libglusterfs: stop special casing "cache-size" in size_t validation

The original situation was as follows:

The function that validates xlator options indicating a size,
xlator_option_validate_sizet(), handles the case when the name
of the option is "cache-size" in a special way.

- Xlator options (things of type volume_option_t) has a
  min and max attribute of type double.
- An xlator option is endowed with a gluster specific type (not
  C type). An instance of an xlator option goes through a validation
  process by a type specific validator function (which are collected
  in option.c).
- Validators of numeric types - size being one of them - make use the
  min and max attributes to perform a range check, except in one case:
  if an option is defined with min = max = 0, then this option will be
  exempt of range checking. (Note: the volume_option_t definition
  features the following comments along the min, max fields:

   double                  min;  /* 0 means no range */
   double                  max;  /* 0 means no range */

  which is slightly misleading as it lets one to conclude that
  zeroing min or max buys exemption from low or high boundary check,
  which is not true -- only *both* being zero buys exemption.)
- Besides this, the validator for options of size type,
  xlator_option_validate_sizet() special cases options
  named "cache-size" so that only min is enforced. (The only consequence
  of a value exceeding max is that glusterd logs a warning about it, but
  the cli user who makes such a setting gets no feedback on it.)
- This was introduced because a hard coded limit is not useful for
  io-cache and quick-read. They rather use a runtime calculated
  upper limit. (See changes
  I7dd4d8c53051b89a293696abf1ee8dc237e39a20
  I9c744b5ace10604d5a814e6218ca0d83c796db80
  about the last two points.)
- As an unintended consequence, the upper limit check of
  cache-size of write-behind, for which a conventional hard coded limit
  is specified, is defeated.

What we do about it:

- Remove the special casing clause for cache-size in
  xlator_option_validate_sizet. Thus the general range
  check policy (as described above) will apply to
  cache-size too.
- To implement a lower bound only check by the validator
  for cache-size of io-cache and quick-read, change the
  max attribute of these options to INFINITY.

The only behavioral difference is the omission of the warnings
about cache-size of io-cache and quick-read exceeding the former max
values. (They were rather heuristic anyway.)

BUG: 1445609
Change-Id: I0bd8bd391fa7d926f76e214a2178833fe4673b4a
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: https://review.gluster.org/17125
Smoke: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Amar Tumballi <amarts@redhat.com>
Tested-by: Raghavendra G <rgowdapp@redhat.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>

1 files changed, 2 insertions, 1 deletions

diff --git a/xlators/performance/quick-read/src/quick-read.c b/xlators/performance/quick-read/src/quick-read.c
index 4e9c6dcf091..8edf495fcdd 100644
--- a/xlators/performance/quick-read/src/quick-read.c
+++ b/xlators/performance/quick-read/src/quick-read.c
@@ -8,6 +8,7 @@
   cases as published by the Free Software Foundation.
 */
 
+#include <math.h>
 #include "quick-read.h"
 #include "statedump.h"
 #include "quick-read-messages.h"
@@ -1144,7 +1145,7 @@ struct volume_options options[] = {
         { .key  = {"cache-size"},
           .type = GF_OPTION_TYPE_SIZET,
           .min  = 0,
-          .max  = 32 * GF_UNIT_GB,
+          .max  = INFINITY,
           .default_value = "128MB",
           .description = "Size of the read cache."
         },