summaryrefslogtreecommitdiffstats
path: root/xlators/features/bit-rot/src
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/features/bit-rot/src')
-rw-r--r--xlators/features/bit-rot/src/bitd/bit-rot-scrub.c373
-rw-r--r--xlators/features/bit-rot/src/bitd/bit-rot-scrub.h6
-rw-r--r--xlators/features/bit-rot/src/bitd/bit-rot.c70
-rw-r--r--xlators/features/bit-rot/src/bitd/bit-rot.h32
-rw-r--r--xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h1
5 files changed, 302 insertions, 180 deletions
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
index 390148fdb06..26ad97a16e8 100644
--- a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
@@ -508,198 +508,217 @@ br_fsscanner_handle_entry (xlator_t *subvol,
return -1;
}
-/*scrubber frequency tunable value in second (day * hour * minut * second)*/
-#define DAILY (1*24*60*60)
-#define WEEKLY (7*24*60*60)
-#define BIWEEKLY (14*24*60*60)
-#define MONTHLY (30*24*60*60)
-
-struct timeval
-br_scrubber_calc_freq (xlator_t *this)
+static inline void
+br_fsscanner_log_time (xlator_t *this, br_child_t *child, const char *sfx)
{
- br_private_t *priv = NULL;
- struct timeval scrub_sec = {0,};
+ struct timeval tv = {0,};
+ char timestr[1024] = {0,};
- priv = this->private;
+ gettimeofday (&tv, NULL);
+ gf_time_fmt (timestr, sizeof (timestr), tv.tv_sec, gf_timefmt_FT);
- /*By default scrubber frequency will be biweekly*/
- if (!strncmp (priv->scrub_freq, "daily", strlen("daily"))) {
- scrub_sec.tv_sec = DAILY;
- } else if (!strncmp (priv->scrub_freq, "weekly", strlen("weekly"))) {
- scrub_sec.tv_sec = WEEKLY;
- } else if (!strncmp (priv->scrub_freq, "monthly", strlen("monthly"))) {
- scrub_sec.tv_sec = MONTHLY;
- } else if (!strncmp (priv->scrub_freq, "biweekly",
- strlen("biweekly"))) {
- scrub_sec.tv_sec = BIWEEKLY;
- } else {
- gf_log (this->name, GF_LOG_ERROR, "Invalid scrub-frequency %s"
- "value.", priv->scrub_freq);
- scrub_sec.tv_sec = -1;
- }
+ gf_log (this->name, GF_LOG_INFO,
+ "Scrubbing \"%s\" %s at %s", child->brick_path, sfx, timestr);
+}
- return scrub_sec;
+static void
+br_fsscanner_wait_until_kicked (struct br_scanfs *fsscan)
+{
+ pthread_mutex_lock (&fsscan->wakelock);
+ {
+ while (!fsscan->kick)
+ pthread_cond_wait (&fsscan->wakecond,
+ &fsscan->wakelock);
+ fsscan->kick = _gf_false;
+ }
+ pthread_mutex_unlock (&fsscan->wakelock);
}
-#define SCRUBBER_SLEEP(freq_diff, elapse_time) do { \
- \
- if (freq_diff < 0) { \
- return 0; \
- } else if (freq_diff <= DAILY) { \
- gf_log (this->name, GF_LOG_INFO, \
- "Scrubber is sleeping for %ld " \
- "sec", freq_diff); \
- sleep (freq_diff); \
- return 0; \
- } else { \
- gf_log (this->name, GF_LOG_INFO, \
- "Scrubber is sleeping for %ld " \
- "sec", freq_diff); \
- sleep (DAILY); \
- elapse_time += DAILY; \
- } \
- } while (0)
-
-static int
-br_scrubber_sleep_check (struct timeval *begin, struct timeval *end,
- xlator_t *this)
+void *
+br_fsscanner (void *arg)
{
- br_private_t *priv = NULL;
- struct timeval elapse_time = {0,};
- struct timeval freq_diff = {0,};
- struct timeval scrub_sec = {0,};
- struct timeval temp = {0,};
+ loc_t loc = {0,};
+ br_child_t *child = NULL;
+ xlator_t *this = NULL;
+ br_private_t *priv = NULL;
+ struct br_scanfs *fsscan = NULL;
+ struct br_scrubber *fsscrub = NULL;
+ child = arg;
+ this = child->this;
priv = this->private;
- scrub_sec = br_scrubber_calc_freq (this);
- if (scrub_sec.tv_sec == -1) {
- gf_log (this->name, GF_LOG_ERROR, "Unable to calculate scrub "
- "frequency %s value", priv->scrub_freq);
- return -1;
- }
+ fsscan = &child->fsscan;
+ fsscrub = &priv->fsscrub;
- if ((end->tv_sec - begin->tv_sec) < scrub_sec.tv_sec) {
- /* Sleep, if scrubber have completed its job before schedule
- * scrub frequency based on current scrub frequency value */
- do {
- scrub_sec = br_scrubber_calc_freq (this);
- freq_diff.tv_sec = scrub_sec.tv_sec - (end->tv_sec -
- begin->tv_sec) -
- elapse_time.tv_sec;
- SCRUBBER_SLEEP(freq_diff.tv_sec, elapse_time.tv_sec);
- } while (1);
-
-
- } else {
- /* Sleep, if scrubber have completed its job after schedule
- * scrub frequency based on current scrub frequency value */
- temp.tv_sec = (end->tv_sec - begin->tv_sec) % scrub_sec.tv_sec;
- if (temp.tv_sec != 0) {
- do {
- scrub_sec = br_scrubber_calc_freq (this);
- freq_diff.tv_sec = scrub_sec.tv_sec
- - temp.tv_sec
- - elapse_time.tv_sec;
- SCRUBBER_SLEEP(freq_diff.tv_sec,
- elapse_time.tv_sec);
- } while (1);
+ THIS = this;
+ loc.inode = child->table->root;
+
+ while (1) {
+ br_fsscanner_wait_until_kicked (fsscan);
+ {
+ /* log start time */
+ br_fsscanner_log_time (this, child, "started");
+
+ /* scrub */
+ (void) syncop_ftw (child->xl,
+ &loc, GF_CLIENT_PID_SCRUB,
+ child, br_fsscanner_handle_entry);
+ if (!list_empty (&fsscan->queued))
+ wait_for_scrubbing (this, fsscan);
+
+ /* log finish time */
+ br_fsscanner_log_time (this, child, "finished");
}
+ br_fsscan_reschedule (this, child, fsscan, fsscrub, _gf_false);
}
- return 0;
+ return NULL;
}
-void *
-br_fsscanner (void *arg)
+void
+br_kickstart_scanner (struct gf_tw_timer_list *timer,
+ void *data, unsigned long calltime)
{
- int32_t ret = -1;
- loc_t loc = {0,};
- char timestr[1024] = {0,};
- xlator_t *this = NULL;
- br_child_t *child = NULL;
- struct br_scanfs *fsscan = NULL;
- br_private_t *priv = NULL;
- struct timeval elapse_time = {0,};
- struct timeval scrub_sec = {0,};
- struct timeval freq_diff = {0,};
+ xlator_t *this = NULL;
+ br_child_t *child = data;
+ struct br_scanfs *fsscan = NULL;
- child = arg;
- this = child->this;
+ THIS = this = child->this;
fsscan = &child->fsscan;
- THIS = this;
+ /* kickstart scanning.. */
+ pthread_mutex_lock (&fsscan->wakelock);
+ {
+ fsscan->kick = _gf_true;
+ pthread_cond_signal (&fsscan->wakecond);
+ }
+ pthread_mutex_unlock (&fsscan->wakelock);
+
+ return;
+
+}
+
+static inline uint32_t
+br_fsscan_calculate_delta (uint32_t boot, uint32_t now, uint32_t times)
+{
+ uint32_t secs = 0;
+ uint32_t diff = 0;
+
+ diff = (now - boot);
+ secs = times * ((diff / times) + 1);
+
+ return (secs - diff);
+}
+
+#define BR_SCRUB_HOURLY (60 * 60)
+#define BR_SCRUB_DAILY (1 * 24 * 60 * 60)
+#define BR_SCRUB_WEEKLY (7 * 24 * 60 * 60)
+#define BR_SCRUB_BIWEEKLY (14 * 24 * 60 * 60)
+#define BR_SCRUB_MONTHLY (30 * 24 * 60 * 60)
+
+static unsigned int
+br_fsscan_calculate_timeout (uint32_t boot, uint32_t now, scrub_freq_t freq)
+{
+ uint32_t timo = 0;
+
+ switch (freq) {
+ case BR_FSSCRUB_FREQ_HOURLY:
+ timo = br_fsscan_calculate_delta (boot, now, BR_SCRUB_HOURLY);
+ break;
+ case BR_FSSCRUB_FREQ_DAILY:
+ timo = br_fsscan_calculate_delta (boot, now, BR_SCRUB_DAILY);
+ break;
+ case BR_FSSCRUB_FREQ_WEEKLY:
+ timo = br_fsscan_calculate_delta (boot, now, BR_SCRUB_WEEKLY);
+ break;
+ case BR_FSSCRUB_FREQ_BIWEEKLY:
+ timo = br_fsscan_calculate_delta (boot, now, BR_SCRUB_BIWEEKLY);
+ break;
+ case BR_FSSCRUB_FREQ_MONTHLY:
+ timo = br_fsscan_calculate_delta (boot, now, BR_SCRUB_MONTHLY);
+ }
+
+ return timo;
+}
+
+int32_t
+br_fsscan_schedule (xlator_t *this, br_child_t *child,
+ struct br_scanfs *fsscan, struct br_scrubber *fsscrub)
+{
+ uint32_t timo = 0;
+ br_private_t *priv = NULL;
+ struct timeval tv = {0,};
+ char timestr[1024] = {0,};
+ struct gf_tw_timer_list *timer = NULL;
priv = this->private;
- loc.inode = child->table->root;
+ (void) gettimeofday (&tv, NULL);
+ fsscan->boot = tv.tv_sec;
- /* Scrubber should start scrubbing the filesystem *after* the
- * schedueled scrub-frequency has expired.*/
- do {
- /*Calculate current scrub frequency value in second*/
- scrub_sec = br_scrubber_calc_freq (this);
- if (scrub_sec.tv_sec == -1) {
- gf_log (this->name, GF_LOG_ERROR, "Unable to calculate "
- "scrub frequency %s value", priv->scrub_freq);
- return NULL;
- }
+ timo = br_fsscan_calculate_timeout (fsscan->boot,
+ fsscan->boot, fsscrub->frequency);
- freq_diff.tv_sec = scrub_sec.tv_sec - elapse_time.tv_sec;
+ fsscan->timer = GF_CALLOC (1, sizeof (*fsscan->timer),
+ gf_br_stub_mt_br_scanner_freq_t);
+ if (!fsscan->timer)
+ goto error_return;
- if (freq_diff.tv_sec < 0) {
- break;
- } else if (freq_diff.tv_sec == DAILY) {
- sleep (DAILY);
- break;
- } else {
- sleep (DAILY);
- elapse_time.tv_sec += DAILY;
- }
+ timer = fsscan->timer;
+ INIT_LIST_HEAD (&timer->entry);
- } while (1);
+ timer->data = child;
+ timer->expires = timo;
+ timer->function = br_kickstart_scanner;
+ gf_tw_add_timer (priv->timer_wheel, timer);
- while (1) {
- /* log scrub start time */
- gettimeofday (&priv->tv_before_scrub, NULL);
- gf_time_fmt (timestr, sizeof timestr,
- priv->tv_before_scrub.tv_sec, gf_timefmt_FT);
- gf_log (this->name, GF_LOG_INFO,
- "Scrubbing \"%s\" started at %s",
- child->brick_path, timestr);
-
- /* scrub */
- (void) syncop_ftw (child->xl, &loc,
- GF_CLIENT_PID_SCRUB,
- child, br_fsscanner_handle_entry);
- if (!list_empty (&fsscan->queued))
- wait_for_scrubbing (this, fsscan);
-
- gettimeofday (&priv->tv_after_scrub, NULL);
- /* log scrub finish time */
- gf_time_fmt (timestr, sizeof timestr,
- priv->tv_after_scrub.tv_sec, gf_timefmt_FT);
+ gf_time_fmt (timestr, sizeof (timestr),
+ (fsscan->boot + timo), gf_timefmt_FT);
+ gf_log (this->name, GF_LOG_INFO, "Scrubbing for %s scheduled to "
+ "run at %s", child->brick_path, timestr);
+
+ return 0;
+
+ error_return:
+ return -1;
+}
+
+int32_t
+br_fsscan_reschedule (xlator_t *this,
+ br_child_t *child, struct br_scanfs *fsscan,
+ struct br_scrubber *fsscrub, gf_boolean_t pendingcheck)
+{
+ int32_t ret = 0;
+ uint32_t timo = 0;
+ char timestr[1024] = {0,};
+ struct timeval now = {0,};
+ br_private_t *priv = NULL;
+
+ priv = this->private;
+
+ (void) gettimeofday (&now, NULL);
+ timo = br_fsscan_calculate_timeout (fsscan->boot,
+ now.tv_sec, fsscrub->frequency);
+
+ gf_time_fmt (timestr, sizeof (timestr),
+ (now.tv_sec + timo), gf_timefmt_FT);
+
+ if (pendingcheck)
+ ret = gf_tw_mod_timer_pending (priv->timer_wheel,
+ fsscan->timer, timo);
+ else
+ ret = gf_tw_mod_timer (priv->timer_wheel, fsscan->timer, timo);
+
+ if (!ret && pendingcheck)
gf_log (this->name, GF_LOG_INFO,
- "Scrubbing \"%s\" finished at %s",
- child->brick_path, timestr);
-
- /* Scrubber should sleep if it have completed scrubbing
- * of filesystem before the scheduled scrub-frequency*/
- ret = br_scrubber_sleep_check (&priv->tv_before_scrub,
- &priv->tv_after_scrub,
- this);
- if (!ret) {
- gf_log (this->name, GF_LOG_DEBUG, "scrubber is crawling"
- " file system with scrubber frequency %s",
- priv->scrub_freq);
- } else {
- gf_log (this->name, GF_LOG_ERROR, "Unable to perform "
- "scrubber sleep check for scrubber frequency");
- return NULL;
- }
- }
+ "Scrubber for %s is currently running and would be "
+ "rescheduled after completion", child->brick_path);
+ else
+ gf_log (this->name, GF_LOG_INFO, "Scrubbing for %s rescheduled "
+ "to run at %s", child->brick_path, timestr);
- return NULL;
+ return 0;
}
#define BR_SCRUB_THREAD_SCALE_LAZY 0
@@ -1092,10 +1111,34 @@ static int32_t
br_scrubber_handle_freq (xlator_t *this, br_private_t *priv, dict_t *options)
{
int32_t ret = -1;
+ char *tmp = NULL;
+ scrub_freq_t frequency = BR_FSSCRUB_FREQ_HOURLY;
+ struct br_scrubber *fsscrub = NULL;
- ret = br_scrubber_fetch_option (this, "scrub-freq", options,
- &priv->scrub_freq);
- return ret;
+ fsscrub = &priv->fsscrub;
+
+ ret = br_scrubber_fetch_option (this, "scrub-freq", options, &tmp);
+ if (ret)
+ goto error_return;
+
+ if (strcasecmp (tmp, "hourly") == 0) {
+ frequency = BR_FSSCRUB_FREQ_HOURLY;
+ } else if (strcasecmp (tmp, "daily") == 0) {
+ frequency = BR_FSSCRUB_FREQ_DAILY;
+ } else if (strcasecmp (tmp, "weekly") == 0) {
+ frequency = BR_FSSCRUB_FREQ_WEEKLY;
+ } else if (strcasecmp (tmp, "biweekly") == 0) {
+ frequency = BR_FSSCRUB_FREQ_BIWEEKLY;
+ } else if (strcasecmp (tmp, "monthly") == 0) {
+ frequency = BR_FSSCRUB_FREQ_MONTHLY;
+ } else
+ goto error_return;
+
+ fsscrub->frequency = frequency;
+ return 0;
+
+ error_return:
+ return -1;
}
int32_t
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
index 4f00020d66a..6c4254a397a 100644
--- a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
@@ -16,6 +16,12 @@
void *br_fsscanner (void *);
+int32_t br_fsscan_schedule (xlator_t *, br_child_t *,
+ struct br_scanfs *, struct br_scrubber *);
+int32_t br_fsscan_reschedule (xlator_t *this,
+ br_child_t *child, struct br_scanfs *,
+ struct br_scrubber *, gf_boolean_t);
+
int32_t br_scrubber_handle_options (xlator_t *, br_private_t *, dict_t *);
int32_t br_scrubber_init (xlator_t *, br_private_t *);
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c
index 2652f02b4ea..e7cfe89e1dd 100644
--- a/xlators/features/bit-rot/src/bitd/bit-rot.c
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.c
@@ -1135,6 +1135,11 @@ br_enact_scrubber (xlator_t *this, br_child_t *child)
INIT_LIST_HEAD (&fsscan->queued);
INIT_LIST_HEAD (&fsscan->ready);
+ /* init scheduler related variables */
+ fsscan->kick = _gf_false;
+ pthread_mutex_init (&fsscan->wakelock, NULL);
+ pthread_cond_init (&fsscan->wakecond, NULL);
+
ret = gf_thread_create (&child->thread, NULL, br_fsscanner, child);
if (ret != 0) {
gf_log (this->name, GF_LOG_ALERT, "failed to spawn bitrot "
@@ -1142,6 +1147,10 @@ br_enact_scrubber (xlator_t *this, br_child_t *child)
goto error_return;
}
+ ret = br_fsscan_schedule (this, child, fsscan, fsscrub);
+ if (ret)
+ goto error_return;
+
/**
* Everything has been setup.. add this subvolume to scrubbers
* list.
@@ -1407,13 +1416,6 @@ br_init_signer (xlator_t *this, br_private_t *priv)
if (ret)
goto out;
- priv->timer_wheel = glusterfs_global_timer_wheel (this);
- if (!priv->timer_wheel) {
- gf_log (this->name, GF_LOG_ERROR,
- "global timer wheel unavailable");
- goto out;
- }
-
pthread_cond_init (&priv->object_cond, NULL);
priv->obj_queue = GF_CALLOC (1, sizeof (*priv->obj_queue),
@@ -1568,6 +1570,13 @@ init (xlator_t *this)
INIT_LIST_HEAD (&priv->children[i].list);
INIT_LIST_HEAD (&priv->bricks);
+ priv->timer_wheel = glusterfs_global_timer_wheel (this);
+ if (!priv->timer_wheel) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "global timer wheel unavailable");
+ goto cleanup_mutex;
+ }
+
this->private = priv;
if (!priv->iamscrubber) {
@@ -1633,12 +1642,55 @@ fini (xlator_t *this)
int
reconfigure (xlator_t *this, dict_t *options)
{
- br_private_t *priv = this->private;
+ int i = 0;
+ int32_t ret = -1;
+ br_child_t *child = NULL;
+ br_private_t *priv = NULL;
+ struct br_scanfs *fsscan = NULL;
+ struct br_scrubber *fsscrub = NULL;
+
+ priv = this->private;
if (!priv->iamscrubber)
return 0;
- return br_scrubber_handle_options (this, priv, options);
+ ret = br_scrubber_handle_options (this, priv, options);
+ if (ret)
+ goto err;
+
+ fsscrub = &priv->fsscrub;
+
+ /* reschedule all _up_ subvolume(s) */
+ pthread_mutex_lock (&priv->lock);
+ {
+ for (; i < priv->child_count; i++) {
+ child = &priv->children[i];
+ if (!child->child_up) {
+ gf_log (this->name, GF_LOG_INFO,
+ "Brick %s is offline, skipping "
+ "rescheduling (scrub would auto- "
+ "schedule when brick is back online).",
+ child->brick_path);
+ continue;
+ }
+
+ fsscan = &child->fsscan;
+ ret = br_fsscan_reschedule (this, child,
+ fsscan, fsscrub, _gf_true);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR, "Could not "
+ "reschedule scrubber for brick: %s. "
+ "Scubbing will continue according to "
+ "old frequency.", child->brick_path);
+ }
+ }
+ }
+ pthread_mutex_unlock (&priv->lock);
+
+ return 0;
+
+ err:
+ return -1;
}
struct xlator_fops fops;
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.h b/xlators/features/bit-rot/src/bitd/bit-rot.h
index 6543be763d6..ec943e9131f 100644
--- a/xlators/features/bit-rot/src/bitd/bit-rot.h
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.h
@@ -46,6 +46,14 @@ typedef enum scrub_throttle {
BR_SCRUB_THROTTLE_STALLED = 3,
} scrub_throttle_t;
+typedef enum scrub_freq {
+ BR_FSSCRUB_FREQ_HOURLY = 1,
+ BR_FSSCRUB_FREQ_DAILY,
+ BR_FSSCRUB_FREQ_WEEKLY,
+ BR_FSSCRUB_FREQ_BIWEEKLY,
+ BR_FSSCRUB_FREQ_MONTHLY,
+} scrub_freq_t;
+
#define signature_size(hl) (sizeof (br_isignature_t) + hl + 1)
struct br_scanfs {
@@ -57,6 +65,15 @@ struct br_scanfs {
unsigned int entries;
struct list_head queued;
struct list_head ready;
+
+ /* scheduler */
+ uint32_t boot;
+ gf_boolean_t kick;
+
+ pthread_mutex_t wakelock;
+ pthread_cond_t wakecond;
+
+ struct gf_tw_timer_list *timer;
};
struct br_child {
@@ -98,13 +115,21 @@ struct br_scrubber {
scrub_throttle_t throttle;
+ /**
+ * frequency of scanning for this subvolume. this should
+ * normally be per-child, but since all childs follow the
+ * same frequency for a volume, this option ends up here
+ * instead of br_child_t.
+ */
+ scrub_freq_t frequency;
+
pthread_mutex_t mutex;
pthread_cond_t cond;
unsigned int nr_scrubbers;
struct list_head scrubbers;
- /*
+ /**
* list of "rotatable" subvolume(s) undergoing scrubbing
*/
struct list_head scrublist;
@@ -139,11 +164,6 @@ struct br_private {
gf_boolean_t iamscrubber; /* function as a fs scrubber */
struct br_scrubber fsscrub; /* scrubbers for this subvolume */
-
- char *scrub_freq; /* Scrubber frequency*/
-
- struct timeval tv_before_scrub; /* time before starting scrubbing*/
- struct timeval tv_after_scrub; /* time after scrubbing completion*/
};
typedef struct br_private br_private_t;
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
index 9f6da89032f..504b8ab3635 100644
--- a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
@@ -29,6 +29,7 @@ enum br_mem_types {
gf_br_mt_br_scrubber_t,
gf_br_mt_br_fsscan_entry_t,
gf_br_stub_mt_br_stub_fd_t,
+ gf_br_stub_mt_br_scanner_freq_t,
gf_br_stub_mt_end,
};
generated by cgit (git 2.34.1) at 2024-11-15 20:19:17 +0000