[coreutils] Patch FIEMAP support for effieicent sparse file copy v1

[jeff.liu]

Add fiemap_copy() to src/copy.c

>From 2d3c01611ef6b2c9255c501b0c882ccb98086921 Mon Sep 17 00:00:00 2001
From: Jie Liu <address@hidden>
Date: Sat, 20 Mar 2010 20:40:57 +0800
Subject: [PATCH] Introduce FIEMAP sparse file efficient copy.

Signed-off-by: Jie Liu <address@hidden>
---
 src/copy.c |  135 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-
 src/copy.h |   27 ++++++++++++
 src/cp.c   |   53 +++++++++++++++++++++++-
 3 files changed, 212 insertions(+), 3 deletions(-)

diff --git a/src/copy.c b/src/copy.c
index 29f37c9..d16ea13 100644
--- a/src/copy.c
+++ b/src/copy.c
@@ -65,6 +65,10 @@
 # include <sys/ioctl.h>
 #endif

+#ifndef HAVE_FIEMAP
+# include "fiemap.h"
+#endif
+
 #ifndef HAVE_FCHOWN
 # define HAVE_FCHOWN false
 # define fchown(fd, uid, gid) (-1)
@@ -151,6 +155,104 @@ clone_file (int dest_fd, int src_fd)
 #endif
 }

+#ifdef __linux__
+# undef FS_IOC_FIEMAP
+# define FS_IOC_FIEMAP _IOWR('f', 11, struct fiemap)
+/* Perform FIEMAP(available in mainline 2.6.27) copy for sparse file backup.
+   Call ioctl(2) with FS_IOC_FIEMAP will return the number of extents and the
+   extent information mapped to a file, it does not includes the holes. so the
+   overhead deal with holes with lseek(2) could be saved.
+   This would result in much faster backups for any kind of sparse file.  */
+static bool
+fiemap_copy (int src_fd, int dest_fd, size_t optimal_buf_size,
+             uint32_t fiemap_flags, char const *src_name,
+             char const *dst_name)
+{
+  int last = 0;
+  unsigned int i;
+  bool return_val = true;
+  char fiemap_buf[4096] = "";
+  struct fiemap *fiemap = (struct fiemap *)fiemap_buf;
+  struct fiemap_extent *fm_ext = &fiemap->fm_extents[0];
+  uint32_t count = (sizeof(fiemap_buf) - sizeof(*fiemap)) /
+                    sizeof(struct fiemap_extent);
+
+  memset (fiemap, 0, sizeof (*fiemap));
+
+  do
+    {
+      fiemap->fm_start = 0ULL;
+      fiemap->fm_length = FIEMAP_MAX_OFFSET;
+      fiemap->fm_flags |= fiemap_flags;
+      fiemap->fm_extent_count = count;
+
+      /* If the underlaying filesystem does not support FIEMAP or
+         the flags specified, fall back to do normal copy if the
+         fiemap_mod == FIEMAP_AUTO.  */
+      if (ioctl (src_fd, FS_IOC_FIEMAP, (unsigned long) fiemap) < 0)
+          return false;
+
+      /* If 0 extents are returned, then more ioctls are not needed.  */
+      if (fiemap->fm_mapped_extents == 0)
+        {
+          return true;
+       }
+
+      for (i = 0; i < fiemap->fm_mapped_extents; i++)
+        {
+          uint64_t ext_logical = fm_ext[i].fe_logical;
+          uint64_t ext_len = fm_ext[i].fe_length;
+
+          /* FIXME: do we need to deal with the 
FIEMAP_EXTENT_UNKNOWN/FIEMAP_EXTENT_ENCODED?  */
+
+          if (lseek (src_fd, ext_logical, SEEK_CUR) < 0LL)
+            {
+              error (0, errno, _("cannot lseek %s"), quote (src_name));
+              return_val = false;
+            }
+
+          uint64_t tot_read;
+          char buf[optimal_buf_size];
+          while (tot_read < ext_len)
+            {
+             memset (buf, 0, sizeof(buf));
+             ssize_t n_read = read (src_fd, buf, optimal_buf_size);
+              if (n_read < 0)
+                {
+#ifdef EINTR
+                  if (errno == EINTR)
+                    continue;
+#endif
+                  error (0, errno, _("reading %s"), quote (src_name));
+                  return_val = false;
+                }
+
+              if (n_read == 0)
+                break;
+
+              if (full_write (dest_fd, buf, n_read) != n_read)
+                {
+                  error (0, errno, _("writing %s"), quote (dst_name));
+                  return_val = false;
+                }
+
+              tot_read += n_read;
+            }
+
+          if (fm_ext[i].fe_flags & FIEMAP_EXTENT_LAST)
+            last = 1;
+
+          fiemap->fm_start = (fm_ext[i-1].fe_logical + fm_ext[i-1].fe_length);
+        }
+
+    } while (last == 0);
+
+  return return_val;
+}
+#else
+static bool fiemap_copy (ignored) { errno == ENOTSUP; return false; }
+#endif
+
 /* FIXME: describe */
 /* FIXME: rewrite this to use a hash table so we avoid the quadratic
    performance hit that's probably noticeable only on trees deeper
@@ -703,6 +805,31 @@ copy_reg (char const *src_name, char const *dst_name,
             buf_size = blcm;
         }

+    if (x->fiemap_mode)
+      {
+        uint32_t fiemap_flags = 0;
+
+        if (x->fiemap_sync)
+          fiemap_flags |= FIEMAP_FLAG_SYNC;
+        if (x->fiemap_xattr)
+          fiemap_flags |= FIEMAP_FLAG_XATTR;
+
+        bool fiemap_copy_ok = fiemap_copy (source_desc, dest_desc, buf_size,
+                                           fiemap_flags, src_name, dst_name);
+        if (fiemap_copy_ok)
+          goto preserve_extra_info;
+        else
+          {
+#ifdef EBADR
+            if ((errno == EBADR) && (x->fiemap_mode == FIEMAP_AUTO))
+                goto normal_copy;
+#endif
+            error (0, errno, _("FIEMAP copy failed %s"), quote (src_name));
+            goto close_src_and_dst_desc;
+         }
+      }
+
+normal_copy:
       /* Make a buffer with space for a sentinel at the end.  */
       buf_alloc = xmalloc (buf_size + buf_alignment_slop);
       buf = ptr_align (buf_alloc, buf_alignment);
@@ -813,6 +940,7 @@ copy_reg (char const *src_name, char const *dst_name,
         }
     }

+preserve_extra_info:
   if (x->preserve_timestamps)
     {
       struct timespec timespec[2];
@@ -901,8 +1029,11 @@ close_src_desc:
       return_val = false;
     }

-  free (buf_alloc);
-  free (name_alloc);
+  if (buf_alloc)
+    free (buf_alloc);
+  if (name_alloc)
+    free (name_alloc);
+
   return return_val;
 }

diff --git a/src/copy.h b/src/copy.h
index bd7359f..0bd407d 100644
--- a/src/copy.h
+++ b/src/copy.h
@@ -56,6 +56,19 @@ enum Reflink_type
   REFLINK_ALWAYS
 };

+/* Control of FIEMAP copy.  */
+enum Fiemap_type
+{
+  /* Default to a standard copy.  */
+  FIEMAP_NEVER,
+
+  /* Try a FIEMAP copy and fall back to a standard copy.  */
+  FIEMAP_AUTO,
+
+  /* Require a FIEMAP copy and fail if not available.  */
+  FIEMAP_ALWAYS
+};
+
 /* This type is used to help mv (via copy.c) distinguish these cases.  */
 enum Interactive
 {
@@ -91,6 +104,11 @@ enum Dereference_symlink
    || (Mode) == REFLINK_AUTO           \
    || (Mode) == REFLINK_ALWAYS)

+# define VALID_FIEMAP_MODE(Mode)        \
+  ((Mode) == FIEMAP_NEVER               \
+   || (Mode) == FIEMAP_AUTO             \
+   || (Mode) == FIEMAP_ALWAYS)
+
 /* These options control how files are copied by at least the
    following programs: mv (when rename doesn't work), cp, install.
    So, if you add a new member, be sure to initialize it in
@@ -237,9 +255,18 @@ struct cp_options
      such a symlink) and returns false.  */
   bool open_dangling_dest_symlink;

+  /* If true, set fiemap ioctl flags with FIEMAP_FLAG_SYNC.  */
+  bool fiemap_sync;
+
+  /* If true, set fiemap ioctl flags with FIEMAP_FLAG_XATTR.  */
+  bool fiemap_xattr;
+
   /* Control creation of COW files.  */
   enum Reflink_type reflink_mode;

+  /* Control of FIEMAP type file copy.  */
+  enum Fiemap_type fiemap_mode;
+
   /* This is a set of destination name/inode/dev triples.  Each such triple
      represents a file we have created corresponding to a source file name
      that was specified on the command line.  Use it to avoid clobbering
diff --git a/src/cp.c b/src/cp.c
index cc958d1..c551b66 100644
--- a/src/cp.c
+++ b/src/cp.c
@@ -78,8 +78,11 @@ enum
   PRESERVE_ATTRIBUTES_OPTION,
   REFLINK_OPTION,
   SPARSE_OPTION,
+  FIEMAP_OPTION,
   STRIP_TRAILING_SLASHES_OPTION,
-  UNLINK_DEST_BEFORE_OPENING
+  UNLINK_DEST_BEFORE_OPENING,
+  FIEMAP_FLAG_SYNC_OPTION,
+  FIEMAP_FLAG_XATTR_OPTION
 };

 /* True if the kernel is SELinux enabled.  */
@@ -112,6 +115,16 @@ static enum Reflink_type const reflink_type[] =
 };
 ARGMATCH_VERIFY (reflink_type_string, reflink_type);

+static char const *const fiemap_type_string[] =
+{
+  "auto", "always", NULL
+};
+static enum Fiemap_type const fiemap_type[] =
+{
+  FIEMAP_AUTO, FIEMAP_ALWAYS
+};
+ARGMATCH_VERIFY (fiemap_type_string, fiemap_type);
+
 static struct option const long_opts[] =
 {
   {"archive", no_argument, NULL, 'a'},
@@ -133,6 +146,9 @@ static struct option const long_opts[] =
   {"remove-destination", no_argument, NULL, UNLINK_DEST_BEFORE_OPENING},
   {"sparse", required_argument, NULL, SPARSE_OPTION},
   {"reflink", optional_argument, NULL, REFLINK_OPTION},
+  {"fiemap", optional_argument, NULL, FIEMAP_OPTION},
+  {"fiemap-sync", optional_argument, NULL, FIEMAP_FLAG_SYNC_OPTION},
+  {"fiemap-xattr", optional_argument, NULL, FIEMAP_FLAG_XATTR_OPTION},
   {"strip-trailing-slashes", no_argument, NULL, STRIP_TRAILING_SLASHES_OPTION},
   {"suffix", required_argument, NULL, 'S'},
   {"symbolic-link", no_argument, NULL, 's'},
@@ -212,6 +228,11 @@ Mandatory arguments to long options are mandatory for 
short options too.\n\
                                  argument\n\
 "), stdout);
       fputs (_("\
+      --fiemap=WHEN            control creation of sparse files. See below\n\
+      --fiemap-sync            sync file data before fiemap\n\
+      --fiemap-xattr           map extended attribute tree\n\
+"), stdout);
+      fputs (_("\
   -s, --symbolic-link          make symbolic links instead of copying\n\
   -S, --suffix=SUFFIX          override the usual backup suffix\n\
   -t, --target-directory=DIRECTORY  copy all SOURCE arguments into DIRECTORY\n\
@@ -237,6 +258,10 @@ Use --sparse=never to inhibit creation of sparse files.\n\
 When --reflink[=always] is specified, perform a lightweight copy, where the\n\
 data blocks are copied only when modified.  If this is not possible the copy\n\
 fails, or if --reflink=auto is specified, fall back to a standard copy.\n\
+\n\
+When --fiemap[=always] is specified, perform a fiemap copy, where the\n\
+allocated data blocks are copied except holes.  If this is not possible the\n\
+copy fails, or if --fiemap=auto is specified, fall back to a standard copy.\n\
 "), stdout);
       fputs (_("\
 \n\
@@ -770,6 +795,10 @@ cp_option_init (struct cp_options *x)
   x->move_mode = false;
   x->one_file_system = false;
   x->reflink_mode = REFLINK_NEVER;
+  x->fiemap_mode = FIEMAP_NEVER;
+
+  x->fiemap_sync = false;
+  x->fiemap_xattr = false;

   x->preserve_ownership = false;
   x->preserve_links = false;
@@ -942,6 +971,22 @@ main (int argc, char **argv)
                                        reflink_type_string, reflink_type);
           break;

+        case FIEMAP_OPTION:
+          if (optarg == NULL)
+            x.fiemap_mode = FIEMAP_ALWAYS;
+          else
+            x.fiemap_mode = XARGMATCH ("--fiemap", optarg,
+                                       fiemap_type_string, fiemap_type);
+          break;
+
+        case FIEMAP_FLAG_SYNC_OPTION:
+          x.fiemap_sync = true;
+          break;
+
+        case FIEMAP_FLAG_XATTR_OPTION:
+          x.fiemap_xattr = true;
+          break;
+
         case 'a':              /* Like -dR --preserve=all with reduced failure 
diagnostics. */
           x.dereference = DEREF_NEVER;
           x.preserve_links = true;
@@ -1108,6 +1153,12 @@ main (int argc, char **argv)
       usage (EXIT_FAILURE);
     }

+  if (x.fiemap_mode == FIEMAP_ALWAYS && x.sparse_mode != SPARSE_NEVER)
+    {
+      error (0, 0, _("--fiemap can be used only with --sparse=NEVER"));
+      usage (EXIT_FAILURE);
+    }
+
   if (backup_suffix_string)
     simple_backup_suffix = xstrdup (backup_suffix_string);

-- 
1.5.4.3