);
+}
+
 // run each test in its own process. run returns 1 if child's exit()
 // indicates success.
 int
@@ -1961,6 +2000,7 @@ main(int argc, char *argv[])
     char *s;
   } tests[] = {
     {pgbug, "pgbug" },
+    {zerosize, "zerosize" },
     {reparent, "reparent" },
     {twochildren, "twochildren"},
     {forkfork, "forkfork"},