LU-435 Add VM_FAULT_RETRY handling

[fs/lustre-release.git] / lustre / llite / llite_mmap.c

diff --git a/lustre/llite/llite_mmap.c b/lustre/llite/llite_mmap.c
index ae43a07..81ee3fb 100644 (file)
--- a/lustre/llite/llite_mmap.c
+++ b/lustre/llite/llite_mmap.c
@@ -119,8 +119,10 @@ struct vm_area_struct * our_vma(unsigned long addr, size_t count)
  * \retval EINVAL if env can't allocated
  * \return other error codes from cl_io_init.
  */
-int ll_fault_io_init(struct vm_area_struct *vma, struct lu_env **env_ret,
-                     struct cl_env_nest *nest, pgoff_t index, unsigned long *ra_flags)
+struct cl_io *ll_fault_io_init(struct vm_area_struct *vma,
+                               struct lu_env **env_ret,
+                               struct cl_env_nest *nest,
+                               pgoff_t index, unsigned long *ra_flags)
 {
         struct file       *file  = vma->vm_file;
         struct inode      *inode = file->f_dentry->d_inode;
@@ -130,8 +132,9 @@ int ll_fault_io_init(struct vm_area_struct *vma, struct lu_env **env_ret,
         struct lu_env     *env;
         ENTRY;
 
+        *env_ret = NULL;
         if (ll_file_nolock(file))
-                RETURN(-EOPNOTSUPP);
+                RETURN(ERR_PTR(-EOPNOTSUPP));
 
         /*
          * page fault can be called when lustre IO is
@@ -141,14 +144,12 @@ int ll_fault_io_init(struct vm_area_struct *vma, struct lu_env **env_ret,
          * one.
          */
         env = cl_env_nested_get(nest);
-        if (IS_ERR(env)) {
-                *env_ret = NULL;
-                 RETURN(-EINVAL);
-        }
+        if (IS_ERR(env))
+                 RETURN(ERR_PTR(-EINVAL));
 
         *env_ret = env;
 
-        io = &ccc_env_info(env)->cti_io;
+        io = ccc_env_thread_io(env);
         io->ci_obj = ll_i2info(inode)->lli_clob;
         LASSERT(io->ci_obj != NULL);
 
@@ -182,7 +183,7 @@ int ll_fault_io_init(struct vm_area_struct *vma, struct lu_env **env_ret,
                 cio->cui_fd  = fd;
         }
 
-        return io->ci_result;
+        return io;
 }
 
 #ifndef HAVE_VM_OP_FAULT
@@ -214,16 +215,15 @@ struct page *ll_nopage(struct vm_area_struct *vma, unsigned long address,
         ENTRY;
 
         pg_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
-        result = ll_fault_io_init(vma, &env,  &nest, pg_offset, &ra_flags);
-        if (env == NULL)
+        io = ll_fault_io_init(vma, &env,  &nest, pg_offset, &ra_flags);
+        if (IS_ERR(io))
                 return NOPAGE_SIGBUS;
 
-        io = &ccc_env_info(env)->cti_io;
+        result = io->ci_result;
         if (result < 0)
                 goto out_err;
 
         vio = vvp_env_io(env);
-
         vio->u.fault.ft_vma            = vma;
         vio->u.fault.nopage.ft_address = address;
         vio->u.fault.nopage.ft_type    = type;
@@ -259,7 +259,7 @@ out_err:
  * \retval VM_FAULT_ERROR on general error
  * \retval NOPAGE_OOM not have memory for allocate new page
  */
-int ll_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+int ll_fault0(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
         struct lu_env           *env;
         struct cl_io            *io;
@@ -270,24 +270,24 @@ int ll_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
         int                      fault_ret = 0;
         ENTRY;
 
-        result = ll_fault_io_init(vma, &env,  &nest, vmf->pgoff, &ra_flags);
-        if (env == NULL)
+        io = ll_fault_io_init(vma, &env,  &nest, vmf->pgoff, &ra_flags);
+        if (IS_ERR(io))
                 RETURN(VM_FAULT_ERROR);
 
-        io = &ccc_env_info(env)->cti_io;
+        result = io->ci_result;
         if (result < 0)
                 goto out_err;
 
         vio = vvp_env_io(env);
-
         vio->u.fault.ft_vma       = vma;
+        vio->u.fault.ft_vmpage    = NULL;
         vio->u.fault.fault.ft_vmf = vmf;
 
         result = cl_io_loop(env, io);
         fault_ret = vio->u.fault.fault.ft_flags;
 
 out_err:
-        if (result != 0)
+        if ((result != 0) && !(fault_ret & VM_FAULT_RETRY))
                 fault_ret |= VM_FAULT_ERROR;
 
         vma->vm_flags |= ra_flags;
@@ -298,6 +298,39 @@ out_err:
         RETURN(fault_ret);
 }
 
+int ll_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+        int count = 0;
+        bool printed = false;
+        int result;
+
+restart:
+        result = ll_fault0(vma, vmf);
+        LASSERT(!(result & VM_FAULT_LOCKED));
+        if (result == 0) {
+                struct page *vmpage = vmf->page;
+
+                /* check if this page has been truncated */
+                lock_page(vmpage);
+                if (unlikely(vmpage->mapping == NULL)) { /* unlucky */
+                        unlock_page(vmpage);
+                        page_cache_release(vmpage);
+                        vmf->page = NULL;
+
+                        if (!printed && ++count > 16) {
+                                CWARN("the page is under heavy contention,"
+                                      "maybe your app(%s) needs revising :-)\n",
+                                      current->comm);
+                                printed = true;
+                        }
+
+                        goto restart;
+                }
+
+                result |= VM_FAULT_LOCKED;
+        }
+        return result;
+}
 #endif
 
 /**