Re: [patch] -sparse predicate

[dean gaudet]

On Sat, 11 Nov 2006, James Youngman wrote:

> Actually, the units in which st_blocks is measures are not st_blksize.
> All that st_blksize tells you is a "preferred" IO request size for
> efficient I/O.  See the RATIONALE section of
> http://www.opengroup.org/onlinepubs/009695399/basedefs/sys/stat.h.html

oh yeah, i should have known that.


> Anyway, that problem is easily fixed I think, since gnulib already has
> a way of figuring out the system's block size.

ok... grabbed cvs findutils and cvs gnulib... hmm... are you thinking of 
the get_fs_usage() operation?  that's the only one i see using 
statfs-related calls.


> Thanks again for the contribution.   At a guess, I would say that it's
> likely to make it into findutils 4.3.2.   It's unlikely to appear in
> 4.2.x because I'm pretty much trying to limit changes in those
> releases to just bugfixes.

that's cool!

so, i've updated the patch using get_fs_usage.  i left the disk parameter 
NULL, which means it'll fail on ultrix... it seems maybe gnulib needs a 
get_fs_blocksize or something which isn't so heavyweight.

i noticed there was already precedence for a simple one filesystem cache 
in the filesystem_type code... so i've done that here as well.

thanks
-dean

diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/doc/find.texi findutils/doc/find.texi
--- findutils.orig/doc/find.texi        2006-08-20 13:39:48.000000000 -0700
+++ findutils/doc/find.texi     2006-11-11 15:49:43.268004118 -0800
@@ -924,6 +924,12 @@ useful with @samp{-depth} (@pxref{Direct
 (@pxref{Single File}).
 @end deffn
 
address@hidden Test -sparse
+True if the file is a regular file and has a size larger than the
+number of blocks allocated multiplied by the blocksize.  This generally
+occurs if the file is sparse.
address@hidden deffn
+
 @node Type
 @section Type
 
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/defs.h findutils/find/defs.h
--- findutils.orig/find/defs.h  2006-08-20 11:18:42.000000000 -0700
+++ findutils/find/defs.h       2006-11-11 15:49:43.260001599 -0800
@@ -513,6 +513,7 @@ boolean pred_readable PARAMS((char *path
 boolean pred_regex PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
 boolean pred_samefile PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
 boolean pred_size PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
+boolean pred_sparse PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
 boolean pred_true PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
 boolean pred_type PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
 boolean pred_uid PARAMS((char *pathname, struct stat *stat_buf, struct 
predicate *pred_ptr));
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/find.1 findutils/find/find.1
--- findutils.orig/find/find.1  2006-08-20 13:39:48.000000000 -0700
+++ findutils/find/find.1       2006-11-11 15:49:43.272005379 -0800
@@ -520,6 +520,10 @@ sparse files that are not actually alloc
 differently.  The `b' suffix always denotes 512-byte blocks and never
 1 Kilobyte blocks, which is different to the behaviour of \-ls.
 
+.IP \-sparse
+True if the file is a regular file and has a size larger than the
+number of blocks allocated multiplied by the blocksize.  This generally
+occurs if the file is sparse.
 .IP \-true
 Always true.
 .IP "\-type \fIc\fR"
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/parser.c findutils/find/parser.c
--- findutils.orig/find/parser.c        2006-08-21 15:35:13.000000000 -0700
+++ findutils/find/parser.c     2006-11-11 15:49:57.932623506 -0800
@@ -136,6 +136,7 @@ static boolean parse_samefile      PARAM
 static boolean parse_show_control_chars PARAMS((const struct parser_table*, 
char *argv[], int *arg_ptr));
 #endif
 static boolean parse_size          PARAMS((const struct parser_table*, char 
*argv[], int *arg_ptr));
+static boolean parse_sparse        PARAMS((const struct parser_table*, char 
*argv[], int *arg_ptr));
 static boolean parse_time          PARAMS((const struct parser_table*, char 
*argv[], int *arg_ptr));
 static boolean parse_true          PARAMS((const struct parser_table*, char 
*argv[], int *arg_ptr));
 static boolean parse_type          PARAMS((const struct parser_table*, char 
*argv[], int *arg_ptr));
@@ -270,6 +271,7 @@ static struct parser_table const parse_t
   PARSE_OPTION     ("show-control-chars",    show_control_chars), /* GNU, 
4.3.0+ */
 #endif
   PARSE_TEST       ("size",                  size),
+  PARSE_TEST       ("sparse",                sparse),       /* GNU */
   PARSE_TEST       ("type",                  type),
   PARSE_TEST       ("uid",                   uid),          /* GNU */
   PARSE_TEST       ("used",                  used),         /* GNU */
@@ -1609,6 +1611,15 @@ parse_regextype (const struct parser_tab
   return parse_noop(entry, argv, arg_ptr);
 }
 
+static boolean
+parse_sparse (const struct parser_table* entry, char **argv, int *arg_ptr)
+{
+  (void) argv;
+  (void) arg_ptr;
+
+  insert_primary (entry);
+  return true;
+}
 
 static boolean
 parse_regex (const struct parser_table* entry, char **argv, int *arg_ptr)
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/pred.c findutils/find/pred.c
--- findutils.orig/find/pred.c  2006-11-07 23:28:50.000000000 -0800
+++ findutils/find/pred.c       2006-11-11 16:28:36.193040437 -0800
@@ -38,6 +38,7 @@
 #include "buildcmd.h"
 #include "yesno.h"
 #include "listfile.h"
+#include "fsusage.h"
 
 #if ENABLE_NLS
 # include <libintl.h>
@@ -213,6 +214,7 @@ struct pred_assoc pred_table[] =
   {pred_regex, "regex   "},
   {pred_samefile,"samefile "},
   {pred_size, "size    "},
+  {pred_sparse, "sparse  "},
   {pred_true, "true    "},
   {pred_type, "type    "},
   {pred_uid, "uid     "},
@@ -1355,6 +1357,31 @@ pred_size (char *pathname, struct stat *
 }
 
 boolean
+pred_sparse (char *pathname, struct stat *stat_buf, struct predicate *pred_ptr)
+{
+  static dev_t cur_dev = -1;
+  static struct fs_usage fsu;
+
+  (void) pathname;
+  (void) pred_ptr;
+
+  if (!S_ISREG (stat_buf->st_mode))
+    return false;
+
+  if (stat_buf->st_dev != cur_dev) {
+    if (get_fs_usage(state.rel_pathname, NULL, &fsu)) {
+      // could produce error here, but it's likely to occur every attempt
+      // on the same filesystem, and that would be quite spammy.
+      //error (0, errno, "get_fs_usage(%s)", pathname);
+      return false;
+    }
+    cur_dev = stat_buf->st_dev;
+  }
+
+  return stat_buf->st_size > stat_buf->st_blocks * fsu.fsu_blocksize;
+}
+
+boolean
 pred_samefile (char *pathname, struct stat *stat_buf, struct predicate 
*pred_ptr)
 {
   /* Potential optimisation: because of the loop protection, we always
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/tree.c findutils/find/tree.c
--- findutils.orig/find/tree.c  2006-01-04 11:22:38.000000000 -0800
+++ findutils/find/tree.c       2006-11-11 16:04:27.645317216 -0800
@@ -1033,6 +1033,7 @@ static struct pred_cost_lookup costlooku
     { pred_regex     ,  NeedsNothing         },
     { pred_samefile  ,  NeedsStatInfo        },
     { pred_size      ,  NeedsStatInfo        },
+    { pred_sparse    ,  NeedsStatInfo        },
     { pred_true             ,  NeedsNothing         },
     { pred_type      ,  NeedsType            },
     { pred_uid       ,  NeedsStatInfo        },
diff -pru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/import-gnulib.sh findutils/import-gnulib.sh
--- findutils.orig/import-gnulib.sh     2006-08-20 11:18:42.000000000 -0700
+++ findutils/import-gnulib.sh  2006-11-11 15:48:53.084196480 -0800
@@ -53,7 +53,7 @@ build-aux/texinfo.tex
 
 # Modules needed for findutils.
 findutils_modules="\
-alloca argmatch dirname error fileblocks fnmatch-gnu fopen-safer fts \
+alloca argmatch dirname error fileblocks fnmatch-gnu fopen-safer fsusage fts \
 getline getopt human idcache lstat malloc memcmp memset mktime \
 modechange pathmax quotearg realloc regex rpmatch savedir \
 stpcpy strdup strftime  strstr strtol strtoul strtoull strtoumax  \
diff -Npru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/testsuite/find.gnu/sparse.exp 
findutils/find/testsuite/find.gnu/sparse.exp
--- findutils.orig/find/testsuite/find.gnu/sparse.exp   1969-12-31 
16:00:00.000000000 -0800
+++ findutils/find/testsuite/find.gnu/sparse.exp        2006-11-11 
15:49:43.272005379 -0800
@@ -0,0 +1,9 @@
+# tests for -sparse
+exec rm -rf tmp
+exec mkdir tmp
+exec echo hi > tmp/notsparse
+# note that some filesystems (such as XFS) won't create a sparse file when the
+# "holes" are too small.  hopefully this hole is large enough.
+exec dd if=/dev/zero of=tmp/sparse seek=10000 bs=4096 count=1 2>/dev/null
+find_start p { tmp -sparse }
+exec rm -rf tmp
diff -Npru -xCVS -xgnulib -xautom4te.cache -xMakefile.in 
findutils.orig/find/testsuite/find.gnu/sparse.xo 
findutils/find/testsuite/find.gnu/sparse.xo
--- findutils.orig/find/testsuite/find.gnu/sparse.xo    1969-12-31 
16:00:00.000000000 -0800
+++ findutils/find/testsuite/find.gnu/sparse.xo 2006-11-11 15:49:43.272005379 
-0800
@@ -0,0 +1 @@
+tmp/sparse