mars/sy_old/mars_light.c

4077 lines
105 KiB
C
Raw Normal View History

2011-02-23 20:48:06 +00:00
// (c) 2011 Thomas Schoebel-Theuer / 1&1 Internet AG
//#define BRICK_DEBUGGING
//#define MARS_DEBUGGING
2011-03-02 09:30:56 +00:00
//#define IO_DEBUGGING
//#define STAT_DEBUGGING // here means: display full statistics
2011-02-23 20:48:06 +00:00
2011-08-31 11:42:04 +00:00
// disable this only for debugging!
#define RUN_PEERS
#define RUN_DATA
#define RUN_LOGINIT
#define RUN_PRIMARY
#define RUN_SYNCSTATUS
#define RUN_LOGFILES
#define RUN_REPLAY
#define RUN_DEVICE
2011-02-23 20:48:06 +00:00
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/string.h>
#include <linux/debug_locks.h>
#include <linux/major.h>
#include <linux/genhd.h>
#include <linux/blkdev.h>
2011-07-22 10:43:40 +00:00
#include "strategy.h"
2011-09-30 11:55:18 +00:00
#include "../buildtag.h"
2011-02-23 20:48:06 +00:00
#include <linux/wait.h>
// used brick types
2011-07-22 10:43:40 +00:00
#include "../mars_server.h"
#include "../mars_client.h"
#include "../mars_copy.h"
#include "../mars_bio.h"
#include "../mars_sio.h"
2011-07-22 10:43:40 +00:00
#include "../mars_aio.h"
#include "../mars_trans_logger.h"
#include "../mars_if.h"
2011-08-25 10:16:32 +00:00
#include "mars_proc.h"
#ifdef CONFIG_MARS_DEBUG // otherwise currently unused
#include "../mars_dummy.h"
#include "../mars_check.h"
#include "../mars_buf.h"
#include "../mars_usebuf.h"
#endif
2011-02-23 20:48:06 +00:00
2011-03-22 14:36:26 +00:00
#if 0
#define inline __attribute__((__noinline__))
#endif
2011-03-10 11:40:06 +00:00
2011-02-23 20:48:06 +00:00
static struct task_struct *main_thread = NULL;
2011-02-28 18:00:32 +00:00
typedef int (*light_worker_fn)(void *buf, struct mars_dent *dent);
2011-02-23 20:48:06 +00:00
struct light_class {
char *cl_name;
int cl_len;
char cl_type;
bool cl_hostcontext;
bool cl_serial;
bool cl_use_channel;
2011-02-23 20:48:06 +00:00
int cl_father;
light_worker_fn cl_prepare;
2011-02-23 20:48:06 +00:00
light_worker_fn cl_forward;
light_worker_fn cl_backward;
};
2011-03-10 11:40:06 +00:00
///////////////////////////////////////////////////////////////////////
// needed for logfile rotation
struct mars_rotate {
struct mars_global *global;
struct copy_brick *sync_brick;
struct mars_dent *replay_link;
struct mars_dent *aio_dent;
struct aio_brick *aio_brick;
struct mars_info aio_info;
struct trans_logger_brick *trans_brick;
struct mars_dent *first_log;
struct mars_dent *relevant_log;
struct mars_brick *relevant_brick;
struct mars_dent *next_relevant_log;
struct mars_brick *next_relevant_brick;
struct mars_dent *next_next_relevant_log;
struct mars_dent *prev_log;
struct mars_dent *next_log;
struct mars_dent *syncstatus_dent;
struct if_brick *if_brick;
const char *copy_path;
const char *parent_path;
struct copy_brick *copy_brick;
struct mars_limiter replay_limiter;
struct mars_limiter sync_limiter;
struct mars_limiter file_limiter;
struct semaphore inf_mutex;
const char *inf_prev_version;
int inf_prev_sequence;
long long switchover_timeout;
long long flip_start;
loff_t dev_size;
loff_t total_space;
loff_t remaining_space;
loff_t start_pos;
loff_t end_pos;
int max_sequence;
int copy_serial;
int relevant_serial;
bool has_error;
bool allow_update;
bool allow_sync;
bool allow_replay;
bool replay_mode;
bool todo_primary;
bool is_primary;
bool old_is_primary;
bool copy_is_done;
};
///////////////////////////////////////////////////////////////////////
2011-03-10 11:40:06 +00:00
// TUNING
int mars_mem_percent = 20;
EXPORT_SYMBOL_GPL(mars_mem_percent);
2012-02-11 22:32:21 +00:00
#define CONF_TRANS_SHADOW_LIMIT (1024 * 128) // don't fill the hashtable too much
2011-03-18 13:15:40 +00:00
#define CONF_TRANS_CHUNKSIZE (128 * 1024)
2011-11-14 14:21:15 +00:00
#define CONF_TRANS_MAX_MREF_SIZE PAGE_SIZE
2011-03-18 13:15:40 +00:00
//#define CONF_TRANS_ALIGN 512
#define CONF_TRANS_ALIGN 0
2011-03-29 14:40:40 +00:00
//#define TRANS_FAKE
2011-03-18 13:15:40 +00:00
2012-02-23 07:32:15 +00:00
#define CONF_TRANS_BATCHLEN 64
2011-03-29 14:40:40 +00:00
#define CONF_TRANS_PRIO MARS_PRIO_HIGH
2011-05-13 11:19:28 +00:00
#define CONF_TRANS_LOG_READS false
//#define CONF_TRANS_LOG_READS true
2011-05-26 14:32:32 +00:00
//#define CONF_TRANS_COMPLETION_SEMANTICS 2
#define CONF_TRANS_COMPLETION_SEMANTICS 0
2011-03-10 11:40:06 +00:00
#define CONF_ALL_BATCHLEN 1
2011-05-19 11:36:00 +00:00
#define CONF_ALL_PRIO MARS_PRIO_NORMAL
2011-03-18 13:15:40 +00:00
2011-03-24 16:05:46 +00:00
#define IF_SKIP_SYNC true
2011-03-30 12:02:50 +00:00
#define IF_MAX_PLUGGED 10000
2011-06-10 13:57:52 +00:00
#define IF_READAHEAD 0
//#define IF_READAHEAD 1
2011-05-06 10:25:52 +00:00
2011-06-10 13:57:52 +00:00
#define BIO_READAHEAD 0
//#define BIO_READAHEAD 1
2011-05-06 10:25:52 +00:00
#define BIO_NOIDLE true
#define BIO_SYNC true
#define BIO_UNPLUG true
#define AIO_WAIT_DURING_FDSYNC false
2011-03-11 13:57:54 +00:00
2011-06-17 11:32:38 +00:00
#define COPY_APPEND_MODE 0
//#define COPY_APPEND_MODE 1 // FIXME: does not work yet
2011-05-26 14:32:32 +00:00
#define COPY_PRIO MARS_PRIO_LOW
#ifdef CONFIG_MARS_MIN_SPACE
#define EXHAUSTED_LIMIT(max) ((max) / 100 * CONFIG_MARS_MIN_SPACE_PERCENT + global_free_space * 1024 * 1024)
#define EXHAUSTED(x,max) ((x) <= EXHAUSTED_LIMIT(max))
#else
#define EXHAUSTED_LIMIT(max) 0
#define EXHAUSTED(x,max) (false)
#endif
#define JAMMED(x) ((x) <= 1024 * 1024)
2011-03-10 11:40:06 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_trans_params(struct mars_brick *_brick, void *private)
2011-03-10 11:40:06 +00:00
{
2011-03-11 13:57:54 +00:00
struct trans_logger_brick *trans_brick = (void*)_brick;
2011-03-18 13:15:40 +00:00
if (_brick->type != (void*)&trans_logger_brick_type) {
MARS_ERR("bad brick type\n");
2011-08-25 10:16:32 +00:00
return -EINVAL;
2011-03-18 13:15:40 +00:00
}
if (!trans_brick->q_phase[1].q_ordering) {
trans_brick->q_phase[0].q_batchlen = CONF_TRANS_BATCHLEN;
trans_brick->q_phase[1].q_batchlen = CONF_ALL_BATCHLEN;
trans_brick->q_phase[2].q_batchlen = CONF_ALL_BATCHLEN;
trans_brick->q_phase[3].q_batchlen = CONF_ALL_BATCHLEN;
2011-04-29 09:36:10 +00:00
trans_brick->q_phase[0].q_io_prio = CONF_TRANS_PRIO;
trans_brick->q_phase[1].q_io_prio = CONF_ALL_PRIO;
trans_brick->q_phase[2].q_io_prio = CONF_ALL_PRIO;
trans_brick->q_phase[3].q_io_prio = CONF_ALL_PRIO;
2011-04-29 09:36:10 +00:00
trans_brick->q_phase[1].q_ordering = true;
trans_brick->q_phase[3].q_ordering = true;
2011-05-26 14:32:32 +00:00
2011-06-30 13:15:52 +00:00
trans_brick->shadow_mem_limit = CONF_TRANS_SHADOW_LIMIT;
2011-04-19 14:46:38 +00:00
trans_brick->log_reads = CONF_TRANS_LOG_READS;
2011-05-26 14:32:32 +00:00
trans_brick->completion_semantics = CONF_TRANS_COMPLETION_SEMANTICS;
2011-03-18 13:15:40 +00:00
#ifdef TRANS_FAKE
trans_brick->debug_shortcut = true;
2011-03-11 13:57:54 +00:00
#endif
2011-03-18 13:15:40 +00:00
trans_brick->max_mref_size = CONF_TRANS_MAX_MREF_SIZE;
2011-03-18 13:15:40 +00:00
trans_brick->align_size = CONF_TRANS_ALIGN;
trans_brick->chunk_size = CONF_TRANS_CHUNKSIZE;
2011-03-10 11:40:06 +00:00
}
2011-05-06 10:25:52 +00:00
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
2011-08-25 10:16:32 +00:00
return 1;
2011-03-10 11:40:06 +00:00
}
struct client_cookie {
bool limit_mode;
bool create_mode;
};
2011-03-22 14:36:26 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_client_params(struct mars_brick *_brick, void *private)
2011-03-22 14:36:26 +00:00
{
2012-01-23 12:39:08 +00:00
struct client_brick *client_brick = (void*)_brick;
struct client_cookie *clc = private;
2012-08-14 14:12:59 +00:00
client_brick->io_timeout = 0;
client_brick->limit_mode = clc ? clc->limit_mode : false;
2011-05-06 10:25:52 +00:00
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
2011-08-25 10:16:32 +00:00
return 1;
2011-03-22 14:36:26 +00:00
}
static
int _set_sio_params(struct mars_brick *_brick, void *private)
{
struct sio_brick *sio_brick = (void*)_brick;
if (_brick->type == (void*)&client_brick_type) {
return _set_client_params(_brick, private);
}
if (_brick->type != (void*)&sio_brick_type) {
MARS_ERR("bad brick type\n");
return -EINVAL;
}
sio_brick->o_direct = false; // important!
sio_brick->o_fdsync = true;
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
return 1;
}
2011-03-11 13:57:54 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_aio_params(struct mars_brick *_brick, void *private)
2011-03-11 13:57:54 +00:00
{
struct aio_brick *aio_brick = (void*)_brick;
struct client_cookie *clc = private;
2011-03-22 14:36:26 +00:00
if (_brick->type == (void*)&client_brick_type) {
2011-08-25 10:16:32 +00:00
return _set_client_params(_brick, private);
2011-03-22 14:36:26 +00:00
}
if (_brick->type == (void*)&sio_brick_type) {
return _set_sio_params(_brick, private);
}
2011-03-18 13:15:40 +00:00
if (_brick->type != (void*)&aio_brick_type) {
MARS_ERR("bad brick type\n");
2011-08-25 10:16:32 +00:00
return -EINVAL;
2011-03-18 13:15:40 +00:00
}
aio_brick->o_creat = clc && clc->create_mode;
2011-03-11 13:57:54 +00:00
aio_brick->o_direct = false; // important!
aio_brick->o_fdsync = true;
2011-04-29 09:36:10 +00:00
aio_brick->wait_during_fdsync = AIO_WAIT_DURING_FDSYNC;
2011-05-06 10:25:52 +00:00
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
2011-08-25 10:16:32 +00:00
return 1;
2011-03-11 13:57:54 +00:00
}
2011-03-18 13:15:40 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_bio_params(struct mars_brick *_brick, void *private)
2011-03-18 13:15:40 +00:00
{
struct bio_brick *bio_brick;
2011-03-22 14:36:26 +00:00
if (_brick->type == (void*)&client_brick_type) {
2011-08-25 10:16:32 +00:00
return _set_client_params(_brick, private);
2011-03-22 14:36:26 +00:00
}
2011-03-18 13:15:40 +00:00
if (_brick->type == (void*)&aio_brick_type) {
2011-08-25 10:16:32 +00:00
return _set_aio_params(_brick, private);
2011-03-18 13:15:40 +00:00
}
if (_brick->type == (void*)&sio_brick_type) {
return _set_sio_params(_brick, private);
}
2011-03-18 13:15:40 +00:00
if (_brick->type != (void*)&bio_brick_type) {
MARS_ERR("bad brick type\n");
2011-08-25 10:16:32 +00:00
return -EINVAL;
2011-03-18 13:15:40 +00:00
}
bio_brick = (void*)_brick;
bio_brick->ra_pages = BIO_READAHEAD;
2011-05-06 10:25:52 +00:00
bio_brick->do_noidle = BIO_NOIDLE;
bio_brick->do_sync = BIO_SYNC;
bio_brick->do_unplug = BIO_UNPLUG;
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
2011-08-25 10:16:32 +00:00
return 1;
2011-03-18 13:15:40 +00:00
}
2011-03-11 13:57:54 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_if_params(struct mars_brick *_brick, void *private)
2011-03-11 13:57:54 +00:00
{
struct if_brick *if_brick = (void*)_brick;
struct mars_rotate *rot = private;
2011-03-18 13:15:40 +00:00
if (_brick->type != (void*)&if_brick_type) {
MARS_ERR("bad brick type\n");
2011-08-25 10:16:32 +00:00
return -EINVAL;
2011-03-18 13:15:40 +00:00
}
if (rot)
if_brick->dev_size = rot->dev_size;
2011-03-30 12:02:50 +00:00
if_brick->max_plugged = IF_MAX_PLUGGED;
2011-03-11 13:57:54 +00:00
if_brick->readahead = IF_READAHEAD;
2011-03-30 12:02:50 +00:00
if_brick->skip_sync = IF_SKIP_SYNC;
MARS_INF("name = '%s' path = '%s' size = %lld\n", _brick->brick_name, _brick->brick_path, if_brick->dev_size);
2011-08-25 10:16:32 +00:00
return 1;
2011-03-11 13:57:54 +00:00
}
2011-03-10 11:40:06 +00:00
2011-08-25 10:16:32 +00:00
struct copy_cookie {
const char *argv[2];
const char *copy_path;
loff_t start_pos;
2012-08-02 08:47:42 +00:00
bool verify_mode;
2011-08-25 10:16:32 +00:00
const char *fullpath[2];
struct mars_output *output[2];
struct mars_info info[2];
};
2011-05-26 14:32:32 +00:00
static
2011-08-25 10:16:32 +00:00
int _set_copy_params(struct mars_brick *_brick, void *private)
2011-05-26 14:32:32 +00:00
{
struct copy_brick *copy_brick = (void*)_brick;
2011-08-25 10:16:32 +00:00
struct copy_cookie *cc = private;
int status = 1;
2011-05-26 14:32:32 +00:00
if (_brick->type != (void*)&copy_brick_type) {
MARS_ERR("bad brick type\n");
2011-08-25 10:16:32 +00:00
status = -EINVAL;
goto done;
2011-05-26 14:32:32 +00:00
}
copy_brick->append_mode = COPY_APPEND_MODE;
copy_brick->io_prio = COPY_PRIO;
2012-08-02 08:47:42 +00:00
copy_brick->verify_mode = cc->verify_mode;
copy_brick->repair_mode = true;
2011-05-26 14:32:32 +00:00
MARS_INF("name = '%s' path = '%s'\n", _brick->brick_name, _brick->brick_path);
2011-08-25 10:16:32 +00:00
/* Determine the copy area, switch on/off when necessary
*/
if (!copy_brick->power.button && copy_brick->power.led_off) {
int i;
copy_brick->copy_last = 0;
for (i = 0; i < 2; i++) {
status = cc->output[i]->ops->mars_get_info(cc->output[i], &cc->info[i]);
if (status < 0) {
MARS_WRN("cannot determine current size of '%s'\n", cc->argv[i]);
2011-08-25 10:16:32 +00:00
goto done;
}
MARS_DBG("%d '%s' current_size = %lld\n", i, cc->fullpath[i], cc->info[i].current_size);
}
copy_brick->copy_start = cc->info[1].current_size;
if (cc->start_pos != -1) {
copy_brick->copy_start = cc->start_pos;
if (unlikely(cc->info[0].current_size != cc->info[1].current_size)) {
MARS_ERR("oops, devices have different size %lld != %lld at '%s'\n", cc->info[0].current_size, cc->info[1].current_size, cc->copy_path);
status = -EINVAL;
goto done;
}
if (unlikely(cc->start_pos > cc->info[0].current_size)) {
MARS_ERR("bad start position %lld is larger than actual size %lld on '%s'\n", cc->start_pos, cc->info[0].current_size, cc->copy_path);
status = -EINVAL;
goto done;
}
}
MARS_DBG("copy_start = %lld\n", copy_brick->copy_start);
copy_brick->copy_end = cc->info[0].current_size;
MARS_DBG("copy_end = %lld\n", copy_brick->copy_end);
if (copy_brick->copy_start < copy_brick->copy_end) {
status = 1;
MARS_DBG("copy switch on\n");
}
} else if (copy_brick->power.button && copy_brick->power.led_on && copy_brick->copy_last == copy_brick->copy_end && copy_brick->copy_end > 0) {
status = 0;
MARS_DBG("copy switch off\n");
}
done:
return status;
2011-05-26 14:32:32 +00:00
}
2011-02-23 20:48:06 +00:00
///////////////////////////////////////////////////////////////////////
// internal helpers
#define MARS_DELIM ','
static
2012-12-17 08:35:49 +00:00
char *_parse_versionlink(const char *str, loff_t *pos)
{
char *res = NULL;
const char *tmp;
int count;
int status;
2012-12-17 08:35:49 +00:00
*pos = 0;
while (*str && *str++ != MARS_DELIM) {
// empty
}
tmp = str;
count = 0;
while (*tmp && *tmp != MARS_DELIM) {
tmp++;
count++;
}
res = brick_string_alloc(count + 1);
if (unlikely(!res)) {
MARS_DBG("bad alloc\n");
goto done;
}
strncpy(res, str, count);
res[count] = '\0';
2012-12-17 08:35:49 +00:00
status = sscanf(tmp, ",%lld", pos);
if (unlikely(status != 1)) {
MARS_DBG("status = %d\n", status);
brick_string_free(res);
res = NULL;
}
done:
return res;
}
2011-02-28 18:00:32 +00:00
static int _parse_args(struct mars_dent *dent, char *str, int count)
2011-02-23 20:48:06 +00:00
{
int i;
int status = -EINVAL;
if (!str)
goto done;
if (!dent->d_args) {
2011-08-12 11:09:48 +00:00
dent->d_args = brick_strdup(str);
2011-02-23 20:48:06 +00:00
if (!dent->d_args) {
status = -ENOMEM;
goto done;
}
}
for (i = 0; i < count; i++) {
char *tmp;
int len;
if (!*str)
goto done;
if (i == count-1) {
len = strlen(str);
} else {
char *tmp = strchr(str, MARS_DELIM);
if (!tmp)
goto done;
len = (tmp - str);
}
2011-08-25 10:16:32 +00:00
tmp = brick_string_alloc(len + 1);
2011-02-23 20:48:06 +00:00
if (!tmp) {
status = -ENOMEM;
goto done;
}
2011-08-12 11:09:48 +00:00
brick_string_free(dent->d_argv[i]);
2011-02-23 20:48:06 +00:00
dent->d_argv[i] = tmp;
strncpy(dent->d_argv[i], str, len);
dent->d_argv[i][len] = '\0';
str += len;
if (i != count-1)
str++;
}
status = 0;
done:
if (status < 0) {
MARS_ERR("bad syntax '%s' (should have %d args), status = %d\n", dent->d_args ? dent->d_args : "", count, status);
}
return status;
}
2011-11-08 17:07:42 +00:00
static
bool _check_switch(struct mars_global *global, const char *path)
{
int res = false;
struct mars_dent *allow_dent;
allow_dent = mars_find_dent(global, path);
if (!allow_dent || !allow_dent->new_link)
goto done;
sscanf(allow_dent->new_link, "%d", &res);
MARS_DBG("'%s' -> %d\n", path, res);
done:
return res;
}
static
bool _check_allow(struct mars_global *global, struct mars_dent *parent, const char *name)
{
int res = false;
char *path = path_make("%s/todo-%s/%s", parent->d_path, my_id(), name);
if (!path)
goto done;
res = _check_switch(global, path);
done:
brick_string_free(path);
return res;
}
2011-09-27 11:57:04 +00:00
///////////////////////////////////////////////////////////////////////
// status display
static
void _update_replay_link(struct mars_rotate *rot, struct trans_logger_info *inf)
{
char *check = NULL;
char *old = NULL;
char *new = NULL;
int status;
old = path_make("log-%09d-%s,%lld,%lld", inf->inf_sequence, inf->inf_host, inf->inf_min_pos, inf->inf_max_pos - inf->inf_min_pos);
if (!old) {
goto out;
}
new = path_make("%s/replay-%s", rot->parent_path, my_id());
if (!new) {
goto out;
}
/* Check whether something really has changed (avoid
* useless/disturbing timestamp updates)
*/
check = mars_readlink(new);
if (check && !strcmp(check, old)) {
MARS_DBG("replay symlink '%s' -> '%s' has not changed\n", old, new);
goto out;
}
status = mars_symlink(old, new, NULL, 0);
if (status < 0) {
MARS_ERR("cannot create symlink '%s' -> '%s' status = %d\n", old, new, status);
} else {
MARS_DBG("made replay symlink '%s' -> '%s' status = %d\n", old, new, status);
}
out:
brick_string_free(new);
brick_string_free(old);
brick_string_free(check);
}
static
void _update_version_link(struct mars_rotate *rot, struct trans_logger_info *inf)
{
char *data = brick_string_alloc(0);
char *old = brick_string_alloc(0);
char *new = NULL;
unsigned char *digest = brick_string_alloc(0);
char *prev = NULL;
char *check = NULL;
char *prev_link = NULL;
2012-12-17 08:35:49 +00:00
char *prev_digest;
int len;
int i;
int status;
if (unlikely(!data || !digest || !old)) {
MARS_ERR("no MEM\n");
goto out;
}
if (likely(inf->inf_sequence > 1)) {
if (unlikely(inf->inf_sequence < rot->inf_prev_sequence && rot->inf_prev_version)) {
MARS_ERR("BACKSKIP in sequence numbers detected: %d < %d\n", inf->inf_sequence, rot->inf_prev_sequence);
goto out;
}
if (unlikely(!rot->inf_prev_version || inf->inf_sequence != rot->inf_prev_sequence + 1)) {
prev = path_make("%s/version-%09d-%s", rot->parent_path, inf->inf_sequence - 1, my_id());
if (unlikely(!prev)) {
MARS_ERR("no MEM\n");
goto out;
}
prev_link = mars_readlink(prev);
if (unlikely(!prev_link)) {
MARS_ERR("cannot find previous version symlink '%s'\n", SAFE_STR(prev));
goto out;
}
brick_string_free(rot->inf_prev_version);
rot->inf_prev_version = brick_strdup(prev_link);
if (unlikely(!rot->inf_prev_version)) {
MARS_ERR("no MEM\n");
goto out;
}
rot->inf_prev_sequence = inf->inf_sequence - 1;
}
}
2012-12-17 08:35:49 +00:00
if (likely(rot->inf_prev_version)) {
char *tmp;
prev_digest = brick_strdup(rot->inf_prev_version);
if (unlikely(!prev_digest)) {
MARS_ERR("no MEM\n");
goto out;
}
// just take the hash part out of it
for (tmp = prev_digest; *tmp; tmp++)
if (*tmp == ',')
break;
*tmp = '\0';
} else {
prev_digest = "";
}
2012-12-17 08:35:49 +00:00
len = sprintf(data, "%s,%d,%lld,%s", inf->inf_host, inf->inf_sequence, inf->inf_log_pos, prev_digest);
if (prev_digest[0])
brick_string_free(prev_digest);
MARS_DBG("data = '%s' len = %d\n", data, len);
mars_digest(digest, data, len);
len = 0;
for (i = 0; i < mars_digest_size; i++) {
len += sprintf(old + len, "%02x", digest[i]);
}
2012-12-17 08:35:49 +00:00
if (likely(rot->inf_prev_version)) {
char *tmp;
prev_digest = brick_strdup(rot->inf_prev_version);
if (unlikely(!prev_digest)) {
MARS_ERR("no MEM\n");
goto out;
}
// take the part before ';'
for (tmp = prev_digest; *tmp; tmp++)
if (*tmp == ';')
break;
*tmp = '\0';
} else {
prev_digest = "";
}
len += sprintf(old + len, ",%s,%lld,%d;%s", inf->inf_host, inf->inf_log_pos, inf->inf_sequence, prev_digest);
if (prev_digest[0])
brick_string_free(prev_digest);
new = path_make("%s/version-%09d-%s", rot->parent_path, inf->inf_sequence, my_id());
if (!new) {
MARS_ERR("no MEM\n");
goto out;
}
/* Check whether something really has changed (avoid
* useless/disturbing timestamp updates)
*/
check = mars_readlink(new);
if (likely(check)) {
if (!strcmp(check, old)) {
MARS_DBG("version symlink '%s' -> '%s' has not changed\n", old, new);
goto out;
}
}
status = mars_symlink(old, new, NULL, 0);
if (unlikely(status < 0)) {
MARS_ERR("cannot create symlink '%s' -> '%s' status = %d\n", old, new, status);
} else {
MARS_DBG("make version symlink '%s' -> '%s' status = %d\n", old, new, status);
}
out:
brick_string_free(new);
brick_string_free(prev);
brick_string_free(data);
brick_string_free(digest);
brick_string_free(old);
brick_string_free(check);
brick_string_free(prev_link);
}
static
void _update_info(struct trans_logger_info *inf)
{
struct mars_rotate *rot = inf->inf_private;
if (unlikely(!rot)) {
MARS_ERR("rot is NULL\n");
goto done;
}
down(&rot->inf_mutex);
MARS_DBG("inf = %p '%s' seq = %d min_pos = %lld max_pos = %lld log_pos = %lld is_writeback = %d is_applying = %d is_logging = %d\n",
inf,
SAFE_STR(inf->inf_host),
inf->inf_sequence,
inf->inf_min_pos,
inf->inf_max_pos,
inf->inf_log_pos,
inf->inf_is_writeback,
inf->inf_is_applying,
inf->inf_is_logging);
if (inf->inf_is_writeback || inf->inf_is_applying) {
_update_replay_link(rot, inf);
}
if (inf->inf_is_logging || inf->inf_is_applying) {
_update_version_link(rot, inf);
}
up(&rot->inf_mutex);
done:;
}
static
void _make_new_replaylink(struct mars_rotate *rot, char *new_host, int new_sequence, loff_t end_pos)
{
struct trans_logger_info inf = {
.inf_private = rot,
.inf_host = new_host,
.inf_sequence = new_sequence,
.inf_min_pos = 0,
.inf_max_pos = 0,
.inf_log_pos = end_pos,
.inf_is_applying = true,
};
down(&rot->inf_mutex);
MARS_DBG("new_host = '%s' new_sequence = %d end_pos = %lld\n", new_host, new_sequence, end_pos);
_update_replay_link(rot, &inf);
_update_version_link(rot, &inf);
up(&rot->inf_mutex);
#ifdef CONFIG_MARS_FAST_TRIGGER
mars_trigger();
mars_remote_trigger();
#endif
}
2011-09-27 11:57:04 +00:00
static
int __show_actual(const char *path, const char *name, int val)
2011-09-27 11:57:04 +00:00
{
char *src;
char *dst = NULL;
int status = -EINVAL;
src = path_make("%d", val);
2011-09-27 11:57:04 +00:00
dst = path_make("%s/actual-%s/%s", path, my_id(), name);
status = -ENOMEM;
if (!dst)
goto done;
2011-09-27 11:57:04 +00:00
MARS_DBG("symlink '%s' -> '%s'\n", dst, src);
status = mars_symlink(src, dst, NULL, 0);
done:
brick_string_free(src);
2011-09-27 11:57:04 +00:00
brick_string_free(dst);
return status;
}
static inline
int _show_actual(const char *path, const char *name, bool val)
{
return __show_actual(path, name, val ? 1 : 0);
}
2011-09-27 11:57:04 +00:00
static
void _show_primary(struct mars_rotate *rot, struct mars_dent *parent)
{
int status;
if (!rot || !parent) {
return;
}
status = _show_actual(parent->d_path, "is-primary", rot->is_primary);
2012-01-17 14:37:14 +00:00
if (rot->is_primary != rot->old_is_primary) {
rot->old_is_primary = rot->is_primary;
mars_remote_trigger();
}
2011-09-27 11:57:04 +00:00
}
static
void _show_brick_status(struct mars_brick *test, bool shutdown)
{
const char *path;
char *src;
char *dst;
int status;
path = test->brick_path;
if (!path) {
MARS_WRN("bad path\n");
return;
}
if (*path != '/') {
MARS_WRN("bogus path '%s'\n", path);
return;
}
src = (test->power.led_on && !shutdown) ? "1" : "0";
dst = backskip_replace(path, '/', true, "/actual-%s/", my_id());
if (!dst) {
return;
}
status = mars_symlink(src, dst, NULL, 0);
MARS_DBG("status symlink '%s' -> '%s' status = %d\n", dst, src, status);
brick_string_free(dst);
}
static
void _show_status_all(struct mars_global *global)
{
struct list_head *tmp;
down_read(&global->brick_mutex);
for (tmp = global->brick_anchor.next; tmp != &global->brick_anchor; tmp = tmp->next) {
struct mars_brick *test;
test = container_of(tmp, struct mars_brick, global_brick_link);
if (!test->show_status)
2011-09-27 11:57:04 +00:00
continue;
_show_brick_status(test, false);
}
up_read(&global->brick_mutex);
}
static
void _show_rate(struct mars_rotate *rot, struct mars_limiter *limiter, bool running, const char *name)
{
int rate = limiter->lim_rate;
if (!running)
rate = 0;
__show_actual(rot->parent_path, name, rate);
}
2011-09-27 11:57:04 +00:00
2011-02-25 11:46:38 +00:00
///////////////////////////////////////////////////////////////////////
static
2011-03-01 09:34:36 +00:00
int __make_copy(
struct mars_global *global,
struct mars_dent *belongs,
2011-03-23 08:09:00 +00:00
const char *switch_path,
2011-03-22 14:36:26 +00:00
const char *copy_path,
2011-03-03 09:02:10 +00:00
const char *parent,
2011-03-01 09:34:36 +00:00
const char *argv[],
2011-03-04 07:41:00 +00:00
loff_t start_pos, // -1 means at EOF
2012-08-02 08:47:42 +00:00
bool verify_mode,
bool limit_mode,
2011-03-01 09:34:36 +00:00
struct copy_brick **__copy)
2011-02-25 11:46:38 +00:00
{
struct mars_brick *copy;
struct copy_brick *_copy;
2011-08-25 10:16:32 +00:00
struct copy_cookie cc = {};
struct client_cookie clc[2] = {
{
.limit_mode = limit_mode,
},
{
.limit_mode = limit_mode,
.create_mode = true,
},
};
2011-02-25 11:46:38 +00:00
int i;
2011-03-23 17:58:02 +00:00
int status = -EINVAL;
2011-11-08 17:07:42 +00:00
if (!switch_path || !global) {
2011-03-23 17:58:02 +00:00
goto done;
}
2011-02-25 11:46:38 +00:00
2011-11-08 17:07:42 +00:00
// don't generate empty aio files if copy does not yet exist
copy = mars_find_brick(global, &copy_brick_type, copy_path);
if (!copy && !_check_switch(global, switch_path))
goto done;
// create/find predecessor aio bricks
2011-02-25 11:46:38 +00:00
for (i = 0; i < 2; i++) {
2011-03-02 09:30:56 +00:00
struct mars_brick *aio;
2011-03-03 09:02:10 +00:00
2011-08-25 10:16:32 +00:00
cc.argv[i] = argv[i];
2011-03-03 09:02:10 +00:00
if (parent) {
2011-08-25 10:16:32 +00:00
cc.fullpath[i] = path_make("%s/%s", parent, argv[i]);
if (!cc.fullpath[i]) {
2011-03-03 09:02:10 +00:00
MARS_ERR("cannot make path '%s/%s'\n", parent, argv[i]);
goto done;
}
} else {
2011-08-25 10:16:32 +00:00
cc.fullpath[i] = argv[i];
2011-03-03 09:02:10 +00:00
}
2011-03-02 09:30:56 +00:00
aio =
2011-03-01 09:34:36 +00:00
make_brick_all(global,
2011-03-11 13:57:54 +00:00
NULL,
2011-09-02 12:17:40 +00:00
false,
_set_bio_params,
&clc[i],
2011-03-01 09:34:36 +00:00
10 * HZ,
2011-03-03 09:02:10 +00:00
NULL,
2011-03-22 14:36:26 +00:00
(const struct generic_brick_type*)&bio_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){},
2011-03-23 08:09:00 +00:00
NULL,
1, // start always
2011-08-25 10:16:32 +00:00
cc.fullpath[i],
2011-03-01 09:34:36 +00:00
(const char *[]){},
2011-03-03 09:02:10 +00:00
0);
2011-03-02 09:30:56 +00:00
if (!aio) {
2011-08-25 10:16:32 +00:00
MARS_DBG("cannot instantiate '%s'\n", cc.fullpath[i]);
2011-02-25 11:46:38 +00:00
goto done;
}
2011-08-25 10:16:32 +00:00
cc.output[i] = aio->outputs[0];
2011-02-25 11:46:38 +00:00
}
2011-08-25 10:16:32 +00:00
cc.copy_path = copy_path;
cc.start_pos = start_pos;
2012-08-02 08:47:42 +00:00
cc.verify_mode = verify_mode;
2011-08-25 10:16:32 +00:00
2011-03-01 09:34:36 +00:00
copy =
make_brick_all(global,
belongs,
2011-09-02 12:17:40 +00:00
false,
2011-05-26 14:32:32 +00:00
_set_copy_params,
2011-08-25 10:16:32 +00:00
&cc,
10 * HZ,
cc.fullpath[1],
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*)&copy_brick_type,
(const struct generic_brick_type*[]){NULL,NULL,NULL,NULL},
2011-03-03 09:02:10 +00:00
"%s",
2012-01-31 15:19:35 +00:00
(global->exhausted || !switch_path[0]) ? -1 : 0,
2011-03-23 08:09:00 +00:00
"%s",
2011-03-03 09:02:10 +00:00
(const char *[]){"%s", "%s", "%s", "%s"},
2011-03-02 09:30:56 +00:00
4,
2011-03-23 08:09:00 +00:00
switch_path,
2011-03-22 14:36:26 +00:00
copy_path,
2011-08-25 10:16:32 +00:00
cc.fullpath[0],
cc.fullpath[0],
cc.fullpath[1],
cc.fullpath[1]);
2011-02-25 11:46:38 +00:00
if (!copy) {
2011-06-17 11:32:38 +00:00
MARS_DBG("creation of copy brick '%s' failed\n", copy_path);
2011-03-01 09:34:36 +00:00
goto done;
2011-02-25 11:46:38 +00:00
}
2011-09-27 11:57:04 +00:00
copy->show_status = _show_brick_status;
2011-02-25 11:46:38 +00:00
_copy = (void*)copy;
if (__copy)
*__copy = _copy;
status = 0;
done:
MARS_DBG("status = %d\n", status);
2011-03-03 09:02:10 +00:00
for (i = 0; i < 2; i++) {
2011-08-25 10:16:32 +00:00
if (cc.fullpath[i] && cc.fullpath[i] != argv[i])
brick_string_free(cc.fullpath[i]);
2011-03-03 09:02:10 +00:00
}
2011-02-25 11:46:38 +00:00
return status;
}
2011-02-23 20:48:06 +00:00
///////////////////////////////////////////////////////////////////////
// remote workers
struct mars_peerinfo {
2011-02-24 16:37:32 +00:00
struct mars_global *global;
2011-02-23 20:48:06 +00:00
char *peer;
char *path;
struct mars_socket socket;
2011-08-25 10:16:32 +00:00
struct task_struct *peer_thread;
2011-03-22 14:36:26 +00:00
spinlock_t lock;
struct list_head remote_dent_list;
2011-02-23 20:48:06 +00:00
int maxdepth;
};
2011-02-24 16:37:32 +00:00
static
2011-03-23 08:09:00 +00:00
bool _is_usable_dir(const char *name)
{
if (!strncmp(name, "resource-", 9)
|| !strncmp(name, "todo-", 5)
|| !strncmp(name, "actual-", 7)
2011-03-23 08:09:00 +00:00
|| !strncmp(name, "defaults", 8)
) {
return true;
}
return false;
}
static
bool _is_peer_logfile(const char *name, const char *id)
2011-03-22 14:36:26 +00:00
{
int len = strlen(name);
int idlen = id ? strlen(id) : 4 + 9 + 1;
if (len <= idlen ||
strncmp(name, "log-", 4) != 0) {
MARS_DBG("not a logfile at all: '%s'\n", name);
return false;
}
if (id &&
name[len - idlen - 1] == '-' &&
strncmp(name + len - idlen, id, idlen) == 0) {
MARS_DBG("not a peer logfile: '%s'\n", name);
return false;
}
MARS_DBG("found peer logfile: '%s'\n", name);
return true;
}
static
int _update_file(struct mars_rotate *rot, const char *switch_path, const char *copy_path, const char *file, const char *peer, loff_t end_pos)
2011-02-24 16:37:32 +00:00
{
struct mars_global *global = rot->global;
2011-03-22 14:36:26 +00:00
const char *tmp = path_make("%s@%s", file, peer);
const char *argv[2] = { tmp, file };
2011-03-04 07:41:00 +00:00
struct copy_brick *copy = NULL;
2011-03-01 18:00:14 +00:00
int status = -ENOMEM;
2011-02-24 16:37:32 +00:00
if (unlikely(!tmp || !global))
2011-03-01 18:00:14 +00:00
goto done;
2011-02-24 16:37:32 +00:00
2011-03-22 14:36:26 +00:00
MARS_DBG("src = '%s' dst = '%s'\n", tmp, file);
status = __make_copy(global, NULL, switch_path, copy_path, NULL, argv, -1, false, false, &copy);
if (status >= 0 && copy) {
copy->copy_limiter = &rot->file_limiter;
if ((!copy->append_mode || copy->power.led_off) &&
end_pos > copy->copy_end) {
2011-03-22 14:36:26 +00:00
MARS_DBG("appending to '%s' %lld => %lld\n", copy_path, copy->copy_end, end_pos);
2011-03-04 07:41:00 +00:00
copy->copy_end = end_pos;
}
}
2011-02-24 16:37:32 +00:00
2011-03-01 18:00:14 +00:00
done:
2011-10-05 14:59:29 +00:00
brick_string_free(tmp);
2011-02-24 16:37:32 +00:00
return status;
}
static
2012-02-21 11:23:38 +00:00
int check_logfile(const char *peer, struct mars_dent *remote_dent, struct mars_dent *local_dent, struct mars_dent *parent, loff_t dst_size)
2011-02-24 16:37:32 +00:00
{
loff_t src_size = remote_dent->new_stat.size;
2011-07-08 07:02:14 +00:00
struct mars_rotate *rot;
2011-03-23 08:09:00 +00:00
const char *switch_path = NULL;
2011-03-22 14:36:26 +00:00
struct copy_brick *copy_brick;
int status = 0;
2011-02-24 16:37:32 +00:00
2011-03-25 10:34:54 +00:00
// plausibility checks
if (unlikely(dst_size > src_size)) {
MARS_WRN("my local copy is larger than the remote one, ignoring\n");
status = -EINVAL;
2011-03-22 14:36:26 +00:00
goto done;
2011-02-24 16:37:32 +00:00
}
2011-03-22 14:36:26 +00:00
// check whether we are participating in that resource
2011-07-08 07:02:14 +00:00
rot = parent->d_private;
if (!rot) {
2011-11-08 17:07:42 +00:00
MARS_WRN("parent has no rot info\n");
2011-07-08 07:02:14 +00:00
status = -EINVAL;
goto done;
}
2012-02-21 11:23:38 +00:00
if (!rot->copy_path) {
MARS_WRN("parent has no copy_path\n");
status = -EINVAL;
goto done;
}
2011-07-08 07:02:14 +00:00
2011-03-22 14:36:26 +00:00
// check whether connection is allowed
2011-09-21 11:30:11 +00:00
switch_path = path_make("%s/todo-%s/connect", parent->d_path, my_id());
2012-02-03 10:25:17 +00:00
// check whether copy is necessary
2012-02-21 11:23:38 +00:00
copy_brick = rot->copy_brick;
MARS_DBG("copy_brick = %p (remote '%s' %d) copy_serial = %d\n", copy_brick, remote_dent->d_path, remote_dent->d_serial, rot->copy_serial);
2012-02-03 10:25:17 +00:00
if (copy_brick) {
2012-02-21 11:23:38 +00:00
if (remote_dent->d_serial == rot->copy_serial) {
2012-02-03 10:25:17 +00:00
// treat copy brick instance underway
2012-02-21 11:23:38 +00:00
status = _update_file(rot, switch_path, rot->copy_path, remote_dent->d_path, peer, src_size);
MARS_DBG("re-update '%s' from peer '%s' status = %d\n", remote_dent->d_path, peer, status);
2012-02-03 10:25:17 +00:00
}
2012-02-21 11:23:38 +00:00
} else if (!rot->copy_serial && rot->allow_update &&
2012-02-03 10:25:17 +00:00
(dst_size < src_size || !local_dent)) {
// start copy brick instance
2012-02-21 11:23:38 +00:00
status = _update_file(rot, switch_path, rot->copy_path, remote_dent->d_path, peer, src_size);
MARS_DBG("update '%s' from peer '%s' status = %d\n", remote_dent->d_path, peer, status);
rot->copy_serial = remote_dent->d_serial;
2011-03-22 14:36:26 +00:00
}
done:
2011-10-05 14:59:29 +00:00
brick_string_free(switch_path);
2011-03-22 14:36:26 +00:00
return status;
2011-02-24 16:37:32 +00:00
}
static
int run_bone(struct mars_peerinfo *peer, struct mars_dent *remote_dent)
2011-02-23 20:48:06 +00:00
{
int status = 0;
2011-02-27 14:17:58 +00:00
struct kstat local_stat = {};
bool stat_ok;
bool update_mtime = true;
bool update_ctime = true;
2011-03-22 14:36:26 +00:00
bool run_trigger = false;
2011-02-24 16:37:32 +00:00
if (!strncmp(remote_dent->d_name, ".tmp", 4)) {
2011-02-24 16:37:32 +00:00
goto done;
}
if (!strncmp(remote_dent->d_name, "ignore", 6)) {
2011-02-24 16:37:32 +00:00
goto done;
}
status = mars_stat(remote_dent->d_path, &local_stat, true);
2011-02-27 14:17:58 +00:00
stat_ok = (status >= 0);
if (stat_ok) {
update_mtime = timespec_compare(&remote_dent->new_stat.mtime, &local_stat.mtime) > 0;
update_ctime = timespec_compare(&remote_dent->new_stat.ctime, &local_stat.ctime) > 0;
2011-02-27 14:17:58 +00:00
2012-01-24 09:06:33 +00:00
MARS_IO("timestamps '%s' remote = %ld.%09ld local = %ld.%09ld\n", remote_dent->d_path, remote_dent->new_stat.mtime.tv_sec, remote_dent->new_stat.mtime.tv_nsec, local_stat.mtime.tv_sec, local_stat.mtime.tv_nsec);
2011-02-27 14:17:58 +00:00
if ((remote_dent->new_stat.mode & S_IRWXU) !=
2011-02-27 14:17:58 +00:00
(local_stat.mode & S_IRWXU) &&
update_ctime) {
mode_t newmode = local_stat.mode;
2012-01-24 09:06:33 +00:00
MARS_IO("chmod '%s' 0x%xd -> 0x%xd\n", remote_dent->d_path, newmode & S_IRWXU, remote_dent->new_stat.mode & S_IRWXU);
2011-02-27 14:17:58 +00:00
newmode &= ~S_IRWXU;
newmode |= (remote_dent->new_stat.mode & S_IRWXU);
mars_chmod(remote_dent->d_path, newmode);
2011-03-22 14:36:26 +00:00
run_trigger = true;
2011-02-27 14:17:58 +00:00
}
if (remote_dent->new_stat.uid != local_stat.uid && update_ctime) {
2012-01-24 09:06:33 +00:00
MARS_IO("lchown '%s' %d -> %d\n", remote_dent->d_path, local_stat.uid, remote_dent->new_stat.uid);
mars_lchown(remote_dent->d_path, remote_dent->new_stat.uid);
2011-03-22 14:36:26 +00:00
run_trigger = true;
2011-02-27 14:17:58 +00:00
}
}
2011-02-23 20:48:06 +00:00
if (S_ISDIR(remote_dent->new_stat.mode)) {
if (!_is_usable_dir(remote_dent->d_name)) {
2012-01-24 09:06:33 +00:00
MARS_IO("ignoring directory '%s'\n", remote_dent->d_path);
2011-02-24 16:37:32 +00:00
goto done;
}
2011-02-27 14:17:58 +00:00
if (!stat_ok) {
status = mars_mkdir(remote_dent->d_path);
2012-01-24 09:06:33 +00:00
MARS_IO("create directory '%s' status = %d\n", remote_dent->d_path, status);
2011-03-22 14:36:26 +00:00
if (status >= 0) {
mars_chmod(remote_dent->d_path, remote_dent->new_stat.mode);
mars_lchown(remote_dent->d_path, remote_dent->new_stat.uid);
2011-03-22 14:36:26 +00:00
}
2011-02-23 20:48:06 +00:00
}
} else if (S_ISLNK(remote_dent->new_stat.mode) && remote_dent->new_link) {
2011-02-27 14:17:58 +00:00
if (!stat_ok || update_mtime) {
status = mars_symlink(remote_dent->new_link, remote_dent->d_path, &remote_dent->new_stat.mtime, remote_dent->new_stat.uid);
2012-01-24 09:06:33 +00:00
MARS_IO("create symlink '%s' -> '%s' status = %d\n", remote_dent->d_path, remote_dent->new_link, status);
2011-03-22 14:36:26 +00:00
run_trigger = true;
2011-02-24 16:37:32 +00:00
}
} else if (S_ISREG(remote_dent->new_stat.mode) && _is_peer_logfile(remote_dent->d_name, my_id())) {
const char *parent_path = backskip_replace(remote_dent->d_path, '/', false, "");
2011-03-22 14:36:26 +00:00
if (likely(parent_path)) {
struct mars_dent *parent = mars_find_dent(peer->global, parent_path);
struct mars_dent *local_dent = mars_find_dent(peer->global, remote_dent->d_path);
2011-03-22 14:36:26 +00:00
if (unlikely(!parent)) {
2012-01-24 09:06:33 +00:00
MARS_IO("ignoring non-existing local resource '%s'\n", parent_path);
#if defined(CONFIG_MARS_LOGDELETE_AUTO)
2012-03-06 11:56:48 +00:00
// don't copy old / outdated logfiles
} else if (parent->d_private &&
((struct mars_rotate *)parent->d_private)->relevant_serial > remote_dent->d_serial) {
MARS_IO("ignoring outdated remote logfile '%s'\n", remote_dent->d_path);
#endif
2011-03-22 14:36:26 +00:00
} else {
2012-02-21 11:23:38 +00:00
status = check_logfile(peer->peer, remote_dent, local_dent, parent, local_stat.size);
2011-02-24 16:37:32 +00:00
}
2011-08-12 11:09:48 +00:00
brick_string_free(parent_path);
2011-02-24 16:37:32 +00:00
}
2011-02-23 20:48:06 +00:00
} else {
2012-01-24 09:06:33 +00:00
MARS_IO("ignoring '%s'\n", remote_dent->d_path);
2011-02-23 20:48:06 +00:00
}
2011-02-24 16:37:32 +00:00
done:
2011-03-22 14:36:26 +00:00
if (status >= 0) {
status = run_trigger ? 1 : 0;
}
return status;
}
static
int run_bones(struct mars_peerinfo *peer)
{
LIST_HEAD(tmp_list);
struct list_head *tmp;
unsigned long flags;
bool run_trigger = false;
int status = 0;
traced_lock(&peer->lock, flags);
list_replace_init(&peer->remote_dent_list, &tmp_list);
traced_unlock(&peer->lock, flags);
2012-01-24 09:00:21 +00:00
MARS_DBG("remote_dent_list list_empty = %d\n", list_empty(&tmp_list));
2011-03-22 14:36:26 +00:00
for (tmp = tmp_list.next; tmp != &tmp_list; tmp = tmp->next) {
struct mars_dent *remote_dent = container_of(tmp, struct mars_dent, dent_link);
if (!remote_dent->d_path) {
2011-03-22 14:36:26 +00:00
MARS_DBG("NULL\n");
continue;
}
2012-01-24 09:06:33 +00:00
MARS_IO("path = '%s'\n", remote_dent->d_path);
status = run_bone(peer, remote_dent);
2011-03-22 14:36:26 +00:00
if (status > 0)
run_trigger = true;
//MARS_DBG("path = '%s' worker status = %d\n", remote_dent->d_path, status);
2011-03-22 14:36:26 +00:00
}
2011-09-27 08:46:14 +00:00
mars_free_dent_all(NULL, &tmp_list);
2012-01-17 10:48:56 +00:00
#ifdef CONFIG_MARS_FAST_TRIGGER
2011-03-22 14:36:26 +00:00
if (run_trigger) {
mars_trigger();
}
#endif
2011-02-23 20:48:06 +00:00
return status;
}
///////////////////////////////////////////////////////////////////////
// remote working infrastructure
2011-02-24 16:37:32 +00:00
static
void _peer_cleanup(struct mars_peerinfo *peer)
2011-02-23 20:48:06 +00:00
{
MARS_DBG("cleanup\n");
if (mars_socket_is_alive(&peer->socket)) {
MARS_DBG("really shutdown socket\n");
mars_shutdown_socket(&peer->socket);
2011-02-23 20:48:06 +00:00
}
mars_put_socket(&peer->socket);
2011-02-23 20:48:06 +00:00
}
2012-01-17 14:37:14 +00:00
static DECLARE_WAIT_QUEUE_HEAD(remote_event);
static atomic_t remote_trigger_count = ATOMIC_INIT(0);
static atomic_t peer_thread_count = ATOMIC_INIT(0);
2011-02-24 16:37:32 +00:00
static
int peer_thread(void *data)
2011-02-23 20:48:06 +00:00
{
struct mars_peerinfo *peer = data;
2011-02-24 16:37:32 +00:00
char *real_peer;
2011-02-23 20:48:06 +00:00
struct sockaddr_storage sockaddr = {};
int pause_time = 0;
bool do_kill = false;
2012-01-24 09:00:21 +00:00
bool flip = false;
2011-02-23 20:48:06 +00:00
int status;
if (!peer)
return -1;
2011-07-22 10:43:40 +00:00
real_peer = mars_translate_hostname(peer->peer);
MARS_INF("-------- peer thread starting on peer '%s' (%s)\n", peer->peer, real_peer);
2011-02-23 20:48:06 +00:00
2011-02-24 16:37:32 +00:00
status = mars_create_sockaddr(&sockaddr, real_peer);
2011-02-23 20:48:06 +00:00
if (unlikely(status < 0)) {
2011-02-24 16:37:32 +00:00
MARS_ERR("unusable remote address '%s' (%s)\n", real_peer, peer->peer);
2011-02-23 20:48:06 +00:00
goto done;
}
2012-01-17 14:37:14 +00:00
atomic_inc(&peer_thread_count);
while (!brick_thread_should_stop()) {
2011-02-23 20:48:06 +00:00
LIST_HEAD(tmp_list);
2011-03-22 14:36:26 +00:00
LIST_HEAD(old_list);
unsigned long flags;
2011-02-23 20:48:06 +00:00
struct mars_cmd cmd = {
.cmd_code = CMD_GETENTS,
.cmd_str1 = peer->path,
.cmd_int1 = peer->maxdepth,
};
if (!mars_socket_is_alive(&peer->socket)) {
if (do_kill) {
do_kill = false;
_peer_cleanup(peer);
brick_msleep(5000);
continue;
}
2012-08-06 13:04:32 +00:00
if (!mars_net_is_alive) {
brick_msleep(1000);
2012-08-06 13:04:32 +00:00
continue;
}
status = mars_create_socket(&peer->socket, &sockaddr, false);
if (unlikely(status < 0)) {
2011-02-24 16:37:32 +00:00
MARS_INF("no connection to '%s'\n", real_peer);
brick_msleep(5000);
2011-02-23 20:48:06 +00:00
continue;
}
do_kill = true;
peer->socket.s_shutdown_on_err = true;
2011-02-24 16:37:32 +00:00
MARS_DBG("successfully opened socket to '%s'\n", real_peer);
brick_msleep(100);
2011-02-23 20:48:06 +00:00
continue;
}
2012-01-17 14:37:14 +00:00
/* This is not completely race-free, but does no harm.
* In worst case, network propagation will just take
* a litte longer (see CONFIG_MARS_PROPAGATE_INTERVAL).
*/
2012-01-24 09:00:21 +00:00
if (!flip && atomic_read(&remote_trigger_count) > 0) {
MARS_DBG("sending notify ... remote_tiogger_count = %d\n", atomic_read(&remote_trigger_count));
2012-01-17 14:37:14 +00:00
atomic_dec(&remote_trigger_count);
cmd.cmd_code = CMD_NOTIFY;
2012-01-24 09:00:21 +00:00
flip = true;
2012-01-17 14:37:14 +00:00
}
status = mars_send_struct(&peer->socket, &cmd, mars_cmd_meta);
2011-02-23 20:48:06 +00:00
if (unlikely(status < 0)) {
2011-08-31 11:42:04 +00:00
MARS_WRN("communication error on send, status = %d\n", status);
if (do_kill) {
do_kill = false;
_peer_cleanup(peer);
}
brick_msleep(2000);
2011-02-23 20:48:06 +00:00
continue;
}
2012-01-24 09:00:21 +00:00
if (cmd.cmd_code == CMD_NOTIFY) {
flip = false;
pause_time = 0;
brick_msleep(1000);
2012-01-17 14:37:14 +00:00
continue;
2012-01-24 09:00:21 +00:00
}
2011-02-23 20:48:06 +00:00
2012-01-24 09:00:21 +00:00
MARS_DBG("fetching remote dentry list\n");
status = mars_recv_dent_list(&peer->socket, &tmp_list);
2011-02-23 20:48:06 +00:00
if (unlikely(status < 0)) {
2011-08-31 11:42:04 +00:00
MARS_WRN("communication error on receive, status = %d\n", status);
if (do_kill) {
do_kill = false;
_peer_cleanup(peer);
}
mars_free_dent_all(NULL, &tmp_list);
brick_msleep(5000);
2011-02-23 20:48:06 +00:00
continue;
}
2012-01-18 14:33:39 +00:00
if (likely(!list_empty(&tmp_list))) {
2012-01-24 09:00:21 +00:00
MARS_DBG("got remote denties\n");
2011-02-23 20:48:06 +00:00
2012-01-18 14:33:39 +00:00
traced_lock(&peer->lock, flags);
2011-03-22 14:36:26 +00:00
2012-01-18 14:33:39 +00:00
list_replace_init(&peer->remote_dent_list, &old_list);
list_replace_init(&tmp_list, &peer->remote_dent_list);
2011-02-23 20:48:06 +00:00
2012-01-18 14:33:39 +00:00
traced_unlock(&peer->lock, flags);
2011-03-22 14:36:26 +00:00
2012-01-18 14:33:39 +00:00
mars_free_dent_all(NULL, &old_list);
}
2011-02-23 20:48:06 +00:00
brick_msleep(1000);
if (!brick_thread_should_stop()) {
if (pause_time < CONFIG_MARS_PROPAGATE_INTERVAL)
pause_time++;
wait_event_interruptible_timeout(remote_event,
atomic_read(&remote_trigger_count) > 0 ||
(mars_global && mars_global->main_trigger),
pause_time * HZ);
}
2011-02-23 20:48:06 +00:00
}
2012-01-19 12:09:22 +00:00
MARS_INF("-------- peer thread terminating\n");
2011-02-23 20:48:06 +00:00
if (do_kill) {
_peer_cleanup(peer);
}
2011-02-23 20:48:06 +00:00
done:
2012-01-17 14:37:14 +00:00
atomic_dec(&peer_thread_count);
2011-08-25 10:16:32 +00:00
brick_string_free(real_peer);
2011-02-23 20:48:06 +00:00
return 0;
}
2012-01-17 14:37:14 +00:00
static
void __mars_remote_trigger(void)
{
int count = atomic_read(&peer_thread_count);
atomic_add(count, &remote_trigger_count);
wake_up_interruptible_all(&remote_event);
}
static
bool is_shutdown(void)
{
bool res = false;
int used;
if ((used = atomic_read(&global_mshadow_count)) > 0) {
MARS_INF("global shutdown delayed: there are %d buffers in use, occupying %ld bytes\n", used, atomic64_read(&global_mshadow_used));
} else {
int rounds = 3;
while ((used = atomic_read(&mars_global_io_flying)) <= 0) {
if (--rounds <= 0) {
res = true;
break;
}
brick_msleep(30);
}
if (!res) {
MARS_INF("global shutdown delayed: there are %d IO requests flying\n", used);
}
}
return res;
}
2011-02-23 20:48:06 +00:00
///////////////////////////////////////////////////////////////////////
// helpers for worker functions
2011-02-28 18:00:32 +00:00
static int _kill_peer(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
LIST_HEAD(tmp_list);
2011-02-23 20:48:06 +00:00
struct mars_global *global = buf;
struct mars_peerinfo *peer = dent->d_private;
unsigned long flags;
2011-02-23 20:48:06 +00:00
if (global->global_power.button) {
return 0;
}
if (!peer) {
return 0;
}
2011-08-31 11:42:04 +00:00
MARS_INF("stopping peer thread...\n");
if (peer->peer_thread) {
brick_thread_stop(peer->peer_thread);
}
traced_lock(&peer->lock, flags);
list_replace_init(&peer->remote_dent_list, &tmp_list);
traced_unlock(&peer->lock, flags);
mars_free_dent_all(NULL, &tmp_list);
brick_string_free(peer->peer);
brick_string_free(peer->path);
2011-02-23 20:48:06 +00:00
dent->d_private = NULL;
2011-08-31 11:42:04 +00:00
brick_mem_free(peer);
2011-02-23 20:48:06 +00:00
return 0;
}
2012-02-03 10:25:17 +00:00
static int _make_peer(struct mars_global *global, struct mars_dent *dent, char *path)
2011-02-23 20:48:06 +00:00
{
static int serial = 0;
struct mars_peerinfo *peer;
2012-02-03 10:25:17 +00:00
char *mypeer;
char *parent_path;
2011-02-23 20:48:06 +00:00
int status = 0;
2012-02-03 10:25:17 +00:00
if (!global || !global->global_power.button || !dent || !dent->new_link || !dent->d_parent || !(parent_path = dent->d_parent->d_path)) {
MARS_DBG("cannot work\n");
2011-02-23 20:48:06 +00:00
return 0;
}
2012-02-03 10:25:17 +00:00
mypeer = dent->d_rest;
2011-02-23 20:48:06 +00:00
if (!mypeer) {
status = _parse_args(dent, dent->new_link, 1);
if (status < 0)
goto done;
mypeer = dent->d_argv[0];
}
MARS_DBG("peer '%s'\n", mypeer);
if (!dent->d_private) {
2011-08-12 11:09:48 +00:00
dent->d_private = brick_zmem_alloc(sizeof(struct mars_peerinfo));
2011-02-23 20:48:06 +00:00
if (!dent->d_private) {
MARS_ERR("no memory for peer structure\n");
2012-02-03 10:25:17 +00:00
status = -ENOMEM;
goto done;
2011-02-23 20:48:06 +00:00
}
peer = dent->d_private;
2011-02-24 16:37:32 +00:00
peer->global = global;
2011-08-31 11:42:04 +00:00
peer->peer = brick_strdup(mypeer);
peer->path = brick_strdup(path);
2011-02-23 20:48:06 +00:00
peer->maxdepth = 2;
2011-03-22 14:36:26 +00:00
spin_lock_init(&peer->lock);
INIT_LIST_HEAD(&peer->remote_dent_list);
2011-02-23 20:48:06 +00:00
}
2011-03-22 14:36:26 +00:00
2011-02-23 20:48:06 +00:00
peer = dent->d_private;
2011-08-25 10:16:32 +00:00
if (!peer->peer_thread) {
peer->peer_thread = brick_thread_create(peer_thread, peer, "mars_peer%d", serial++);
if (unlikely(!peer->peer_thread)) {
MARS_ERR("cannot start peer thread\n");
2011-02-23 20:48:06 +00:00
return -1;
}
MARS_DBG("started peer thread\n");
2011-02-23 20:48:06 +00:00
}
2011-09-27 08:46:14 +00:00
/* This must be called by the main thread in order to
* avoid nasty races.
* The peer thread does nothing but fetching the dent list.
*/
2011-03-22 14:36:26 +00:00
status = run_bones(peer);
2011-02-23 20:48:06 +00:00
done:
return status;
}
2011-02-28 18:00:32 +00:00
static int kill_scan(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
return _kill_peer(buf, dent);
}
2011-02-28 18:00:32 +00:00
static int make_scan(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
2011-02-24 16:37:32 +00:00
MARS_DBG("path = '%s' peer = '%s'\n", dent->d_path, dent->d_rest);
2011-03-22 14:36:26 +00:00
// don't connect to myself
2011-02-24 16:37:32 +00:00
if (!strcmp(dent->d_rest, my_id())) {
return 0;
}
2012-02-03 10:25:17 +00:00
return _make_peer(buf, dent, "/mars");
2011-02-23 20:48:06 +00:00
}
static
2011-08-25 10:16:32 +00:00
int kill_any(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
2011-08-25 10:16:32 +00:00
struct list_head *tmp;
2011-02-23 20:48:06 +00:00
if (global->global_power.button || !is_shutdown()) {
2011-02-23 20:48:06 +00:00
return 0;
}
2011-08-25 10:16:32 +00:00
for (tmp = dent->brick_list.next; tmp != &dent->brick_list; tmp = tmp->next) {
struct mars_brick *brick = container_of(tmp, struct mars_brick, dent_brick_link);
if (brick->nr_outputs > 0 && brick->outputs[0] && brick->outputs[0]->nr_connected) {
MARS_DBG("cannot kill dent '%s' because brick '%s' is wired\n", dent->d_path, brick->brick_path);
return 0;
}
}
2011-03-01 18:00:14 +00:00
MARS_DBG("killing dent = '%s'\n", dent->d_path);
2011-03-23 08:09:00 +00:00
mars_kill_dent(dent);
2011-09-27 11:57:04 +00:00
return 1;
2011-02-23 20:48:06 +00:00
}
///////////////////////////////////////////////////////////////////////
// handlers / helpers for logfile rotation
static
2011-03-23 17:58:02 +00:00
void _create_new_logfile(const char *path)
2011-02-23 20:48:06 +00:00
{
struct file *f;
const int flags = O_RDWR | O_CREAT | O_EXCL;
const int prot = 0600;
mm_segment_t oldfs;
oldfs = get_fs();
set_fs(get_ds());
f = filp_open(path, flags, prot);
set_fs(oldfs);
2011-03-03 09:02:10 +00:00
if (IS_ERR(f)) {
2011-03-25 10:34:54 +00:00
int err = PTR_ERR(f);
if (err == -EEXIST) {
MARS_INF("logfile '%s' already exists\n", path);
} else {
MARS_ERR("could not create logfile '%s' status = %d\n", path, err);
}
2011-03-03 09:02:10 +00:00
} else {
2011-02-23 20:48:06 +00:00
MARS_DBG("created empty logfile '%s'\n", path);
2011-03-03 09:02:10 +00:00
filp_close(f, NULL);
2011-02-23 20:48:06 +00:00
mars_trigger();
}
}
2011-10-05 14:59:29 +00:00
static
int _check_versionlink(struct mars_global *global, const char *parent_path, int sequence, loff_t target_end_pos)
2011-07-15 10:12:06 +00:00
{
char *my_version = NULL;
char *my_version_link = NULL;
char *other_version = NULL;
char *other_version_link = NULL;
char *from_host = NULL;
2012-12-17 08:35:49 +00:00
loff_t pos = 0;
2011-07-15 10:12:06 +00:00
int status = -ENOMEM;
my_version = path_make("%s/version-%09d-%s", parent_path, sequence, my_id());
if (!my_version) {
MARS_WRN("out of memory");
2011-07-15 10:12:06 +00:00
goto out;
}
status = -ENOENT;
my_version_link = mars_readlink(my_version);
if (!my_version_link) {
MARS_WRN("cannot find my own version symlink '%s'\n", my_version);
2011-07-20 13:11:44 +00:00
goto out;
}
2012-12-17 08:35:49 +00:00
from_host = _parse_versionlink(my_version_link, &pos);
if (!from_host) {
MARS_WRN("cannot parse '%s'\n", my_version_link);
goto out;
}
if (!strcmp(from_host, my_id())) {
MARS_DBG("found version stemming from myself, no check of other version necessary.\n");
status = 1;
2012-12-17 08:35:49 +00:00
if (unlikely(pos != target_end_pos)) {
MARS_WRN("pos = %lld != target_end_pos = %lld\n", pos, target_end_pos);
status = 0;
}
goto out;
}
status = -ENOMEM;
other_version = path_make("%s/version-%09d-%s", parent_path, sequence, from_host);
if (!other_version) {
MARS_WRN("out of memory");
goto out;
}
status = -ENOENT;
other_version_link = mars_readlink(other_version);
if (!other_version_link) {
MARS_WRN("cannot find other version symlink '%s'\n", other_version);
goto out;
}
if (!strcmp(my_version_link, other_version_link)) {
MARS_DBG("VERSION OK '%s'\n", my_version_link);
status = 1;
2011-07-15 10:12:06 +00:00
} else {
MARS_WRN("VERSION MISMATCH '%s' != '%s' => check for SPLIT BRAIN!\n", my_version_link, other_version_link);
status = 0;
2011-07-15 10:12:06 +00:00
}
out:
brick_string_free(my_version);
brick_string_free(my_version_link);
brick_string_free(other_version);
brick_string_free(other_version_link);
brick_string_free(from_host);
2011-07-20 13:11:44 +00:00
return status;
}
static
void rot_destruct(void *_rot)
{
struct mars_rotate *rot = _rot;
if (likely(rot)) {
brick_string_free(rot->copy_path);
brick_string_free(rot->parent_path);
brick_string_free(rot->inf_prev_version);
rot->copy_path = NULL;
rot->parent_path = NULL;
rot->inf_prev_version = NULL;
}
}
2011-02-23 20:48:06 +00:00
/* This must be called once at every round of logfile checking.
*/
static
2011-07-08 07:02:14 +00:00
int make_log_init(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
2011-07-08 07:02:14 +00:00
struct mars_dent *parent = dent->d_parent;
2011-02-23 20:48:06 +00:00
struct mars_brick *aio_brick;
struct mars_brick *trans_brick;
struct mars_rotate *rot = parent->d_private;
2011-02-28 18:00:32 +00:00
struct mars_dent *replay_link;
struct mars_dent *aio_dent;
2011-02-23 20:48:06 +00:00
struct mars_output *output;
const char *parent_path;
2011-03-23 17:58:02 +00:00
const char *replay_path = NULL;
const char *aio_path = NULL;
const char *switch_path = NULL;
int status = 0;
2011-02-23 20:48:06 +00:00
2011-08-25 10:16:32 +00:00
if (!global->global_power.button) {
goto done;
2011-08-25 10:16:32 +00:00
}
status = -EINVAL;
CHECK_PTR(parent, done);
parent_path = parent->d_path;
CHECK_PTR(parent_path, done);
2011-08-25 10:16:32 +00:00
2011-02-23 20:48:06 +00:00
if (!rot) {
2012-02-21 11:23:38 +00:00
const char *copy_path;
2011-08-12 11:09:48 +00:00
rot = brick_zmem_alloc(sizeof(struct mars_rotate));
2012-02-21 11:23:38 +00:00
if (unlikely(!rot)) {
2011-02-23 20:48:06 +00:00
MARS_ERR("cannot allocate rot structure\n");
status = -ENOMEM;
goto done;
}
sema_init(&rot->inf_mutex, 1);
2012-02-21 11:23:38 +00:00
copy_path = path_make("%s/logfile-update", parent_path);
if (unlikely(!copy_path)) {
MARS_ERR("cannot create copy_path\n");
brick_mem_free(rot);
status = -ENOMEM;
goto done;
}
rot->copy_path = copy_path;
2011-03-01 18:00:14 +00:00
rot->global = global;
2012-02-21 11:23:38 +00:00
parent->d_private = rot;
parent->d_private_destruct = rot_destruct;
2011-02-23 20:48:06 +00:00
}
rot->replay_link = NULL;
rot->aio_dent = NULL;
rot->aio_brick = NULL;
2012-03-06 11:56:48 +00:00
rot->first_log = NULL;
2011-02-23 20:48:06 +00:00
rot->relevant_log = NULL;
2011-03-01 18:00:14 +00:00
rot->relevant_brick = NULL;
2011-06-17 11:32:38 +00:00
rot->next_relevant_log = NULL;
rot->next_next_relevant_log = NULL;
2011-02-23 20:48:06 +00:00
rot->prev_log = NULL;
rot->next_log = NULL;
rot->max_sequence = 0;
rot->has_error = false;
if (dent->new_link)
sscanf(dent->new_link, "%lld", &rot->dev_size);
if (!rot->parent_path)
rot->parent_path =brick_strdup(parent_path);
mars_remaining_space(parent_path, &rot->total_space, &rot->remaining_space);
2011-02-23 20:48:06 +00:00
/* Fetch the replay status symlink.
* It must exist, and its value will control everything.
*/
replay_path = path_make("%s/replay-%s", parent_path, my_id());
2011-03-01 09:34:36 +00:00
if (unlikely(!replay_path)) {
MARS_ERR("cannot make path\n");
status = -ENOMEM;
goto done;
}
2011-02-23 20:48:06 +00:00
2011-03-01 09:34:36 +00:00
replay_link = (void*)mars_find_dent(global, replay_path);
if (unlikely(!replay_link || !replay_link->new_link)) {
2011-03-22 14:36:26 +00:00
MARS_DBG("replay status symlink '%s' does not exist (%p)\n", replay_path, replay_link);
rot->allow_update = false;
2011-02-23 20:48:06 +00:00
status = -ENOENT;
goto done;
}
2011-05-13 11:19:28 +00:00
status = _parse_args(replay_link, replay_link->new_link, 3);
2011-03-01 09:34:36 +00:00
if (unlikely(status < 0)) {
2011-02-23 20:48:06 +00:00
goto done;
}
rot->replay_link = replay_link;
2011-11-15 17:32:20 +00:00
/* Fetch AIO dentry of the logfile.
2011-02-23 20:48:06 +00:00
*/
if (rot->trans_brick) {
struct trans_logger_input *trans_input = rot->trans_brick->inputs[rot->trans_brick->old_input_nr];
if (trans_input && trans_input->is_operating && trans_input->inf.inf_host) {
aio_path = path_make("%s/log-%09d-%s", parent_path, trans_input->inf.inf_sequence, trans_input->inf.inf_host);
MARS_DBG("using logfile '%s' from trans_input %d (new=%d)\n", SAFE_STR(aio_path), rot->trans_brick->old_input_nr, rot->trans_brick->log_input_nr);
2011-11-15 17:32:20 +00:00
}
}
if (!aio_path) {
aio_path = path_make("%s/%s", parent_path, replay_link->d_argv[0]);
MARS_DBG("using logfile '%s' from replay symlink\n", SAFE_STR(aio_path));
}
2011-03-01 09:34:36 +00:00
if (unlikely(!aio_path)) {
MARS_ERR("cannot make path\n");
status = -ENOMEM;
goto done;
}
2011-02-23 20:48:06 +00:00
2011-03-01 09:34:36 +00:00
aio_dent = (void*)mars_find_dent(global, aio_path);
2011-02-23 20:48:06 +00:00
if (unlikely(!aio_dent)) {
2011-03-22 14:36:26 +00:00
MARS_DBG("logfile '%s' does not exist\n", aio_path);
2011-02-23 20:48:06 +00:00
status = -ENOENT;
2011-09-27 08:46:14 +00:00
if (rot->todo_primary) { // try to create an empty logfile
2011-03-01 09:34:36 +00:00
_create_new_logfile(aio_path);
2011-02-23 20:48:06 +00:00
}
goto done;
}
rot->aio_dent = aio_dent;
/* Fetch / make the AIO brick instance
*/
2011-03-01 09:34:36 +00:00
aio_brick =
make_brick_all(global,
2011-03-02 09:30:56 +00:00
aio_dent,
2011-09-02 12:17:40 +00:00
false,
_set_aio_params,
2011-03-11 13:57:54 +00:00
NULL,
2011-03-02 09:30:56 +00:00
10 * HZ,
2011-03-01 09:34:36 +00:00
aio_path,
2011-03-07 10:27:38 +00:00
(const struct generic_brick_type*)&aio_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){},
2011-03-23 08:09:00 +00:00
NULL,
1, // start always
2011-11-15 17:32:20 +00:00
"%s",
2011-03-01 09:34:36 +00:00
(const char *[]){},
0,
2011-11-15 17:32:20 +00:00
aio_path);
2011-02-23 20:48:06 +00:00
if (!aio_brick) {
2011-03-01 09:34:36 +00:00
MARS_ERR("cannot access '%s'\n", aio_path);
status = -EIO;
goto done;
2011-02-23 20:48:06 +00:00
}
rot->aio_brick = (void*)aio_brick;
/* Fetch the actual logfile size
*/
output = aio_brick->outputs[0];
status = output->ops->mars_get_info(output, &rot->aio_info);
if (status < 0) {
2011-03-01 09:34:36 +00:00
MARS_ERR("cannot get info on '%s'\n", aio_path);
2011-02-23 20:48:06 +00:00
goto done;
}
2011-03-01 09:34:36 +00:00
MARS_DBG("logfile '%s' size = %lld\n", aio_path, rot->aio_info.current_size);
2011-02-23 20:48:06 +00:00
2012-02-28 11:02:35 +00:00
#if defined(CONFIG_MARS_LOGROT_AUTO) && CONFIG_MARS_LOGROT_AUTO > 0
if (rot->is_primary &&
unlikely(rot->aio_info.current_size >= (loff_t)CONFIG_MARS_LOGROT_AUTO * 1024 * 1024 * 1024)) {
char *new_path = path_make("%s/log-%09d-%s", parent_path, aio_dent->d_serial + 1, my_id());
if (likely(new_path && !mars_find_dent(global, new_path))) {
MARS_INF("old logfile size = %lld, creating new logfile '%s'\n", rot->aio_info.current_size, new_path);
_create_new_logfile(new_path);
}
brick_string_free(new_path);
}
#endif
2011-03-23 17:58:02 +00:00
// check whether attach is allowed
switch_path = path_make("%s/todo-%s/attach", parent_path, my_id());
2011-03-23 17:58:02 +00:00
2011-03-01 09:34:36 +00:00
/* Fetch / make the transaction logger.
* We deliberately "forget" to connect the log input here.
2011-08-25 10:16:32 +00:00
* Will be carried out later in make_log_step().
2011-03-01 09:34:36 +00:00
* The final switch-on will be started in make_log_finalize().
2011-02-23 20:48:06 +00:00
*/
2011-03-01 09:34:36 +00:00
trans_brick =
make_brick_all(global,
2011-08-25 10:16:32 +00:00
dent,
2011-09-02 12:17:40 +00:00
false,
2011-03-11 13:57:54 +00:00
_set_trans_params,
NULL,
2011-03-01 09:34:36 +00:00
0,
aio_path,
(const struct generic_brick_type*)&trans_logger_brick_type,
2011-03-18 13:15:40 +00:00
(const struct generic_brick_type*[]){NULL},
2011-03-23 17:58:02 +00:00
switch_path,
0, // let switch decide
2011-03-03 09:02:10 +00:00
"%s/logger",
(const char *[]){"%s/data-%s"},
2011-03-01 09:34:36 +00:00
1,
parent_path,
parent_path,
2011-03-01 09:34:36 +00:00
my_id());
status = -ENOENT;
2011-02-23 20:48:06 +00:00
if (!trans_brick) {
2011-03-01 09:34:36 +00:00
goto done;
2011-02-23 20:48:06 +00:00
}
rot->trans_brick = (void*)trans_brick;
rot->trans_brick->replay_limiter = &rot->replay_limiter;
2011-02-23 20:48:06 +00:00
/* For safety, default is to try an (unnecessary) replay in case
* something goes wrong later.
*/
2012-08-03 08:42:51 +00:00
rot->replay_mode = true;
2011-02-23 20:48:06 +00:00
status = 0;
done:
2011-10-05 14:59:29 +00:00
brick_string_free(aio_path);
brick_string_free(replay_path);
brick_string_free(switch_path);
2011-02-23 20:48:06 +00:00
return status;
}
2011-06-17 11:32:38 +00:00
/* Note: this is strictly called in d_serial order.
* This is important!
2011-02-23 20:48:06 +00:00
*/
static
2011-08-25 10:16:32 +00:00
int make_log_step(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
2011-06-17 11:32:38 +00:00
struct mars_global *global = buf;
2011-02-28 18:00:32 +00:00
struct mars_dent *parent = dent->d_parent;
struct mars_rotate *rot;
2011-06-17 11:32:38 +00:00
struct trans_logger_brick *trans_brick;
struct mars_dent *prev_log;
2011-02-23 20:48:06 +00:00
int status = -EINVAL;
CHECK_PTR(parent, err);
rot = parent->d_private;
2012-12-03 08:43:26 +00:00
if (!rot)
goto err;
2011-06-17 11:32:38 +00:00
CHECK_PTR(rot, err);
2011-02-23 20:48:06 +00:00
status = 0;
2011-06-17 11:32:38 +00:00
trans_brick = rot->trans_brick;
if (!global->global_power.button || !dent->d_parent || !trans_brick || rot->has_error) {
MARS_DBG("nothing to do rot_error = %d\n", rot->has_error);
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-17 11:32:38 +00:00
/* Check for consecutiveness of logfiles
*/
prev_log = rot->next_log;
if (prev_log && prev_log->d_serial + 1 != dent->d_serial) {
MARS_WRN("transaction logs are not consecutive at '%s' (%d ~> %d)\n", dent->d_path, prev_log->d_serial, dent->d_serial);
2011-06-17 11:32:38 +00:00
status = -EINVAL;
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-17 11:32:38 +00:00
if (dent->d_serial > rot->max_sequence) {
rot->max_sequence = dent->d_serial;
}
2012-03-06 11:56:48 +00:00
if (!rot->first_log)
rot->first_log = dent;
2011-06-17 11:32:38 +00:00
/* Skip any logfiles after the relevant one.
* This should happen only when replaying multiple logfiles
* in sequence, or when starting a new logfile for writing.
*/
status = 0;
if (rot->relevant_log) {
if (!rot->next_relevant_log) {
rot->next_relevant_log = dent;
} else if (!rot->next_next_relevant_log) {
rot->next_next_relevant_log = dent;
2011-06-17 11:32:38 +00:00
}
MARS_DBG("next_relevant_log = %p next_next_relevant_log = %p\n", rot->next_relevant_log, rot->next_next_relevant_log);
2011-06-17 11:32:38 +00:00
goto ok;
}
/* Preconditions
*/
if (!rot->replay_link || !rot->aio_dent || !rot->aio_brick) {
MARS_DBG("nothing to do on '%s'\n", dent->d_path);
2011-06-17 11:32:38 +00:00
goto ok;
}
2011-10-05 14:59:29 +00:00
/* Remember the relevant log.
2011-06-17 11:32:38 +00:00
*/
if (rot->aio_dent->d_serial == dent->d_serial) {
2012-03-06 11:56:48 +00:00
rot->relevant_serial = dent->d_serial;
2011-06-17 11:32:38 +00:00
rot->relevant_log = dent;
}
ok:
/* All ok: switch over the indicators.
*/
MARS_DBG("next_log = '%s'\n", dent->d_path);
2011-06-17 11:32:38 +00:00
rot->prev_log = rot->next_log;
rot->next_log = dent;
done:
if (status < 0) {
MARS_DBG("rot_error status = %d\n", status);
rot->has_error = true;
}
err:
return status;
}
/* Internal helper. Return codes:
* ret < 0 : error
* ret == 0 : not relevant
* ret == 1 : relevant, no transaction replay, switch to the next
* ret == 2 : relevant for transaction replay
* ret == 3 : relevant for appending
*/
static
int _check_logging_status(struct mars_rotate *rot, long long *oldpos_start, long long *oldpos_end, long long *newpos)
{
struct mars_dent *dent = rot->relevant_log;
struct mars_dent *parent;
2011-09-27 08:46:14 +00:00
struct mars_global *global = NULL;
2011-06-17 11:32:38 +00:00
int status = 0;
if (!dent)
goto done;
status = -EINVAL;
parent = dent->d_parent;
CHECK_PTR(parent, done);
global = rot->global;
2011-08-31 11:42:04 +00:00
CHECK_PTR_NULL(global, done);
2011-06-17 11:32:38 +00:00
CHECK_PTR(rot->replay_link, done);
CHECK_PTR(rot->aio_brick, done);
2011-05-13 11:19:28 +00:00
if (sscanf(rot->replay_link->d_argv[1], "%lld", oldpos_start) != 1) {
MARS_ERR("bad start position argument '%s'\n", rot->replay_link->d_argv[1]);
goto done;
}
if (sscanf(rot->replay_link->d_argv[2], "%lld", oldpos_end) != 1) {
MARS_ERR("bad end position argument '%s'\n", rot->replay_link->d_argv[2]);
2011-02-23 20:48:06 +00:00
goto done;
}
2011-05-13 11:19:28 +00:00
*oldpos_end += *oldpos_start;
if (unlikely(*oldpos_end < *oldpos_start)) {
MARS_ERR("end_pos %lld < start_pos %lld\n", *oldpos_end, *oldpos_start);
// safety: use the smaller value, it does not hurt
*oldpos_start = *oldpos_end;
if (unlikely(*oldpos_start < 0))
*oldpos_start = 0;
2011-05-13 11:19:28 +00:00
}
2011-02-23 20:48:06 +00:00
*newpos = rot->aio_info.current_size;
2011-05-13 11:19:28 +00:00
if (unlikely(rot->aio_info.current_size < *oldpos_start)) {
MARS_ERR("oops, bad replay position attempted at logfile '%s' (file length %lld should never be smaller than requested position %lld, is your filesystem corrupted?) => please repair this by hand\n", rot->aio_dent->d_path, rot->aio_info.current_size, *oldpos_start);
2011-06-17 11:32:38 +00:00
status = -EBADF;
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-17 11:32:38 +00:00
status = 0;
if (rot->aio_info.current_size > *oldpos_start) {
2011-05-13 11:19:28 +00:00
MARS_DBG("transaction log replay is necessary on '%s' from %lld to %lld (dirty region ends at %lld)\n", rot->aio_dent->d_path, *oldpos_start, rot->aio_info.current_size, *oldpos_end);
2011-02-23 20:48:06 +00:00
status = 2;
2011-06-17 11:32:38 +00:00
} else if (rot->next_relevant_log) {
MARS_DBG("transaction log '%s' is already applied, and the next one is available for switching\n", rot->aio_dent->d_path);
2011-02-23 20:48:06 +00:00
status = 1;
2011-09-27 08:46:14 +00:00
} else if (rot->todo_primary) {
2011-10-05 14:59:29 +00:00
if (rot->aio_info.current_size > 0 || strcmp(dent->d_rest, my_id()) != 0) {
2011-06-17 11:32:38 +00:00
MARS_DBG("transaction log '%s' is already applied (would be usable for appending at position %lld, but a fresh logfile will be used for safety reasons)\n", rot->aio_dent->d_path, *oldpos_end);
status = 1;
} else {
MARS_DBG("empty transaction log '%s' is usable for me as a primary node\n", rot->aio_dent->d_path);
status = 3;
}
2011-02-23 20:48:06 +00:00
} else {
2011-06-17 11:32:38 +00:00
MARS_DBG("transaction log '%s' is the last one, currently fully applied\n", rot->aio_dent->d_path);
status = 0;
2011-02-23 20:48:06 +00:00
}
done:
return status;
}
static
2011-06-17 11:32:38 +00:00
int _make_logging_status(struct mars_rotate *rot)
2011-02-23 20:48:06 +00:00
{
2011-06-17 11:32:38 +00:00
struct mars_dent *dent = rot->relevant_log;
struct mars_dent *parent;
2011-09-27 08:46:14 +00:00
struct mars_global *global = NULL;
2011-02-23 20:48:06 +00:00
struct trans_logger_brick *trans_brick;
loff_t start_pos = 0;
2011-05-13 11:19:28 +00:00
loff_t dirty_pos = 0;
2011-02-23 20:48:06 +00:00
loff_t end_pos = 0;
2011-06-17 11:32:38 +00:00
int status = 0;
2011-02-23 20:48:06 +00:00
2011-06-17 11:32:38 +00:00
if (!dent)
goto done;
status = -EINVAL;
parent = dent->d_parent;
CHECK_PTR(parent, done);
global = rot->global;
2011-08-31 11:42:04 +00:00
CHECK_PTR_NULL(global, done);
2011-02-23 20:48:06 +00:00
status = 0;
trans_brick = rot->trans_brick;
2011-06-17 11:32:38 +00:00
if (!global->global_power.button || !trans_brick || rot->has_error) {
2011-03-01 09:34:36 +00:00
MARS_DBG("nothing to do rot_error = %d\n", rot->has_error);
2011-02-23 20:48:06 +00:00
goto done;
}
/* Find current logging status.
*/
2011-06-17 11:32:38 +00:00
status = _check_logging_status(rot, &start_pos, &dirty_pos, &end_pos);
2011-02-23 20:48:06 +00:00
if (status < 0) {
goto done;
}
if (unlikely(start_pos < 0 || dirty_pos < start_pos || end_pos < dirty_pos)) {
MARS_ERR("implausible values: start_pos = %lld dirty_pos = %lld end_pos = %lld\n", start_pos, dirty_pos, end_pos);
}
2011-02-23 20:48:06 +00:00
/* Relevant or not?
*/
switch (status) {
case 0: // not relevant
goto ok;
2011-04-08 09:52:46 +00:00
case 1: /* Relevant, and transaction replay already finished.
2011-06-17 11:32:38 +00:00
* Allow switching over to a new logfile.
2011-03-07 05:55:10 +00:00
*/
2011-10-05 14:59:29 +00:00
if (!trans_brick->power.button && !trans_brick->power.led_on && trans_brick->power.led_off) {
if (rot->next_relevant_log) {
MARS_DBG("check switchover from '%s' to '%s' (size = %lld, next_next = %p, allow_replay = %d)\n", dent->d_path, rot->next_relevant_log->d_path, rot->next_relevant_log->new_stat.size, rot->next_next_relevant_log, rot->allow_replay);
2012-02-02 16:25:50 +00:00
if ((rot->next_relevant_log->new_stat.size > 0 || rot->next_next_relevant_log || (long long)jiffies > rot->switchover_timeout + 30 * HZ) &&
rot->allow_replay &&
_check_versionlink(global, parent->d_path, dent->d_serial, end_pos) > 0) {
2012-02-02 16:25:50 +00:00
rot->switchover_timeout = 0;
MARS_DBG("switching over from '%s' to next relevant transaction log '%s'\n", dent->d_path, rot->next_relevant_log->d_path);
_make_new_replaylink(rot, rot->next_relevant_log->d_rest, rot->next_relevant_log->d_serial, rot->next_relevant_log->new_stat.size);
2012-02-02 16:25:50 +00:00
} else {
MARS_DBG("waiting for logfile to become stable\n");
if (!rot->switchover_timeout)
rot->switchover_timeout = jiffies;
2011-10-05 14:59:29 +00:00
}
} else if (rot->todo_primary) {
MARS_DBG("preparing new transaction log '%s' from version %d to %d\n", dent->d_path, dent->d_serial, dent->d_serial + 1);
_make_new_replaylink(rot, my_id(), dent->d_serial + 1, 0);
2011-10-05 14:59:29 +00:00
} else {
MARS_DBG("nothing to do on last transaction log '%s'\n", dent->d_path);
}
2011-02-23 20:48:06 +00:00
}
status = -EAGAIN;
goto done;
case 2: // relevant for transaction replay
MARS_DBG("replaying transaction log '%s' from %lld to %lld\n", dent->d_path, start_pos, end_pos);
2012-08-03 08:42:51 +00:00
rot->replay_mode = true;
2011-02-23 20:48:06 +00:00
rot->start_pos = start_pos;
rot->end_pos = end_pos;
break;
case 3: // relevant for appending
MARS_DBG("appending to transaction log '%s'\n", dent->d_path);
2012-08-03 08:42:51 +00:00
rot->replay_mode = false;
2011-02-23 20:48:06 +00:00
rot->start_pos = 0;
rot->end_pos = 0;
break;
default:
MARS_ERR("bad internal status %d\n", status);
status = -EINVAL;
goto done;
}
ok:
/* All ok: switch over the indicators.
*/
rot->prev_log = rot->next_log;
rot->next_log = dent;
done:
if (status < 0) {
2011-03-01 18:00:14 +00:00
MARS_DBG("rot_error status = %d\n", status);
2011-02-23 20:48:06 +00:00
rot->has_error = true;
}
return status;
}
static
void _init_trans_input(struct trans_logger_input *trans_input, struct mars_dent *log_dent, struct mars_rotate *rot)
{
if (unlikely(trans_input->connect || trans_input->is_operating)) {
2011-11-15 17:32:20 +00:00
MARS_ERR("this should not happen\n");
return;
}
brick_string_free(trans_input->inf.inf_host); // just for safety
2011-11-15 17:32:20 +00:00
memset(&trans_input->inf, 0, sizeof(trans_input->inf));
2011-11-15 17:32:20 +00:00
trans_input->inf.inf_host = brick_strdup(log_dent->d_rest);
trans_input->inf.inf_sequence = log_dent->d_serial;
trans_input->inf.inf_private = rot;
trans_input->inf.inf_callback = _update_info;
MARS_DBG("initialized '%s' %d\n", trans_input->inf.inf_host, trans_input->inf.inf_sequence);
}
#ifdef CONFIG_MARS_LOGROT
static
int _get_free_input(struct trans_logger_brick *trans_brick)
{
int nr = (((trans_brick->log_input_nr - TL_INPUT_LOG1) + 1) % 2) + TL_INPUT_LOG1;
struct trans_logger_input *candidate;
candidate = trans_brick->inputs[nr];
if (unlikely(!candidate)) {
MARS_ERR("input nr = %d is corrupted!\n", nr);
return -EEXIST;
}
if (unlikely(candidate->is_operating || candidate->connect)) {
MARS_DBG("nr = %d unusable! is_operating = %d connect = %p\n", nr, candidate->is_operating, candidate->connect);
return -EEXIST;
}
MARS_DBG("got nr = %d\n", nr);
return nr;
}
static
void _rotate_trans(struct mars_rotate *rot)
2011-02-23 20:48:06 +00:00
{
struct trans_logger_brick *trans_brick = rot->trans_brick;
int old_nr = trans_brick->old_input_nr;
int log_nr = trans_brick->log_input_nr;
int next_nr;
MARS_DBG("log_input_nr = %d old_input_nr = %d next_relevant_log = %p\n", log_nr, old_nr, rot->next_relevant_log);
// try to cleanup old log
if (log_nr != old_nr) {
struct trans_logger_input *trans_input = trans_brick->inputs[old_nr];
if (!trans_input->connect) {
MARS_DBG("ignoring unused input %d\n", old_nr);
} else if (trans_input->is_operating &&
trans_input->inf.inf_min_pos == trans_input->inf.inf_max_pos &&
list_empty(&trans_input->pos_list)) {
int status;
MARS_INF("cleanup old transaction log (%d -> %d)\n", old_nr, log_nr);
status = generic_disconnect((void*)trans_input);
if (unlikely(status < 0)) {
MARS_ERR("disconnect failed\n");
} else {
2012-01-17 14:37:14 +00:00
mars_remote_trigger();
}
} else {
MARS_DBG("old transaction replay not yet finished: is_operating = %d pos %lld != %lld\n",
trans_input->is_operating,
trans_input->inf.inf_min_pos,
trans_input->inf.inf_max_pos);
}
} else
// try to setup new log
if (log_nr == trans_brick->new_input_nr &&
rot->next_relevant_log &&
(next_nr = _get_free_input(trans_brick)) >= 0) {
struct trans_logger_input *trans_input;
int status;
MARS_DBG("start switchover %d -> %d\n", old_nr, next_nr);
rot->next_relevant_brick =
make_brick_all(rot->global,
rot->next_relevant_log,
false,
_set_aio_params,
NULL,
10 * HZ,
rot->next_relevant_log->d_path,
(const struct generic_brick_type*)&aio_brick_type,
(const struct generic_brick_type*[]){},
NULL,
0, // let switch decide
rot->next_relevant_log->d_path,
(const char *[]){},
0);
if (unlikely(!rot->next_relevant_brick)) {
MARS_ERR("could not open next transaction log '%s'\n", rot->next_relevant_log->d_path);
goto done;
}
trans_input = trans_brick->inputs[next_nr];
if (unlikely(!trans_input)) {
MARS_ERR("log input does not exist\n");
goto done;
}
2011-02-23 20:48:06 +00:00
_init_trans_input(trans_input, rot->next_relevant_log, rot);
status = generic_connect((void*)trans_input, (void*)rot->next_relevant_brick->outputs[0]);
if (unlikely(status < 0)) {
MARS_ERR("connect failed\n");
goto done;
}
trans_brick->new_input_nr = next_nr;
MARS_INF("started switchover to '%s'\n", rot->next_relevant_log->d_path);
}
done: ;
}
#endif
static
void _change_trans(struct mars_rotate *rot)
{
struct trans_logger_brick *trans_brick = rot->trans_brick;
2012-08-03 08:42:51 +00:00
MARS_DBG("replay_mode = %d start_pos = %lld end_pos = %lld\n", trans_brick->replay_mode, rot->start_pos, rot->end_pos);
2011-07-11 14:01:28 +00:00
2012-08-03 08:42:51 +00:00
if (trans_brick->replay_mode) {
2011-06-01 12:55:06 +00:00
trans_brick->replay_start_pos = rot->start_pos;
trans_brick->replay_end_pos = rot->end_pos;
} else {
#ifdef CONFIG_MARS_LOGROT
_rotate_trans(rot);
#endif
2011-02-23 20:48:06 +00:00
}
2011-06-01 12:55:06 +00:00
}
static
int _start_trans(struct mars_rotate *rot)
{
struct trans_logger_brick *trans_brick;
2011-06-01 12:55:06 +00:00
struct trans_logger_input *trans_input;
int nr;
2011-06-01 12:55:06 +00:00
int status;
2011-02-23 20:48:06 +00:00
/* Internal safety checks
*/
status = -EINVAL;
if (unlikely(!rot)) {
MARS_ERR("rot is NULL\n");
2011-06-01 12:55:06 +00:00
goto done;
}
if (unlikely(!rot->aio_brick || !rot->relevant_log)) {
MARS_ERR("aio %p or relevant log %p is missing, this should not happen\n", rot->aio_brick, rot->relevant_log);
2011-06-01 12:55:06 +00:00
goto done;
}
trans_brick = rot->trans_brick;
if (unlikely(!trans_brick)) {
MARS_ERR("logger instance does not exist\n");
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-01 12:55:06 +00:00
/* Update status when already working
*/
if (trans_brick->power.button || !trans_brick->power.led_off) {
_change_trans(rot);
status = 0;
2011-03-01 18:00:14 +00:00
goto done;
}
2011-06-01 12:55:06 +00:00
/* Further safety checks.
2011-06-01 12:55:06 +00:00
*/
if (unlikely(rot->relevant_brick)) {
MARS_ERR("log aio brick already present, this should not happen\n");
2011-05-13 11:19:28 +00:00
goto done;
}
if (unlikely(trans_brick->inputs[TL_INPUT_LOG1]->is_operating || trans_brick->inputs[TL_INPUT_LOG2]->is_operating)) {
MARS_ERR("some input is operating, this should not happen\n");
goto done;
}
2011-02-23 20:48:06 +00:00
/* Allocate new input slot
2011-02-23 20:48:06 +00:00
*/
nr = _get_free_input(trans_brick);
if (unlikely(nr < TL_INPUT_LOG1 || nr > TL_INPUT_LOG2)) {
MARS_ERR("bad new_input_nr = %d\n", nr);
goto done;
}
trans_brick->new_input_nr = nr;
trans_brick->old_input_nr = nr;
trans_brick->log_input_nr = nr;
trans_input = trans_brick->inputs[nr];
if (unlikely(!trans_input)) {
MARS_ERR("log input %d does not exist\n", nr);
goto done;
2011-02-23 20:48:06 +00:00
}
2011-03-01 18:00:14 +00:00
/* Open new transaction log
*/
rot->relevant_brick =
make_brick_all(rot->global,
2011-03-02 09:30:56 +00:00
rot->relevant_log,
2011-09-02 12:17:40 +00:00
false,
_set_aio_params,
2011-03-11 13:57:54 +00:00
NULL,
2011-03-01 18:00:14 +00:00
10 * HZ,
rot->relevant_log->d_path,
2011-03-07 10:27:38 +00:00
(const struct generic_brick_type*)&aio_brick_type,
2011-03-01 18:00:14 +00:00
(const struct generic_brick_type*[]){},
2011-03-23 08:09:00 +00:00
NULL,
1, // start always
2011-03-01 18:00:14 +00:00
rot->relevant_log->d_path,
(const char *[]){},
0);
if (unlikely(!rot->relevant_brick)) {
MARS_ERR("log aio brick '%s' not open\n", rot->relevant_log->d_path);
2011-03-01 18:00:14 +00:00
goto done;
}
/* Supply all relevant parameters
*/
trans_brick->replay_mode = rot->replay_mode;
_init_trans_input(trans_input, rot->relevant_log, rot);
2011-02-23 20:48:06 +00:00
/* Connect to new transaction log
*/
2011-05-13 11:19:28 +00:00
status = generic_connect((void*)trans_input, (void*)rot->relevant_brick->outputs[0]);
if (unlikely(status < 0)) {
MARS_ERR("initial connect failed\n");
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-01 12:55:06 +00:00
_change_trans(rot);
2011-02-23 20:48:06 +00:00
/* Switch on....
*/
2011-03-23 17:58:02 +00:00
status = mars_power_button((void*)trans_brick, true, false);
2011-02-23 20:48:06 +00:00
MARS_DBG("status = %d\n", status);
done:
return status;
}
static
int _stop_trans(struct mars_rotate *rot, const char *parent_path)
2011-02-23 20:48:06 +00:00
{
struct trans_logger_brick *trans_brick = rot->trans_brick;
int status = 0;
2011-06-01 12:55:06 +00:00
if (!trans_brick) {
2011-02-23 20:48:06 +00:00
goto done;
}
2011-03-23 17:58:02 +00:00
/* Switch off temporarily....
2011-02-23 20:48:06 +00:00
*/
2011-03-23 17:58:02 +00:00
status = mars_power_button((void*)trans_brick, false, false);
2011-02-23 20:48:06 +00:00
MARS_DBG("status = %d\n", status);
2011-06-01 12:55:06 +00:00
if (status < 0) {
2011-02-23 20:48:06 +00:00
goto done;
}
2011-06-01 12:55:06 +00:00
/* Disconnect old connection(s)
2011-02-23 20:48:06 +00:00
*/
2011-06-01 12:55:06 +00:00
if (trans_brick->power.led_off) {
int i;
for (i = TL_INPUT_LOG1; i <= TL_INPUT_LOG2; i++) {
2011-06-01 12:55:06 +00:00
struct trans_logger_input *trans_input;
trans_input = trans_brick->inputs[i];
if (trans_input && !trans_input->is_operating) {
2011-11-15 17:32:20 +00:00
if (trans_input->connect)
(void)generic_disconnect((void*)trans_input);
2011-06-01 12:55:06 +00:00
}
}
2011-02-23 20:48:06 +00:00
}
done:
return status;
}
static
void override_all_syncstatus(struct mars_global *global, struct mars_rotate *rot, char *parent_path)
{
char *prefix = NULL;
struct mars_dent **table = NULL;
int count;
int i;
prefix = path_make("%s/syncstatus-", parent_path);
if (!prefix)
goto done;
count = mars_find_dent_all(global, prefix, &table);
MARS_DBG("prefix='%s' count=%d\n", prefix, count);
for (i = 0; i < count; i++) {
struct mars_dent *dent = table[i];
int status;
if (!dent || !dent->d_path || dent == rot->syncstatus_dent)
continue;
status = mars_symlink("0", dent->d_path, NULL, 0);
MARS_DBG("clearing syncstatus link '%s' status=%d\n", dent->d_path, status);
}
done:
if (table)
brick_mem_free(table);
if (prefix)
brick_string_free(prefix);
}
2011-02-23 20:48:06 +00:00
static
2011-08-25 10:16:32 +00:00
int make_log_finalize(struct mars_global *global, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
2011-08-25 10:16:32 +00:00
struct mars_dent *parent = dent->d_parent;
struct mars_rotate *rot;
2011-02-23 20:48:06 +00:00
struct trans_logger_brick *trans_brick;
2012-02-21 11:23:38 +00:00
struct copy_brick *copy_brick;
2011-02-23 20:48:06 +00:00
int status = -EINVAL;
CHECK_PTR(parent, err);
rot = parent->d_private;
2012-12-03 08:43:26 +00:00
if (!rot)
goto err;
CHECK_PTR(rot, err);
2011-02-23 20:48:06 +00:00
trans_brick = rot->trans_brick;
status = 0;
if (!trans_brick) {
MARS_DBG("nothing to do\n");
goto done;
}
2011-03-22 14:36:26 +00:00
/* Handle jamming (a very exceptional state)
*/
if (global->jammed) {
if (rot->todo_primary || rot->is_primary)
trans_brick->cease_logging = true;
} else if (!rot->todo_primary && !rot->is_primary) {
trans_brick->cease_logging = false;
}
if (trans_brick->cease_logging)
override_all_syncstatus(global, rot, parent->d_path);
2012-02-21 11:23:38 +00:00
// check whether some copy has finished
copy_brick = (struct copy_brick*)mars_find_brick(global, &copy_brick_type, rot->copy_path);
MARS_DBG("copy_path = '%s' copy_brick = %p\n", rot->copy_path, copy_brick);
if (copy_brick && (copy_brick->copy_last == copy_brick->copy_end || copy_brick->power.led_off)) {
status = mars_kill_brick((void*)copy_brick);
if (status < 0) {
MARS_ERR("could not kill copy_brick, status = %d\n", status);
goto done;
}
copy_brick = NULL;
}
rot->copy_brick = copy_brick;
if (!copy_brick)
rot->copy_serial = 0;
#if defined(CONFIG_MARS_LOGDELETE_AUTO)
2012-03-06 11:56:48 +00:00
#define LIMIT1 ((loff_t)EXHAUSTED_LIMIT(rot->total_space))
#define LIMIT2 ((loff_t)global_logdel_auto * 1024 * 1024)
2012-03-06 11:56:48 +00:00
if (rot->remaining_space <= LIMIT1 + LIMIT2) {
MARS_WRN("filesystem space = %lld kiB is lower than %lld + %lld = %lld\n", rot->remaining_space, LIMIT1, LIMIT2, LIMIT1 + LIMIT2);
if (rot->first_log && rot->first_log != rot->relevant_log) {
MARS_DBG("freeing old logfile '%s'\n", rot->first_log->d_path);
mars_unlink(rot->first_log->d_path);
rot->first_log->d_killme = true;
}
}
#endif
2011-02-23 20:48:06 +00:00
/* Stopping is also possible in case of errors
*/
if (trans_brick->power.button && trans_brick->power.led_on && !trans_brick->power.led_off) {
2011-05-13 11:19:28 +00:00
bool do_stop = true;
2012-08-03 08:42:51 +00:00
if (trans_brick->replay_mode) {
2011-11-08 17:07:42 +00:00
do_stop = trans_brick->replay_code != 0 || !_check_allow(global, parent, "allow-replay");
2011-05-13 11:19:28 +00:00
} else {
do_stop = !rot->is_primary;
2011-05-13 11:19:28 +00:00
}
2011-06-17 11:32:38 +00:00
2012-12-04 08:26:19 +00:00
MARS_DBG("replay_mode = %d replay_code = %d is_primary = %d do_stop = %d\n", trans_brick->replay_mode, trans_brick->replay_code, rot->is_primary, (int)do_stop);
2011-02-23 20:48:06 +00:00
if (do_stop) {
status = _stop_trans(rot, parent->d_path);
#ifdef CONFIG_MARS_LOGROT
} else {
_change_trans(rot);
#endif
2011-02-23 20:48:06 +00:00
}
goto done;
}
2011-06-17 11:32:38 +00:00
2011-05-13 11:19:28 +00:00
/* Starting is only possible when no error occurred.
2011-02-23 20:48:06 +00:00
*/
if (!rot->relevant_log || rot->has_error) {
MARS_DBG("nothing to do\n");
goto done;
}
/* Start when necessary
*/
if (!trans_brick->power.button && !trans_brick->power.led_on && trans_brick->power.led_off) {
2011-06-17 11:32:38 +00:00
bool do_start;
status = _make_logging_status(rot);
if (status <= 0) {
goto done;
}
2012-08-03 08:42:51 +00:00
do_start = (!rot->replay_mode ||
(rot->start_pos != rot->end_pos &&
rot->allow_replay &&
_check_allow(global, parent, "allow-replay")));
2012-08-03 08:42:51 +00:00
MARS_DBG("rot->replay_mode = %d rot->start_pos = %lld rot->end_pos = %lld rot->allow_replay = %d | do_start = %d\n", rot->replay_mode, rot->start_pos, rot->end_pos, rot->allow_replay, do_start);
2011-02-23 20:48:06 +00:00
if (do_start) {
status = _start_trans(rot);
}
}
done:
2012-08-02 12:59:55 +00:00
rot->allow_sync = (!rot->trans_brick || rot->trans_brick->power.led_off);
if (rot->trans_brick)
_show_rate(rot, &rot->replay_limiter, rot->trans_brick->power.led_on, "replay_rate");
if (rot->copy_brick)
_show_rate(rot, &rot->file_limiter, rot->copy_brick->power.led_on, "file_rate");
if (rot->sync_brick)
_show_rate(rot, &rot->sync_limiter, rot->sync_brick->power.led_on, "sync_rate");
err:
2011-02-23 20:48:06 +00:00
return status;
}
///////////////////////////////////////////////////////////////////////
// specific handlers
static
2011-02-28 18:00:32 +00:00
int make_primary(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
2011-08-25 10:16:32 +00:00
struct mars_global *global = buf;
2011-02-28 18:00:32 +00:00
struct mars_dent *parent;
2011-02-27 14:17:58 +00:00
struct mars_rotate *rot;
2011-02-23 20:48:06 +00:00
int status = -EINVAL;
2011-08-25 10:16:32 +00:00
if (!global->global_power.button) {
status = 0;
goto done;
}
2011-02-27 14:17:58 +00:00
parent = dent->d_parent;
CHECK_PTR(parent, done);
rot = parent->d_private;
2012-12-03 08:43:26 +00:00
if (!rot)
goto done;
2011-02-23 20:48:06 +00:00
CHECK_PTR(rot, done);
2011-09-27 08:46:14 +00:00
rot->todo_primary =
global->global_power.button && dent->new_link && !strcmp(dent->new_link, my_id());
2011-07-20 13:11:44 +00:00
rot->is_primary =
2011-09-27 08:46:14 +00:00
rot->if_brick && !rot->if_brick->power.led_off;
2011-02-23 20:48:06 +00:00
status = 0;
done:
return status;
}
static
2011-03-18 13:15:40 +00:00
int make_bio(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
struct mars_brick *brick;
2011-02-27 14:17:58 +00:00
int status = 0;
2011-02-23 20:48:06 +00:00
2011-03-23 08:09:00 +00:00
if (!global->global_power.button) {
2011-02-27 14:17:58 +00:00
goto done;
2011-02-23 20:48:06 +00:00
}
2012-08-08 09:50:24 +00:00
brick = mars_find_brick(global, NULL, dent->d_path);
if (brick) {
goto check;
2011-02-23 20:48:06 +00:00
}
2011-03-01 09:34:36 +00:00
brick =
make_brick_all(global,
2011-03-02 09:30:56 +00:00
dent,
2011-09-02 12:17:40 +00:00
false,
2011-03-18 13:15:40 +00:00
_set_bio_params,
2011-03-11 13:57:54 +00:00
NULL,
2011-03-01 09:34:36 +00:00
10 * HZ,
dent->d_path,
2011-03-18 13:15:40 +00:00
(const struct generic_brick_type*)&bio_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){},
2011-03-23 08:09:00 +00:00
NULL,
1, // start always
2011-03-01 09:34:36 +00:00
dent->d_path,
(const char *[]){},
0);
2011-02-27 14:17:58 +00:00
if (unlikely(!brick)) {
status = -ENXIO;
goto done;
}
2011-02-23 20:48:06 +00:00
brick->outputs[0]->output_name = dent->d_path;
2011-03-23 17:58:02 +00:00
status = mars_power_button((void*)brick, true, false);
2011-02-23 20:48:06 +00:00
if (status < 0) {
2011-08-25 10:16:32 +00:00
kill_any(buf, dent);
2012-08-08 09:50:24 +00:00
goto done;
}
check:
/* Report the actual size of the device.
* It may be larger than the global size.
*/
if (brick && brick->power.led_on && dent->d_parent) {
struct mars_info info = {};
struct mars_output *output;
char *src = NULL;
char *dst = NULL;
output = brick->outputs[0];
status = output->ops->mars_get_info(output, &info);
if (status < 0) {
MARS_ERR("cannot get info on '%s'\n", dent->d_path);
goto done;
}
src = path_make("%lld", info.current_size);
dst = path_make("%s/actsize-%s", dent->d_parent->d_path, my_id());
if (src && dst) {
(void)mars_symlink(src, dst, NULL, 0);
}
brick_string_free(src);
brick_string_free(dst);
2011-02-23 20:48:06 +00:00
}
2012-08-08 09:50:24 +00:00
2011-02-27 14:17:58 +00:00
done:
2011-02-23 20:48:06 +00:00
return status;
}
2011-06-01 12:55:06 +00:00
static int make_replay(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
2011-02-28 18:00:32 +00:00
struct mars_dent *parent = dent->d_parent;
2011-02-23 20:48:06 +00:00
int status = 0;
2011-03-22 14:36:26 +00:00
if (!global->global_power.button || !parent || !dent->new_link) {
2011-02-23 20:48:06 +00:00
MARS_DBG("nothing to do\n");
goto done;
}
2011-08-25 10:16:32 +00:00
status = make_log_finalize(global, dent);
2011-02-23 20:48:06 +00:00
if (status < 0) {
MARS_DBG("logger not initialized\n");
goto done;
}
2011-06-01 12:55:06 +00:00
done:
return status;
}
2011-06-17 11:32:38 +00:00
static
int make_dev(void *buf, struct mars_dent *dent)
2011-06-01 12:55:06 +00:00
{
struct mars_global *global = buf;
struct mars_dent *parent = dent->d_parent;
2011-06-17 11:32:38 +00:00
struct mars_rotate *rot = NULL;
2011-06-01 12:55:06 +00:00
struct mars_brick *dev_brick;
struct if_brick *_dev_brick;
2011-07-15 10:12:06 +00:00
bool switch_on;
2011-06-01 12:55:06 +00:00
int status = 0;
2011-06-17 11:32:38 +00:00
if (!parent || !dent->new_link) {
MARS_ERR("nothing to do\n");
return -EINVAL;
}
2011-03-22 14:36:26 +00:00
rot = parent->d_private;
2011-06-17 11:32:38 +00:00
if (!rot) {
MARS_DBG("nothing to do\n");
2011-02-23 20:48:06 +00:00
goto done;
}
2011-07-15 10:12:06 +00:00
if (!rot->trans_brick) {
MARS_DBG("transaction logger does not exist\n");
2011-02-23 20:48:06 +00:00
goto done;
}
2011-09-27 11:57:04 +00:00
if (!global->global_power.button &&
(!rot->if_brick || rot->if_brick->power.led_off)) {
MARS_DBG("nothing to do\n");
goto done;
}
if (rot->dev_size <= 0) {
MARS_WRN("trying to create device '%s' with zero size\n", dent->d_path);
goto done;
}
2011-02-23 20:48:06 +00:00
status = _parse_args(dent, dent->new_link, 1);
if (status < 0) {
2011-03-01 09:34:36 +00:00
MARS_DBG("fail\n");
2011-02-23 20:48:06 +00:00
goto done;
}
2011-03-01 18:00:14 +00:00
2011-07-15 10:12:06 +00:00
switch_on =
2011-07-20 13:11:44 +00:00
(rot->if_brick && atomic_read(&rot->if_brick->inputs[0]->open_count) > 0) ||
2011-09-27 08:46:14 +00:00
(rot->todo_primary &&
2012-08-03 08:42:51 +00:00
!rot->trans_brick->replay_mode &&
2011-07-20 13:11:44 +00:00
rot->trans_brick->power.led_on);
if (!global->global_power.button || global->exhausted) {
2011-09-27 11:57:04 +00:00
switch_on = false;
}
2011-07-15 10:12:06 +00:00
2011-03-01 09:34:36 +00:00
dev_brick =
make_brick_all(global,
dent,
2011-09-02 12:17:40 +00:00
false,
2011-03-11 13:57:54 +00:00
_set_if_params,
rot,
2011-03-01 09:34:36 +00:00
10 * HZ,
2011-03-01 18:00:14 +00:00
dent->d_argv[0],
2011-03-07 10:27:38 +00:00
(const struct generic_brick_type*)&if_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){(const struct generic_brick_type*)&trans_logger_brick_type},
NULL,
switch_on ? 1 : -1,
2011-09-27 11:57:04 +00:00
"%s/device-%s",
2011-03-03 09:02:10 +00:00
(const char *[]){"%s/logger"},
2011-03-01 18:00:14 +00:00
1,
2011-03-22 14:36:26 +00:00
parent->d_path,
2011-03-03 09:02:10 +00:00
dent->d_argv[0],
2011-03-22 14:36:26 +00:00
parent->d_path);
2011-06-17 11:32:38 +00:00
rot->if_brick = (void*)dev_brick;
2011-02-23 20:48:06 +00:00
if (!dev_brick) {
2011-06-17 11:32:38 +00:00
MARS_DBG("device not shown\n");
goto done;
2011-02-23 20:48:06 +00:00
}
2011-09-27 11:57:04 +00:00
dev_brick->show_status = _show_brick_status;
2011-03-08 16:45:52 +00:00
_dev_brick = (void*)dev_brick;
#if 0
if (_dev_brick->has_closed) {
_dev_brick->has_closed = false;
MARS_INF("rotating logfile for '%s'\n", parent->d_name);
status = mars_power_button((void*)rot->trans_brick, false);
rot->relevant_log = NULL;
}
#endif
2011-02-23 20:48:06 +00:00
done:
2011-06-17 11:32:38 +00:00
_show_primary(rot, parent);
2011-02-23 20:48:06 +00:00
return status;
}
2011-09-27 11:57:04 +00:00
static
int kill_dev(void *buf, struct mars_dent *dent)
{
struct mars_dent *parent = dent->d_parent;
int status = kill_any(buf, dent);
if (status > 0 && parent) {
struct mars_rotate *rot = parent->d_private;
if (rot) {
rot->if_brick = NULL;
}
}
return status;
}
2011-02-28 18:00:32 +00:00
static int _make_direct(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
struct mars_brick *brick;
2011-03-31 16:16:00 +00:00
char *src_path = NULL;
2011-02-23 20:48:06 +00:00
int status;
2011-03-31 16:16:00 +00:00
bool do_dealloc = false;
2011-02-23 20:48:06 +00:00
if (!global->global_power.button || !dent->d_parent || !dent->new_link) {
return 0;
}
status = _parse_args(dent, dent->new_link, 2);
if (status < 0) {
2011-03-11 13:57:54 +00:00
MARS_DBG("parse status = %d\n", status);
2011-02-23 20:48:06 +00:00
goto done;
}
2011-03-31 16:16:00 +00:00
src_path = dent->d_argv[0];
if (src_path[0] != '/') {
src_path = path_make("%s/%s", dent->d_parent->d_path, dent->d_argv[0]);
2011-08-12 11:09:48 +00:00
if (!src_path) {
MARS_DBG("fail\n");
status = -ENOMEM;
goto done;
}
2011-03-31 16:16:00 +00:00
do_dealloc = true;
}
2011-03-01 09:34:36 +00:00
brick =
make_brick_all(global,
dent,
2011-09-02 12:17:40 +00:00
false,
2011-03-18 13:15:40 +00:00
_set_bio_params,
2011-03-11 13:57:54 +00:00
NULL,
2011-03-01 09:34:36 +00:00
10 * HZ,
2011-03-31 16:16:00 +00:00
src_path,
2011-03-18 13:15:40 +00:00
(const struct generic_brick_type*)&bio_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){},
2011-03-23 08:09:00 +00:00
NULL,
0,
2011-03-31 16:16:00 +00:00
"%s",
2011-03-01 09:34:36 +00:00
(const char *[]){},
2011-03-02 09:30:56 +00:00
0,
2011-03-31 16:16:00 +00:00
src_path);
2011-03-01 09:34:36 +00:00
status = -1;
2011-02-23 20:48:06 +00:00
if (!brick) {
2011-03-01 09:34:36 +00:00
MARS_DBG("fail\n");
goto done;
2011-02-23 20:48:06 +00:00
}
2011-03-01 09:34:36 +00:00
brick =
make_brick_all(global,
dent,
2011-09-02 12:17:40 +00:00
false,
2011-03-11 13:57:54 +00:00
_set_if_params,
NULL,
2011-03-01 09:34:36 +00:00
10 * HZ,
dent->d_argv[1],
2011-03-07 10:27:38 +00:00
(const struct generic_brick_type*)&if_brick_type,
2011-03-01 09:34:36 +00:00
(const struct generic_brick_type*[]){NULL},
2011-03-23 08:09:00 +00:00
NULL,
0,
2011-09-27 11:57:04 +00:00
"%s/directdevice-%s",
2011-03-31 16:16:00 +00:00
(const char *[]){ "%s" },
2011-03-02 09:30:56 +00:00
1,
2011-03-03 09:02:10 +00:00
dent->d_parent->d_path,
dent->d_argv[1],
2011-03-31 16:16:00 +00:00
src_path);
2011-03-01 09:34:36 +00:00
status = -1;
2011-02-23 20:48:06 +00:00
if (!brick) {
2011-03-01 09:34:36 +00:00
MARS_DBG("fail\n");
goto done;
2011-02-23 20:48:06 +00:00
}
2011-03-01 09:34:36 +00:00
2011-02-23 20:48:06 +00:00
status = 0;
done:
MARS_DBG("status = %d\n", status);
2011-08-12 11:09:48 +00:00
if (do_dealloc && src_path)
brick_string_free(src_path);
2011-02-23 20:48:06 +00:00
return status;
}
2011-02-28 18:00:32 +00:00
static int _make_copy(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
2011-02-25 11:46:38 +00:00
struct mars_global *global = buf;
2011-03-23 08:09:00 +00:00
const char *switch_path = NULL;
2011-03-22 14:36:26 +00:00
const char *copy_path = NULL;
2011-02-25 11:46:38 +00:00
int status;
2011-02-23 20:48:06 +00:00
2011-02-25 11:46:38 +00:00
if (!global->global_power.button || !dent->d_parent || !dent->new_link) {
return 0;
2011-02-23 20:48:06 +00:00
}
2011-02-25 11:46:38 +00:00
status = _parse_args(dent, dent->new_link, 2);
if (status < 0) {
goto done;
2011-02-23 20:48:06 +00:00
}
2011-03-22 14:36:26 +00:00
copy_path = backskip_replace(dent->d_path, '/', true, "/copy-");
if (unlikely(!copy_path)) {
status = -ENOMEM;
goto done;
}
2011-03-23 17:58:02 +00:00
// check whether connection is allowed
2011-09-21 11:30:11 +00:00
switch_path = path_make("%s/todo-%s/connect", dent->d_parent->d_path, my_id());
2011-02-23 20:48:06 +00:00
status = __make_copy(global, dent, switch_path, copy_path, dent->d_parent->d_path, (const char**)dent->d_argv, -1, false, true, NULL);
2011-02-23 20:48:06 +00:00
done:
MARS_DBG("status = %d\n", status);
2011-03-22 14:36:26 +00:00
if (copy_path)
2011-08-12 11:09:48 +00:00
brick_string_free(copy_path);
2011-03-23 17:58:02 +00:00
if (switch_path)
2011-08-12 11:09:48 +00:00
brick_string_free(switch_path);
2011-02-23 20:48:06 +00:00
return status;
}
2011-02-28 18:00:32 +00:00
static int make_sync(void *buf, struct mars_dent *dent)
2011-02-23 20:48:06 +00:00
{
struct mars_global *global = buf;
2011-07-08 07:02:14 +00:00
struct mars_rotate *rot;
2011-02-25 11:46:38 +00:00
loff_t start_pos = 0;
2011-07-08 07:02:14 +00:00
loff_t end_pos = 0;
struct mars_dent *size_dent;
2011-02-28 18:00:32 +00:00
struct mars_dent *connect_dent;
2011-02-25 11:46:38 +00:00
char *peer;
struct copy_brick *copy = NULL;
2011-03-02 09:30:56 +00:00
char *tmp = NULL;
2011-03-23 08:09:00 +00:00
const char *switch_path = NULL;
2011-03-22 14:36:26 +00:00
const char *copy_path = NULL;
const char *src = NULL;
const char *dst = NULL;
bool do_start = true;
2011-02-23 20:48:06 +00:00
int status;
2011-03-23 08:09:00 +00:00
if (!global->global_power.button || !dent->d_parent || !dent->new_link) {
2011-02-23 20:48:06 +00:00
return 0;
}
2011-02-25 11:46:38 +00:00
/* Analyze replay position
*/
status = sscanf(dent->new_link, "%lld", &start_pos);
if (status != 1) {
MARS_ERR("bad syncstatus symlink syntax '%s' (%s)\n", dent->new_link, dent->d_path);
status = -EINVAL;
goto done;
}
2011-07-08 07:02:14 +00:00
rot = dent->d_parent->d_private;
if (rot) {
rot->allow_update = true;
rot->syncstatus_dent = dent;
2011-07-08 07:02:14 +00:00
}
/* Sync necessary?
*/
tmp = path_make("%s/size", dent->d_parent->d_path);
status = -ENOMEM;
if (unlikely(!tmp))
goto done;
size_dent = (void*)mars_find_dent(global, tmp);
if (!size_dent || !size_dent->new_link) {
2011-07-28 11:41:06 +00:00
MARS_ERR("cannot determine size '%s'\n", tmp);
2011-07-08 07:02:14 +00:00
status = -ENOENT;
goto done;
}
status = sscanf(size_dent->new_link, "%lld", &end_pos);
if (status != 1) {
MARS_ERR("bad size symlink syntax '%s' (%s)\n", size_dent->new_link, tmp);
status = -EINVAL;
goto done;
}
if (start_pos == end_pos) {
MARS_DBG("no data sync necessary, size = %lld\n", start_pos);
do_start = false;
2011-07-08 07:02:14 +00:00
}
2011-08-12 11:09:48 +00:00
brick_string_free(tmp);
2011-07-08 07:02:14 +00:00
2011-02-25 11:46:38 +00:00
/* Determine peer
*/
2011-03-03 09:02:10 +00:00
tmp = path_make("%s/connect-%s", dent->d_parent->d_path, my_id());
2011-03-02 09:30:56 +00:00
status = -ENOMEM;
if (unlikely(!tmp))
goto done;
connect_dent = (void*)mars_find_dent(global, tmp);
2011-02-25 11:46:38 +00:00
if (!connect_dent || !connect_dent->new_link) {
MARS_WRN("cannot determine peer, symlink '%s' is missing => assuming that I am STANDALONE\n", tmp);
rot->allow_replay = true;
status = 0;
2011-02-23 20:48:06 +00:00
goto done;
}
2011-02-25 11:46:38 +00:00
peer = connect_dent->new_link;
/* Disallow contemporary sync & logfile_apply
*/
2012-08-02 12:59:55 +00:00
if (do_start && !rot->allow_sync) {
MARS_WRN("cannot start sync because logfile application is running!\n");
do_start = false;
}
#if defined(CONFIG_MARS_SYNC_FLIP_INTERVAL) && CONFIG_MARS_SYNC_FLIP_INTERVAL > 0
/* Flip between replay and sync
*/
if (do_start && rot->replay_mode && rot->end_pos > rot->start_pos) {
if (!rot->flip_start) {
rot->flip_start = jiffies;
} else if ((long long)jiffies - rot->flip_start > CONFIG_MARS_SYNC_FLIP_INTERVAL * HZ) {
do_start = false;
rot->flip_start = jiffies + CONFIG_MARS_SYNC_FLIP_INTERVAL * HZ;
}
} else {
rot->flip_start = 0;
}
#endif
2011-02-25 11:46:38 +00:00
/* Start copy
*/
2011-03-03 09:02:10 +00:00
src = path_make("data-%s@%s", peer, peer);
dst = path_make("data-%s", my_id());
2011-03-22 14:36:26 +00:00
copy_path = backskip_replace(dent->d_path, '/', true, "/copy-");
2011-03-01 18:00:14 +00:00
2011-03-23 17:58:02 +00:00
// check whether connection is allowed
2011-09-21 11:30:11 +00:00
switch_path = path_make("%s/todo-%s/sync", dent->d_parent->d_path, my_id());
2011-03-23 17:58:02 +00:00
2011-06-17 11:32:38 +00:00
status = -ENOMEM;
if (unlikely(!src || !dst || !copy_path || !switch_path))
goto done;
MARS_DBG("initial sync '%s' => '%s' rot->allow_sync = %d do_start = %d\n", src, dst, rot->allow_sync, do_start);
2011-02-25 11:46:38 +00:00
2011-03-01 18:00:14 +00:00
{
const char *argv[2] = { src, dst };
2012-08-02 08:47:42 +00:00
#ifdef CONFIG_MARS_FAST_FULLSYNC
# define VERIFY_MODE true
#else
# define VERIFY_MODE false
#endif
status = __make_copy(global, dent, do_start ? switch_path : "", copy_path, dent->d_parent->d_path, argv, start_pos, VERIFY_MODE, true, &copy);
if (copy)
copy->copy_limiter = &rot->sync_limiter;
rot->sync_brick = copy;
rot->allow_replay = (!copy || copy->power.led_off);
2011-03-01 18:00:14 +00:00
}
2011-02-23 20:48:06 +00:00
2011-02-25 11:46:38 +00:00
/* Update syncstatus symlink
*/
2011-06-17 11:32:38 +00:00
if (status >= 0 && copy &&
((copy->power.button && copy->power.led_on) ||
(copy->copy_last == copy->copy_end && copy->copy_end > 0))) {
2011-08-12 11:09:48 +00:00
brick_string_free(src);
brick_string_free(dst);
2011-03-03 09:02:10 +00:00
src = path_make("%lld", copy->copy_last);
dst = path_make("%s/syncstatus-%s", dent->d_parent->d_path, my_id());
2011-03-02 09:30:56 +00:00
status = -ENOMEM;
if (unlikely(!src || !dst))
goto done;
2011-02-27 14:17:58 +00:00
status = mars_symlink(src, dst, NULL, 0);
brick_string_free(src);
brick_string_free(dst);
src = path_make("%lld,%lld", copy->verify_ok_count, copy->verify_error_count);
dst = path_make("%s/verifystatus-%s", dent->d_parent->d_path, my_id());
status = -ENOMEM;
if (unlikely(!src || !dst))
goto done;
status = mars_symlink(src, dst, NULL, 0);
2011-02-25 11:46:38 +00:00
}
2011-02-23 20:48:06 +00:00
done:
MARS_DBG("status = %d\n", status);
brick_string_free(tmp);
brick_string_free(src);
brick_string_free(dst);
brick_string_free(copy_path);
brick_string_free(switch_path);
2011-02-23 20:48:06 +00:00
return status;
}
static int prepare_delete(void *buf, struct mars_dent *dent)
{
struct mars_global *global = buf;
struct mars_dent *target;
struct mars_dent *response;
const char *response_path = NULL;
int max_serial = 0;
if (!global || !dent || !dent->new_link || !dent->d_path) {
goto done;
}
target = _mars_find_dent(global, dent->new_link);
if (target) {
mars_unlink(dent->new_link);
target->d_killme = true;
MARS_DBG("target '%s' deleted and marked for removal\n", dent->new_link);
} else {
MARS_DBG("target '%s' does no longer exist\n", dent->new_link);
if (dent->d_serial < global->deleted_border) {
MARS_DBG("removing deletion symlink '%s'\n", dent->d_path);
dent->d_killme = true;
mars_unlink(dent->d_path);
}
}
response_path = path_make("/mars/todo-global/deleted-%s", my_id());
if (!response_path) {
MARS_ERR("cannot build response path for '%s'\n", dent->new_link);
goto done;
}
response = _mars_find_dent(global, response_path);
if (response && response->new_link) {
sscanf(response->new_link, "%d", &max_serial);
}
if (dent->d_serial > max_serial) {
char response_val[16];
max_serial = dent->d_serial;
snprintf(response_val, sizeof(response_val), "%09d", max_serial);
mars_symlink(response_val, response_path, NULL, 0);
}
done:
brick_string_free(response_path);
return 0;
}
static int check_deleted(void *buf, struct mars_dent *dent)
{
struct mars_global *global = buf;
int serial = 0;
int status;
if (!global || !dent || !dent->new_link) {
goto done;
}
status = sscanf(dent->new_link, "%d", &serial);
if (status != 1 || serial <= 0) {
MARS_WRN("cannot parse symlink '%s' -> '%s'\n", dent->d_path, dent->new_link);
goto done;
}
/* Compute the minimum of the deletion progress among
* the resource members.
*/
if (serial < global->deleted_min || !global->deleted_min)
global->deleted_min = serial;
done:
return 0;
}
2011-02-23 20:48:06 +00:00
///////////////////////////////////////////////////////////////////////
// the order is important!
enum {
2011-02-25 11:46:38 +00:00
// root element: this must have index 0
CL_ROOT,
// global userspace
CL_GLOBAL_USERSPACE,
CL_GLOBAL_USERSPACE_ITEMS,
// global todos
CL_GLOBAL_TODO,
CL_GLOBAL_TODO_DELETE,
CL_GLOBAL_TODO_DELETED,
2011-02-25 11:46:38 +00:00
// replacement for DNS in kernelspace
2011-02-23 20:48:06 +00:00
CL_IPS,
CL_PEERS,
CL_ALIVE,
CL_EXHAUSTED,
CL_JAMMED,
CL_REST_SPACE,
2011-02-25 11:46:38 +00:00
// resource definitions
2011-02-23 20:48:06 +00:00
CL_RESOURCE,
CL_RESOURCE_USERSPACE,
CL_RESOURCE_USERSPACE_ITEMS,
2011-06-10 13:57:52 +00:00
CL_DEFAULTS0,
CL_DEFAULTS,
CL_DEFAULTS_ITEMS0,
CL_DEFAULTS_ITEMS,
2011-09-21 11:30:11 +00:00
CL_TODO,
CL_TODO_ITEMS,
2011-06-10 13:57:52 +00:00
CL_ACTUAL,
CL_ACTUAL_ITEMS,
2011-02-25 11:46:38 +00:00
CL_CONNECT,
2011-02-23 20:48:06 +00:00
CL_DATA,
2011-07-08 07:02:14 +00:00
CL_SIZE,
2012-08-08 09:50:24 +00:00
CL_ACTSIZE,
2011-02-23 20:48:06 +00:00
CL_PRIMARY,
CL__FILE,
CL_SYNC,
CL_VERIF,
2011-02-23 20:48:06 +00:00
CL__COPY,
CL__DIRECT,
2011-05-13 11:19:28 +00:00
CL_VERSION,
2011-02-23 20:48:06 +00:00
CL_LOG,
2011-06-01 12:55:06 +00:00
CL_REPLAYSTATUS,
2011-02-23 20:48:06 +00:00
CL_DEVICE,
};
/* Please keep the order the same as in the enum.
*/
static const struct light_class light_classes[] = {
/* Placeholder for root node /mars/
*/
[CL_ROOT] = {
},
/* Subdirectory for global userspace items...
*/
[CL_GLOBAL_USERSPACE] = {
.cl_name = "userspace",
.cl_len = 9,
.cl_type = 'd',
.cl_hostcontext = false,
.cl_father = CL_ROOT,
},
[CL_GLOBAL_USERSPACE_ITEMS] = {
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
.cl_father = CL_GLOBAL_USERSPACE,
},
/* Subdirectory for global controlling items...
*/
[CL_GLOBAL_TODO] = {
.cl_name = "todo-global",
.cl_len = 11,
.cl_type = 'd',
.cl_hostcontext = false,
.cl_father = CL_ROOT,
},
/* ... and its contents
*/
[CL_GLOBAL_TODO_DELETE] = {
.cl_name = "delete-",
.cl_len = 7,
.cl_type = 'l',
.cl_serial = true,
.cl_father = CL_GLOBAL_TODO,
.cl_prepare = prepare_delete,
},
[CL_GLOBAL_TODO_DELETED] = {
.cl_name = "deleted-",
.cl_len = 8,
.cl_type = 'l',
.cl_father = CL_GLOBAL_TODO,
.cl_prepare = check_deleted,
},
2011-02-23 20:48:06 +00:00
/* Directory containing the addresses of all peers
*/
[CL_IPS] = {
.cl_name = "ips",
.cl_len = 3,
.cl_type = 'd',
.cl_father = CL_ROOT,
},
/* Anyone participating in a MARS cluster must
* be named here (symlink pointing to the IP address).
* We have no DNS in kernel space.
*/
[CL_PEERS] = {
.cl_name = "ip-",
.cl_len = 3,
.cl_type = 'l',
.cl_father = CL_IPS,
2011-08-31 11:42:04 +00:00
#ifdef RUN_PEERS
2011-02-24 16:37:32 +00:00
.cl_forward = make_scan,
#endif
2011-08-31 11:42:04 +00:00
.cl_backward = kill_scan,
2011-02-23 20:48:06 +00:00
},
/* Indicate aliveness of all cluster paritcipants
* by the timestamp of this link.
*/
[CL_ALIVE] = {
.cl_name = "alive-",
.cl_len = 6,
.cl_type = 'l',
.cl_father = CL_ROOT,
},
/* Indicate whether filesystem is full
*/
[CL_EXHAUSTED] = {
.cl_name = "exhausted-",
.cl_len = 10,
.cl_type = 'l',
.cl_father = CL_ROOT,
},
/* dto as percentage
*/
2012-01-24 09:00:21 +00:00
[CL_REST_SPACE] = {
.cl_name = "rest-space-",
2012-01-24 08:30:50 +00:00
.cl_len = 11,
.cl_type = 'l',
.cl_father = CL_ROOT,
},
[CL_JAMMED] = {
.cl_name = "jammed-",
.cl_len = 7,
.cl_type = 'l',
.cl_father = CL_ROOT,
},
2011-02-23 20:48:06 +00:00
/* Directory containing all items of a resource
*/
[CL_RESOURCE] = {
.cl_name = "resource-",
.cl_len = 9,
.cl_type = 'd',
.cl_use_channel = true,
2011-02-23 20:48:06 +00:00
.cl_father = CL_ROOT,
},
2011-06-10 13:57:52 +00:00
/* Subdirectory for resource-specific userspace items...
*/
[CL_RESOURCE_USERSPACE] = {
.cl_name = "userspace",
.cl_len = 9,
.cl_type = 'd',
.cl_hostcontext = false,
.cl_father = CL_RESOURCE,
},
[CL_RESOURCE_USERSPACE_ITEMS] = {
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
.cl_father = CL_RESOURCE_USERSPACE,
},
2011-06-10 13:57:52 +00:00
/* Subdirectory for defaults...
*/
[CL_DEFAULTS0] = {
.cl_name = "defaults",
.cl_len = 8,
.cl_type = 'd',
.cl_hostcontext = false,
.cl_father = CL_RESOURCE,
},
[CL_DEFAULTS] = {
.cl_name = "defaults-",
.cl_len = 9,
.cl_type = 'd',
2011-07-15 10:12:06 +00:00
.cl_hostcontext = false,
2011-06-10 13:57:52 +00:00
.cl_father = CL_RESOURCE,
},
/* ... and its contents
*/
[CL_DEFAULTS_ITEMS0] = {
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
.cl_father = CL_DEFAULTS0,
},
[CL_DEFAULTS_ITEMS] = {
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
.cl_father = CL_DEFAULTS,
},
2011-03-23 17:58:02 +00:00
/* Subdirectory for controlling items...
*/
2011-09-21 11:30:11 +00:00
[CL_TODO] = {
.cl_name = "todo-",
.cl_len = 5,
2011-03-23 17:58:02 +00:00
.cl_type = 'd',
2011-07-15 10:12:06 +00:00
.cl_hostcontext = false,
2011-03-23 17:58:02 +00:00
.cl_father = CL_RESOURCE,
},
/* ... and its contents
*/
2011-09-21 11:30:11 +00:00
[CL_TODO_ITEMS] = {
2011-03-23 17:58:02 +00:00
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
2011-09-21 11:30:11 +00:00
.cl_father = CL_TODO,
2011-03-23 17:58:02 +00:00
},
2011-06-10 13:57:52 +00:00
/* Subdirectory for actual state
*/
[CL_ACTUAL] = {
.cl_name = "actual-",
.cl_len = 7,
.cl_type = 'd',
2011-07-15 10:12:06 +00:00
.cl_hostcontext = false,
2011-06-10 13:57:52 +00:00
.cl_father = CL_RESOURCE,
},
/* ... and its contents
*/
[CL_ACTUAL_ITEMS] = {
.cl_name = "",
.cl_len = 0, // catch any
.cl_type = 'l',
.cl_father = CL_ACTUAL,
},
2011-02-25 11:46:38 +00:00
/* Symlink indicating the current peer
*/
[CL_CONNECT] = {
.cl_name = "connect-",
.cl_len = 8,
.cl_type = 'l',
2011-06-17 11:32:38 +00:00
.cl_hostcontext = false, // not used here
2011-02-25 11:46:38 +00:00
.cl_father = CL_RESOURCE,
},
2011-02-23 20:48:06 +00:00
/* File or symlink to the real device / real (sparse) file
* when hostcontext is missing, the corresponding peer will
* not participate in that resource.
*/
[CL_DATA] = {
.cl_name = "data-",
.cl_len = 5,
.cl_type = 'F',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_DATA
2011-03-18 13:15:40 +00:00
.cl_forward = make_bio,
2011-08-31 11:42:04 +00:00
#endif
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
2011-07-08 07:02:14 +00:00
/* Symlink indicating the (common) size of the resource
*/
[CL_SIZE] = {
.cl_name = "size",
.cl_len = 4,
.cl_type = 'l',
.cl_hostcontext = false,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_LOGINIT
2011-07-08 07:02:14 +00:00
.cl_forward = make_log_init,
2011-08-31 11:42:04 +00:00
#endif
.cl_backward = kill_any,
2011-07-08 07:02:14 +00:00
},
2012-08-08 09:50:24 +00:00
/* Dito for each individual size
*/
[CL_ACTSIZE] = {
.cl_name = "actsize-",
.cl_len = 8,
.cl_type = 'l',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
},
2011-02-23 20:48:06 +00:00
/* Symlink pointing to the name of the primary node
*/
[CL_PRIMARY] = {
.cl_name = "primary",
.cl_len = 7,
.cl_type = 'l',
.cl_hostcontext = false,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_PRIMARY
2011-02-23 20:48:06 +00:00
.cl_forward = make_primary,
2011-08-31 11:42:04 +00:00
#endif
2011-02-23 20:48:06 +00:00
.cl_backward = NULL,
},
/* Only for testing: open local file
*/
[CL__FILE] = {
.cl_name = "_file-",
.cl_len = 6,
.cl_type = 'F',
.cl_serial = true,
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
2011-03-18 13:15:40 +00:00
.cl_forward = make_bio,
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
/* symlink indicating the current status / end
* of initial data sync.
*/
[CL_SYNC] = {
.cl_name = "syncstatus-",
.cl_len = 11,
.cl_type = 'l',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_SYNCSTATUS
2011-02-23 20:48:06 +00:00
.cl_forward = make_sync,
2011-08-31 11:42:04 +00:00
#endif
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
/* informational symlink for verify status
* of initial data sync.
*/
[CL_VERIF] = {
.cl_name = "verifystatus-",
.cl_len = 13,
.cl_type = 'l',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
},
2011-02-23 20:48:06 +00:00
/* Only for testing: make a copy instance
*/
[CL__COPY] = {
.cl_name = "_copy-",
.cl_len = 6,
.cl_type = 'l',
.cl_serial = true,
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
.cl_forward = _make_copy,
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
/* Only for testing: access local data
*/
[CL__DIRECT] = {
.cl_name = "_direct-",
.cl_len = 8,
.cl_type = 'l',
.cl_serial = true,
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
.cl_forward = _make_direct,
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
2011-05-13 11:19:28 +00:00
/* Passive symlink indicating the split-brain crypto hash
*/
[CL_VERSION] = {
.cl_name = "version-",
.cl_len = 8,
.cl_type = 'l',
.cl_serial = true,
2011-07-15 10:12:06 +00:00
.cl_hostcontext = false,
2011-05-13 11:19:28 +00:00
.cl_father = CL_RESOURCE,
},
2011-02-23 20:48:06 +00:00
/* Logfiles for transaction logger
*/
[CL_LOG] = {
.cl_name = "log-",
.cl_len = 4,
.cl_type = 'F',
.cl_serial = true,
2011-10-05 14:59:29 +00:00
.cl_hostcontext = false,
2011-02-23 20:48:06 +00:00
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_LOGFILES
2011-08-25 10:16:32 +00:00
.cl_forward = make_log_step,
2011-02-23 20:48:06 +00:00
#endif
2011-08-31 11:42:04 +00:00
.cl_backward = kill_any,
2011-02-23 20:48:06 +00:00
},
2011-06-01 12:55:06 +00:00
/* Symlink indicating the last state of
* transaction log replay.
*/
[CL_REPLAYSTATUS] = {
.cl_name = "replay-",
.cl_len = 7,
.cl_type = 'l',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_REPLAY
2011-06-01 12:55:06 +00:00
.cl_forward = make_replay,
2011-08-31 11:42:04 +00:00
#endif
2011-08-25 10:16:32 +00:00
.cl_backward = kill_any,
2011-06-01 12:55:06 +00:00
},
2011-02-23 20:48:06 +00:00
/* Name of the device appearing at the primary
*/
[CL_DEVICE] = {
.cl_name = "device-",
.cl_len = 7,
.cl_type = 'l',
.cl_hostcontext = true,
.cl_father = CL_RESOURCE,
2011-08-31 11:42:04 +00:00
#ifdef RUN_DEVICE
2011-02-23 20:48:06 +00:00
.cl_forward = make_dev,
2011-08-31 11:42:04 +00:00
#endif
2011-09-27 11:57:04 +00:00
.cl_backward = kill_dev,
2011-02-23 20:48:06 +00:00
},
{}
};
/* Helper routine to pre-determine the relevance of a name from the filesystem.
*/
2012-12-18 05:31:24 +00:00
int light_checker(struct mars_dent *parent, const char *_name, int namlen, unsigned int d_type, int *prefix, int *serial, bool *use_channel)
2011-02-23 20:48:06 +00:00
{
int class;
2011-03-01 18:00:14 +00:00
int status = -2;
#ifdef MARS_DEBUGGING
const char *name = brick_strndup(_name, namlen);
2011-03-01 18:00:14 +00:00
if (!name)
return -ENOMEM;
#else
const char *name = _name;
#endif
2011-02-23 20:48:06 +00:00
//MARS_DBG("trying '%s' '%s'\n", path, name);
for (class = CL_ROOT + 1; ; class++) {
const struct light_class *test = &light_classes[class];
int len = test->cl_len;
2011-03-23 17:58:02 +00:00
if (!test->cl_name) { // end of table
2011-02-23 20:48:06 +00:00
break;
2011-03-23 17:58:02 +00:00
}
2011-02-23 20:48:06 +00:00
//MARS_DBG(" testing class '%s'\n", test->cl_name);
2011-03-23 17:58:02 +00:00
#ifdef MARS_DEBUGGING
2011-02-23 20:48:06 +00:00
if (len != strlen(test->cl_name)) {
2011-03-23 17:58:02 +00:00
MARS_ERR("internal table '%s' mismatch: %d != %d\n", test->cl_name, len, (int)strlen(test->cl_name));
2011-02-23 20:48:06 +00:00
len = strlen(test->cl_name);
}
#endif
2011-03-23 17:58:02 +00:00
if (test->cl_father &&
(!parent || parent->d_class != test->cl_father)) {
continue;
}
if (len > 0 &&
(namlen < len || memcmp(name, test->cl_name, len))) {
continue;
}
//MARS_DBG("path '%s/%s' matches class %d '%s'\n", path, name, class, test->cl_name);
// check special contexts
if (test->cl_serial) {
int plus = 0;
int count;
count = sscanf(name+len, "%d%n", serial, &plus);
if (count < 1) {
//MARS_DBG("'%s' serial number mismatch at '%s'\n", name, name+len);
continue;
2011-02-23 20:48:06 +00:00
}
2011-06-01 12:55:06 +00:00
//MARS_DBG("'%s' serial number = %d\n", name, *serial);
2011-03-23 17:58:02 +00:00
len += plus;
if (name[len] == '-')
len++;
}
if (prefix)
2011-02-23 20:48:06 +00:00
*prefix = len;
2011-03-23 17:58:02 +00:00
if (test->cl_hostcontext) {
if (memcmp(name+len, my_id(), namlen-len)) {
//MARS_DBG("context mismatch '%s' at '%s'\n", name, name+len);
continue;
2011-02-23 20:48:06 +00:00
}
}
2011-03-23 17:58:02 +00:00
// all ok
status = class;
*use_channel = test->cl_use_channel;
2011-02-23 20:48:06 +00:00
}
2011-03-23 17:58:02 +00:00
2011-03-01 18:00:14 +00:00
#ifdef MARS_DEBUGGING
2011-08-12 11:09:48 +00:00
brick_string_free(name);
2011-03-01 18:00:14 +00:00
#endif
return status;
2011-02-23 20:48:06 +00:00
}
/* Do some syntactic checks, then delegate work to the real worker functions
* from the light_classes[] table.
*/
static int light_worker(struct mars_global *global, struct mars_dent *dent, bool prepare, bool direction)
2011-02-23 20:48:06 +00:00
{
light_worker_fn worker;
int class = dent->d_class;
if (class < 0 || class >= sizeof(light_classes)/sizeof(struct light_class)) {
MARS_ERR_ONCE(dent, "bad internal class %d of '%s'\n", class, dent->d_path);
return -EINVAL;
}
switch (light_classes[class].cl_type) {
case 'd':
if (!S_ISDIR(dent->new_stat.mode)) {
MARS_ERR_ONCE(dent, "'%s' should be a directory, but is something else\n", dent->d_path);
return -EINVAL;
}
break;
case 'f':
if (!S_ISREG(dent->new_stat.mode)) {
MARS_ERR_ONCE(dent, "'%s' should be a regular file, but is something else\n", dent->d_path);
return -EINVAL;
}
break;
case 'F':
if (!S_ISREG(dent->new_stat.mode) && !S_ISLNK(dent->new_stat.mode)) {
MARS_ERR_ONCE(dent, "'%s' should be a regular file or a symlink, but is something else\n", dent->d_path);
return -EINVAL;
}
break;
case 'l':
if (!S_ISLNK(dent->new_stat.mode)) {
MARS_ERR_ONCE(dent, "'%s' should be a symlink, but is something else\n", dent->d_path);
return -EINVAL;
}
break;
}
if (likely(class > CL_ROOT)) {
int father = light_classes[class].cl_father;
if (father == CL_ROOT) {
if (unlikely(dent->d_parent)) {
MARS_ERR_ONCE(dent, "'%s' class %d is not at the root of the hierarchy\n", dent->d_path, class);
2011-02-23 20:48:06 +00:00
return -EINVAL;
}
} else if (unlikely(!dent->d_parent || dent->d_parent->d_class != father)) {
MARS_ERR_ONCE(dent, "last component '%s' from '%s' is at the wrong position in the hierarchy (class = %d, parent_class = %d, parent = '%s')\n", dent->d_name, dent->d_path, father, dent->d_parent ? dent->d_parent->d_class : -9999, dent->d_parent ? dent->d_parent->d_path : "");
return -EINVAL;
}
}
if (prepare) {
worker = light_classes[class].cl_prepare;
} else if (direction) {
2011-02-23 20:48:06 +00:00
worker = light_classes[class].cl_backward;
} else {
worker = light_classes[class].cl_forward;
}
if (worker) {
int status;
2012-01-17 10:48:56 +00:00
if (!direction)
MARS_DBG("--- start working %s on '%s' rest='%s'\n", direction ? "backward" : "forward", dent->d_path, dent->d_rest);
2011-02-23 20:48:06 +00:00
status = worker(global, (void*)dent);
2012-01-17 10:48:56 +00:00
MARS_DBG("--- done, worked %s on '%s', status = %d\n", direction ? "backward" : "forward", dent->d_path, status);
2011-02-23 20:48:06 +00:00
return status;
}
return 0;
}
2011-03-03 09:02:10 +00:00
#ifdef STAT_DEBUGGING
2011-08-31 11:42:04 +00:00
static
void _show_one(struct mars_brick *test, int *brick_count)
{
int i;
if (*brick_count) {
MARS_STAT("---------\n");
}
MARS_STAT("BRICK type = %s path = '%s' name = '%s' "
"size_hint=%d "
"mrefs_alloc = %d "
"mrefs_apsect_alloc = %d "
"total_mrefs_alloc = %d "
"total_mrefs_aspects = %d "
"button = %d off = %d on = %d\n",
SAFE_STR(test->type->type_name),
SAFE_STR(test->brick_path),
SAFE_STR(test->brick_name),
test->mref_object_layout.size_hint,
atomic_read(&test->mref_object_layout.alloc_count),
atomic_read(&test->mref_object_layout.aspect_count),
atomic_read(&test->mref_object_layout.total_alloc_count),
atomic_read(&test->mref_object_layout.total_aspect_count),
test->power.button,
test->power.led_off,
test->power.led_on);
2011-08-31 11:42:04 +00:00
(*brick_count)++;
if (test->ops && test->ops->brick_statistics) {
char *info = test->ops->brick_statistics(test, 0);
if (info) {
MARS_STAT(" %s", info);
brick_string_free(info);
}
}
2011-10-05 14:59:29 +00:00
for (i = 0; i < test->type->max_inputs; i++) {
2011-08-31 11:42:04 +00:00
struct mars_input *input = test->inputs[i];
struct mars_output *output = input ? input->connect : NULL;
if (output) {
MARS_STAT(" input %d connected with %s path = '%s' name = '%s'\n", i, SAFE_STR(output->brick->type->type_name), SAFE_STR(output->brick->brick_path), SAFE_STR(output->brick->brick_name));
} else {
MARS_STAT(" input %d not connected\n", i);
}
}
2011-10-05 14:59:29 +00:00
for (i = 0; i < test->type->max_outputs; i++) {
2011-08-31 11:42:04 +00:00
struct mars_output *output = test->outputs[i];
2011-10-05 14:59:29 +00:00
if (output) {
MARS_STAT(" output %d nr_connected = %d\n", i, output->nr_connected);
}
2011-08-31 11:42:04 +00:00
}
}
2011-03-03 09:02:10 +00:00
static
void _show_statist(struct mars_global *global)
{
struct list_head *tmp;
int dent_count = 0;
int brick_count = 0;
2011-08-12 11:09:48 +00:00
brick_mem_statistics();
2011-06-10 13:57:52 +00:00
down_read(&global->brick_mutex);
2011-08-31 11:42:04 +00:00
MARS_STAT("================================== ordinary bricks:\n");
2011-03-03 09:02:10 +00:00
for (tmp = global->brick_anchor.next; tmp != &global->brick_anchor; tmp = tmp->next) {
struct mars_brick *test;
test = container_of(tmp, struct mars_brick, global_brick_link);
2011-08-31 11:42:04 +00:00
_show_one(test, &brick_count);
}
MARS_STAT("================================== server bricks:\n");
for (tmp = global->server_anchor.next; tmp != &global->server_anchor; tmp = tmp->next) {
struct mars_brick *test;
test = container_of(tmp, struct mars_brick, global_brick_link);
_show_one(test, &brick_count);
2011-03-03 09:02:10 +00:00
}
2011-06-10 13:57:52 +00:00
up_read(&global->brick_mutex);
2011-03-03 09:02:10 +00:00
2011-08-25 10:16:32 +00:00
MARS_STAT("================================== dents:\n");
down_read(&global->dent_mutex);
for (tmp = global->dent_anchor.next; tmp != &global->dent_anchor; tmp = tmp->next) {
struct mars_dent *dent;
struct list_head *sub;
dent = container_of(tmp, struct mars_dent, dent_link);
2011-08-31 11:42:04 +00:00
MARS_STAT("dent %d '%s' '%s' stamp=%ld.%09ld\n", dent->d_class, SAFE_STR(dent->d_path), SAFE_STR(dent->new_link), dent->new_stat.mtime.tv_sec, dent->new_stat.mtime.tv_nsec);
2011-08-25 10:16:32 +00:00
dent_count++;
for (sub = dent->brick_list.next; sub != &dent->brick_list; sub = sub->next) {
struct mars_brick *test;
test = container_of(sub, struct mars_brick, dent_brick_link);
2011-08-31 11:42:04 +00:00
MARS_STAT(" owner of brick '%s'\n", SAFE_STR(test->brick_path));
2011-08-25 10:16:32 +00:00
}
}
up_read(&global->dent_mutex);
MARS_INF("==================== STATISTICS: %d dents, %d bricks, %lld KB free\n", dent_count, brick_count, global->remaining_space);
2011-03-03 09:02:10 +00:00
}
#endif
static
void _make_alivelink(const char *name, loff_t val)
{
char *src = path_make("%lld", val);
char *dst = path_make("/mars/%s-%s", name, my_id());
2012-01-24 09:28:47 +00:00
if (!src || !dst) {
MARS_ERR("cannot make symlink paths\n");
goto err;
}
2012-01-31 15:19:35 +00:00
MARS_DBG("'%s' -> '%s'\n", src, dst);
mars_symlink(src, dst, NULL, 0);
2012-01-24 09:28:47 +00:00
err:
brick_string_free(dst);
brick_string_free(src);
}
2011-08-31 11:42:04 +00:00
static struct mars_global _global = {
.dent_anchor = LIST_HEAD_INIT(_global.dent_anchor),
.brick_anchor = LIST_HEAD_INIT(_global.brick_anchor),
.server_anchor = LIST_HEAD_INIT(_global.server_anchor),
.global_power = {
.button = true,
},
.dent_mutex = __RWSEM_INITIALIZER(_global.dent_mutex),
.brick_mutex = __RWSEM_INITIALIZER(_global.brick_mutex),
.main_event = __WAIT_QUEUE_HEAD_INITIALIZER(_global.main_event),
};
2011-02-23 20:48:06 +00:00
static int light_thread(void *data)
{
2012-11-16 11:50:49 +00:00
long long last_rollover = jiffies;
2011-02-23 20:48:06 +00:00
char *id = my_id();
int status = 0;
2011-08-31 11:42:04 +00:00
mars_global = &_global;
2011-02-23 20:48:06 +00:00
if (!id || strlen(id) < 2) {
MARS_ERR("invalid hostname\n");
status = -EFAULT;
goto done;
}
MARS_INF("-------- starting as host '%s' ----------\n", id);
2011-08-31 11:42:04 +00:00
while (_global.global_power.button || !list_empty(&_global.brick_anchor)) {
2011-02-23 20:48:06 +00:00
int status;
loff_t rest_space;
bool exhausted;
bool jammed;
2012-01-17 10:48:56 +00:00
MARS_DBG("-------- NEW ROUND ---------\n");
if (mars_mem_percent < 0)
mars_mem_percent = 0;
if (mars_mem_percent > 70)
mars_mem_percent = 70;
brick_global_memlimit = (long long)brick_global_memavail * mars_mem_percent / 100;
brick_msleep(100);
2012-01-17 10:48:56 +00:00
if (brick_thread_should_stop()) {
2012-08-06 13:04:32 +00:00
_global.global_power.button = false;
mars_net_is_alive = false;
}
_make_alivelink("alive", _global.global_power.button ? 1 : 0);
mars_remaining_space("/mars", &_global.total_space, &_global.remaining_space);
exhausted = EXHAUSTED(_global.remaining_space, _global.total_space);
_global.exhausted = exhausted;
_make_alivelink("exhausted", exhausted ? 1 : 0);
if (exhausted)
MARS_WRN("EXHAUSTED filesystem space = %lld\n", _global.remaining_space);
jammed = JAMMED(_global.remaining_space);
_global.jammed = jammed;
_make_alivelink("jammed", jammed ? 1 : 0);
if (jammed)
MARS_WRN("JAMMED filesystem space = %lld, STOPPING TRANSACTION LOGGING\n", _global.remaining_space);
rest_space = _global.remaining_space - EXHAUSTED_LIMIT(_global.total_space);
_make_alivelink("rest-space", rest_space);
if (!_global.global_power.button && is_shutdown()) {
MARS_INF("global shutdown of all bricks...\n");
brick_msleep(500);
mars_kill_brick_all(&_global, &_global.server_anchor, false);
2011-08-31 11:42:04 +00:00
}
2011-02-23 20:48:06 +00:00
2012-01-17 10:48:56 +00:00
MARS_DBG("-------- start worker ---------\n");
_global.deleted_min = 0;
2011-08-31 11:42:04 +00:00
status = mars_dent_work(&_global, "/mars", sizeof(struct mars_dent), light_checker, light_worker, &_global, 3);
_global.deleted_border = _global.deleted_min;
MARS_DBG("-------- worker deleted_min = %d status = %d\n", _global.deleted_min, status);
2011-02-23 20:48:06 +00:00
if (!_global.global_power.button) {
status = mars_kill_brick_when_possible(&_global, &_global.brick_anchor, false, (void*)&copy_brick_type, false);
MARS_DBG("kill copy bricks (when possible) = %d\n", status);
}
status = mars_kill_brick_when_possible(&_global, &_global.brick_anchor, false, (void*)&client_brick_type, false);
MARS_DBG("kill client bricks (when possible) = %d\n", status);
status = mars_kill_brick_when_possible(&_global, &_global.brick_anchor, false, (void*)&aio_brick_type, false);
MARS_DBG("kill aio bricks (when possible) = %d\n", status);
status = mars_kill_brick_when_possible(&_global, &_global.brick_anchor, false, (void*)&sio_brick_type, false);
MARS_DBG("kill sio bricks (when possible) = %d\n", status);
status = mars_kill_brick_when_possible(&_global, &_global.server_anchor, false, (void*)&aio_brick_type, false);
MARS_DBG("kill server aio bricks (when possible) = %d\n", status);
status = mars_kill_brick_when_possible(&_global, &_global.server_anchor, false, (void*)&sio_brick_type, false);
MARS_DBG("kill server sio bricks (when possible) = %d\n", status);
2012-11-16 11:50:49 +00:00
if ((long long)jiffies + rollover_time * HZ >= last_rollover) {
last_rollover = jiffies;
rollover_all();
}
2011-09-27 11:57:04 +00:00
_show_status_all(&_global);
2011-03-03 09:02:10 +00:00
#ifdef STAT_DEBUGGING
2011-08-31 11:42:04 +00:00
_show_statist(&_global);
2011-03-03 09:02:10 +00:00
#endif
MARS_DBG("ban_count = %d ban_renew_count = %d\n", mars_global_ban.ban_count, mars_global_ban.ban_renew_count);
brick_msleep(500);
2011-02-27 14:17:58 +00:00
2012-01-17 10:48:56 +00:00
wait_event_interruptible_timeout(_global.main_event, _global.main_trigger, CONFIG_MARS_SCAN_INTERVAL * HZ);
2011-08-31 11:42:04 +00:00
_global.main_trigger = false;
2011-02-23 20:48:06 +00:00
}
done:
MARS_INF("-------- cleaning up ----------\n");
2012-01-17 14:37:14 +00:00
mars_remote_trigger();
brick_msleep(2000);
2011-02-23 20:48:06 +00:00
2011-08-31 11:42:04 +00:00
mars_kill_brick_all(&_global, &_global.server_anchor, false);
2011-09-27 08:46:14 +00:00
mars_free_dent_all(&_global, &_global.dent_anchor);
2011-08-31 11:42:04 +00:00
mars_kill_brick_all(&_global, &_global.brick_anchor, false);
2011-02-23 20:48:06 +00:00
2011-09-27 11:57:04 +00:00
_show_status_all(&_global);
2011-08-25 10:16:32 +00:00
#ifdef STAT_DEBUGGING
2011-08-31 11:42:04 +00:00
_show_statist(&_global);
2011-08-25 10:16:32 +00:00
#endif
2011-02-23 20:48:06 +00:00
mars_global = NULL;
MARS_INF("-------- done status = %d ----------\n", status);
2011-03-02 09:30:56 +00:00
//cleanup_mm();
2011-02-23 20:48:06 +00:00
return status;
}
static struct mem_reservation global_reserve = {
.amount = {
[1] = 32,
[2] = 32,
[3] = 32,
2012-02-07 13:15:07 +00:00
[4] = 64,
[5] = 64,
[6] = 64,
2012-01-23 10:23:41 +00:00
[7] = 2,
[8] = 2,
[9] = 1,
[10] = 1,
[11] = 0,
},
};
2011-08-25 10:16:32 +00:00
#ifdef CONFIG_MARS_HAVE_BIGMODULE
#define INIT_MAX 32
static char *exit_names[INIT_MAX] = {};
static void (*exit_fn[INIT_MAX])(void) = {};
static int exit_fn_nr = 0;
#define DO_INIT(name) \
MARS_DBG("=== starting module " #name "...\n"); \
2011-08-25 10:16:32 +00:00
do { \
if ((status = init_##name()) < 0) goto done; \
exit_names[exit_fn_nr] = #name; \
exit_fn[exit_fn_nr++] = exit_##name; \
} while (0)
#endif
2012-01-17 14:37:14 +00:00
void (*_mars_remote_trigger)(void);
EXPORT_SYMBOL_GPL(_mars_remote_trigger);
2011-02-23 20:48:06 +00:00
static void __exit exit_light(void)
{
2011-08-25 10:16:32 +00:00
MARS_DBG("====================== stopping everything...\n");
2011-02-23 20:48:06 +00:00
// TODO: make this thread-safe.
if (main_thread) {
2011-08-25 10:16:32 +00:00
MARS_DBG("=== stopping light thread...\n");
2012-01-17 10:48:56 +00:00
mars_trigger();
MARS_INF("stopping main thread...\n");
brick_thread_stop(main_thread);
2011-02-23 20:48:06 +00:00
}
2011-08-25 10:16:32 +00:00
2012-01-17 14:37:14 +00:00
_mars_remote_trigger = NULL;
brick_allow_freelist = false;
2011-10-04 11:45:09 +00:00
#ifdef CONFIG_MARS_HAVE_BIGMODULE
2011-08-25 10:16:32 +00:00
while (exit_fn_nr > 0) {
MARS_DBG("=== stopping module %s ...\n", exit_names[exit_fn_nr - 1]);
exit_fn[--exit_fn_nr]();
}
2011-10-04 11:45:09 +00:00
#endif
2011-08-25 10:16:32 +00:00
MARS_DBG("====================== stopped everything.\n");
2012-01-10 10:29:22 +00:00
exit_say();
2011-02-23 20:48:06 +00:00
}
int global_logrot_auto = CONFIG_MARS_LOGROT_AUTO;
EXPORT_SYMBOL_GPL(global_logrot_auto);
int global_logdel_auto = CONFIG_MARS_LOGDELETE_AUTO;
EXPORT_SYMBOL_GPL(global_logdel_auto);
int global_free_space = CONFIG_MARS_MIN_SPACE_BASE;
EXPORT_SYMBOL_GPL(global_free_space);
2011-02-23 20:48:06 +00:00
static int __init init_light(void)
{
2011-08-25 10:16:32 +00:00
int status = 0;
init_say(); // this must come first
2011-08-25 10:16:32 +00:00
#ifdef CONFIG_MARS_HAVE_BIGMODULE
/* be careful: order is important!
*/
DO_INIT(brick_mem);
DO_INIT(brick);
DO_INIT(mars);
#ifdef CONFIG_MARS_DEBUG // otherwise currently unused
DO_INIT(mars_dummy);
DO_INIT(mars_check);
DO_INIT(mars_buf);
DO_INIT(mars_usebuf);
#endif
2011-08-25 10:16:32 +00:00
DO_INIT(log_format);
DO_INIT(mars_net);
DO_INIT(mars_server);
DO_INIT(mars_client);
DO_INIT(mars_aio);
DO_INIT(mars_sio);
2011-08-25 10:16:32 +00:00
DO_INIT(mars_bio);
DO_INIT(mars_if);
DO_INIT(mars_copy);
DO_INIT(mars_trans_logger);
DO_INIT(sy);
DO_INIT(sy_net);
DO_INIT(mars_proc);
#endif
brick_mem_reserve(&global_reserve);
main_thread = brick_thread_create(light_thread, NULL, "mars_light");
if (unlikely(!main_thread)) {
status = -ENOENT;
2011-08-25 10:16:32 +00:00
goto done;
2011-02-23 20:48:06 +00:00
}
2011-08-25 10:16:32 +00:00
done:
if (status < 0) {
MARS_ERR("module init failed with status = %d, exiting.\n", status);
exit_light();
}
2012-01-17 14:37:14 +00:00
_mars_remote_trigger = __mars_remote_trigger;
2011-08-25 10:16:32 +00:00
return status;
2011-02-23 20:48:06 +00:00
}
2011-03-03 09:02:10 +00:00
// force module loading
const void *dummy1 = &client_brick_type;
const void *dummy2 = &server_brick_type;
2011-02-23 20:48:06 +00:00
MODULE_DESCRIPTION("MARS Light");
MODULE_AUTHOR("Thomas Schoebel-Theuer <tst@1und1.de>");
2011-10-04 11:45:09 +00:00
MODULE_VERSION(BUILDTAG " (" BUILDHOST " " BUILDDATE ")");
2011-02-23 20:48:06 +00:00
MODULE_LICENSE("GPL");
module_init(init_light);
module_exit(exit_light);