1
0
mirror of https://github.com/schoebel/mars synced 2024-12-30 19:02:51 +00:00
mars/kernel/mars_copy.c

1033 lines
26 KiB
C
Raw Normal View History

2011-02-23 20:48:06 +00:00
// (c) 2010 Thomas Schoebel-Theuer / 1&1 Internet AG
// Copy brick (just for demonstration)
//#define BRICK_DEBUGGING
//#define MARS_DEBUGGING
//#define IO_DEBUGGING
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/string.h>
#include "mars.h"
#include "lib_limiter.h"
2011-02-23 20:48:06 +00:00
2012-02-22 13:34:04 +00:00
#ifndef READ
#define READ 0
#define WRITE 1
#endif
#define COPY_CHUNK (PAGE_SIZE)
#define NR_COPY_REQUESTS (32 * 1024 * 1024 / COPY_CHUNK)
#define STATES_PER_PAGE (PAGE_SIZE / sizeof(struct copy_state))
#define MAX_SUB_TABLES (NR_COPY_REQUESTS / STATES_PER_PAGE + (NR_COPY_REQUESTS % STATES_PER_PAGE ? 1 : 0))
#define MAX_COPY_REQUESTS (PAGE_SIZE / sizeof(struct copy_state*) * STATES_PER_PAGE)
#define GET_STATE(brick,index) \
((brick)->st[(index) / STATES_PER_PAGE][(index) % STATES_PER_PAGE])
2011-02-23 20:48:06 +00:00
///////////////////////// own type definitions ////////////////////////
#include "mars_copy.h"
int mars_copy_overlap = 1;
EXPORT_SYMBOL_GPL(mars_copy_overlap);
int mars_copy_read_prio = MARS_PRIO_NORMAL;
EXPORT_SYMBOL_GPL(mars_copy_read_prio);
int mars_copy_write_prio = MARS_PRIO_NORMAL;
EXPORT_SYMBOL_GPL(mars_copy_write_prio);
2013-07-22 07:15:53 +00:00
int mars_copy_read_max_fly = 0;
EXPORT_SYMBOL_GPL(mars_copy_read_max_fly);
int mars_copy_write_max_fly = 0;
EXPORT_SYMBOL_GPL(mars_copy_write_max_fly);
#define is_read_limited(brick) \
(mars_copy_read_max_fly > 0 && atomic_read(&(brick)->copy_read_flight) >= mars_copy_read_max_fly)
#define is_write_limited(brick) \
(mars_copy_write_max_fly > 0 && atomic_read(&(brick)->copy_write_flight) >= mars_copy_write_max_fly)
2011-02-23 20:48:06 +00:00
///////////////////////// own helper functions ////////////////////////
/* TODO:
* The clash logic is untested / alpha stage (Feb. 2011).
*
* For now, the output is never used, so this cannot do harm.
*
* In order to get the output really working / enterprise grade,
* some larger test effort should be invested.
*/
static inline
void _clash(struct copy_brick *brick)
{
brick->trigger = true;
set_bit(0, &brick->clash);
2012-12-20 10:22:41 +00:00
atomic_inc(&brick->total_clash_count);
2011-02-23 20:48:06 +00:00
wake_up_interruptible(&brick->event);
}
static inline
int _clear_clash(struct copy_brick *brick)
{
int old;
old = test_and_clear_bit(0, &brick->clash);
return old;
}
/* Current semantics:
*
* All writes are always going to the original input A. They are _not_
* replicated to B.
*
* In order to get B really uptodate, you have to replay the right
* transaction logs there (at the right time).
* [If you had no writes on A at all during the copy, of course
* this is not necessary]
*
2011-05-26 14:32:32 +00:00
* When utilize_mode is on, reads can utilize the already copied
2011-02-23 20:48:06 +00:00
* region from B, but only as long as this region has not been
* invalidated by writes (indicated by low_dirty).
*
* TODO: implement replicated writes, together with some transaction
* replay logic applying the transaction logs _only_ after
* crashes during inconsistency caused by partial replication of writes.
*/
static
int _determine_input(struct copy_brick *brick, struct mref_object *mref)
{
int rw;
int below;
int behind;
loff_t ref_end;
2011-05-26 14:32:32 +00:00
if (!brick->utilize_mode || brick->low_dirty)
2011-02-23 20:48:06 +00:00
return INPUT_A_IO;
ref_end = mref->ref_pos + mref->ref_len;
below = ref_end <= brick->copy_start;
behind = !brick->copy_end || mref->ref_pos >= brick->copy_end;
rw = mref->ref_may_write | mref->ref_rw;
if (rw) {
if (!behind) {
brick->low_dirty = true;
if (!below) {
_clash(brick);
wake_up_interruptible(&brick->event);
}
}
return INPUT_A_IO;
}
if (below)
return INPUT_B_IO;
return INPUT_A_IO;
}
#define GET_INDEX(pos) (((pos) / COPY_CHUNK) % NR_COPY_REQUESTS)
2011-06-10 13:57:52 +00:00
#define GET_OFFSET(pos) ((pos) % COPY_CHUNK)
2011-02-23 20:48:06 +00:00
2012-01-23 12:39:08 +00:00
static
void __clear_mref(struct copy_brick *brick, struct mref_object *mref, int queue)
{
struct copy_input *input;
input = queue ? brick->inputs[INPUT_B_COPY] : brick->inputs[INPUT_A_COPY];
GENERIC_INPUT_CALL(input, mref_put, mref);
}
static
void _clear_mref(struct copy_brick *brick, int index, int queue)
{
struct copy_state *st = &GET_STATE(brick, index);
2012-12-20 16:54:22 +00:00
struct mref_object *mref = st->table[queue];
2012-01-23 12:39:08 +00:00
if (mref) {
2012-12-20 16:54:22 +00:00
if (unlikely(st->active[queue])) {
MARS_ERR("clearing active mref, index = %d queue = %d\n", index, queue);
st->active[queue] = false;
}
2012-01-23 12:39:08 +00:00
__clear_mref(brick, mref, queue);
st->table[queue] = NULL;
2012-01-23 12:39:08 +00:00
}
}
static
void _clear_all_mref(struct copy_brick *brick)
{
int i;
for (i = 0; i < NR_COPY_REQUESTS; i++) {
GET_STATE(brick, i).state = COPY_STATE_START;
2012-01-23 12:39:08 +00:00
_clear_mref(brick, i, 0);
_clear_mref(brick, i, 1);
}
}
static
void _clear_state_table(struct copy_brick *brick)
{
int i;
for (i = 0; i < MAX_SUB_TABLES; i++) {
struct copy_state *sub_table = brick->st[i];
memset(sub_table, 0, PAGE_SIZE);
}
}
2011-02-23 20:48:06 +00:00
static
void copy_endio(struct generic_callback *cb)
{
struct copy_mref_aspect *mref_a;
struct mref_object *mref;
struct copy_brick *brick;
2011-06-10 13:57:52 +00:00
struct copy_state *st;
2011-02-23 20:48:06 +00:00
int index;
int queue;
2011-06-10 13:57:52 +00:00
int error = 0;
2011-02-23 20:48:06 +00:00
mref_a = cb->cb_private;
CHECK_PTR(mref_a, err);
mref = mref_a->object;
CHECK_PTR(mref, err);
brick = mref_a->brick;
CHECK_PTR(brick, err);
queue = mref_a->queue;
2011-06-10 13:57:52 +00:00
index = GET_INDEX(mref->ref_pos);
st = &GET_STATE(brick, index);
2011-06-10 13:57:52 +00:00
2011-02-23 20:48:06 +00:00
MARS_IO("queue = %d index = %d pos = %lld status = %d\n", queue, index, mref->ref_pos, cb->cb_error);
if (unlikely(queue < 0 || queue >= 2)) {
MARS_ERR("bad queue %d\n", queue);
2011-06-10 13:57:52 +00:00
error = -EINVAL;
2011-02-23 20:48:06 +00:00
goto exit;
}
2012-12-20 16:54:22 +00:00
st->active[queue] = false;
2011-06-10 13:57:52 +00:00
if (unlikely(st->table[queue])) {
MARS_ERR("table corruption at %d %d (%p => %p)\n", index, queue, st->table[queue], mref);
error = -EEXIST;
2011-02-23 20:48:06 +00:00
goto exit;
}
if (unlikely(cb->cb_error < 0)) {
2011-06-10 13:57:52 +00:00
error = cb->cb_error;
2012-08-22 12:37:34 +00:00
__clear_mref(brick, mref, queue);
/* This is racy, but does no harm.
* Worst case just produces more error output.
*/
if (!brick->copy_error_count++) {
MARS_WRN("IO error %d on index %d, old state = %d\n", cb->cb_error, index, st->state);
}
} else {
if (unlikely(st->table[queue])) {
MARS_ERR("overwriting index %d, state = %d\n", index, st->state);
_clear_mref(brick, index, queue);
}
2011-06-10 13:57:52 +00:00
st->table[queue] = mref;
2011-02-23 20:48:06 +00:00
}
exit:
if (unlikely(error < 0)) {
2011-06-10 13:57:52 +00:00
st->error = error;
_clash(brick);
}
2013-07-22 07:15:53 +00:00
if (mref->ref_rw) {
atomic_dec(&brick->copy_write_flight);
} else {
atomic_dec(&brick->copy_read_flight);
}
2011-02-23 20:48:06 +00:00
brick->trigger = true;
wake_up_interruptible(&brick->event);
return;
err:
MARS_FAT("cannot handle callback\n");
}
static
2012-08-02 08:16:55 +00:00
int _make_mref(struct copy_brick *brick, int index, int queue, void *data, loff_t pos, loff_t end_pos, int rw, int cs_mode)
2011-02-23 20:48:06 +00:00
{
struct mref_object *mref;
struct copy_mref_aspect *mref_a;
struct copy_input *input;
2011-06-10 13:57:52 +00:00
int offset;
2011-02-23 20:48:06 +00:00
int len;
2012-02-22 13:34:04 +00:00
int status = -EAGAIN;
2011-02-23 20:48:06 +00:00
2012-02-22 13:34:04 +00:00
if (brick->clash || end_pos <= 0)
2011-02-23 20:48:06 +00:00
goto done;
2012-02-02 15:25:43 +00:00
mref = copy_alloc_mref(brick);
2011-02-23 20:48:06 +00:00
status = -ENOMEM;
if (unlikely(!mref))
goto done;
mref_a = copy_mref_get_aspect(brick, mref);
2011-02-23 20:48:06 +00:00
if (unlikely(!mref_a)) {
2011-08-12 11:09:48 +00:00
MARS_FAT("cannot get own apsect\n");
2011-02-23 20:48:06 +00:00
goto done;
}
mref_a->brick = brick;
mref_a->queue = queue;
mref->ref_may_write = rw;
mref->ref_rw = rw;
mref->ref_data = data;
mref->ref_pos = pos;
2012-08-02 08:16:55 +00:00
mref->ref_cs_mode = cs_mode;
2011-06-10 13:57:52 +00:00
offset = GET_OFFSET(pos);
len = COPY_CHUNK - offset;
2012-02-22 13:34:04 +00:00
if (pos + len > end_pos) {
len = end_pos - pos;
2011-02-23 20:48:06 +00:00
}
mref->ref_len = len;
mref->ref_prio = rw ?
mars_copy_write_prio :
mars_copy_read_prio;
if (mref->ref_prio < MARS_PRIO_HIGH || mref->ref_prio > MARS_PRIO_LOW)
mref->ref_prio = brick->io_prio;
SETUP_CALLBACK(mref, copy_endio, mref_a);
2011-02-23 20:48:06 +00:00
input = queue ? brick->inputs[INPUT_B_COPY] : brick->inputs[INPUT_A_COPY];
status = GENERIC_INPUT_CALL(input, mref_get, mref);
if (unlikely(status < 0)) {
MARS_ERR("status = %d\n", status);
mars_free_mref(mref);
goto done;
}
2011-03-22 14:36:26 +00:00
if (unlikely(mref->ref_len < len)) {
2011-06-10 13:57:52 +00:00
MARS_DBG("shorten len %d < %d\n", mref->ref_len, len);
2011-03-22 14:36:26 +00:00
}
2012-02-22 13:34:04 +00:00
if (queue == 0) {
GET_STATE(brick, index).len = mref->ref_len;
} else if (unlikely(mref->ref_len < GET_STATE(brick, index).len)) {
2013-08-06 08:29:12 +00:00
MARS_DBG("shorten len %d < %d at index %d\n", mref->ref_len, GET_STATE(brick, index).len, index);
GET_STATE(brick, index).len = mref->ref_len;
2012-02-22 13:34:04 +00:00
}
2011-02-23 20:48:06 +00:00
//MARS_IO("queue = %d index = %d pos = %lld len = %d rw = %d\n", queue, index, mref->ref_pos, mref->ref_len, rw);
2011-02-23 20:48:06 +00:00
GET_STATE(brick, index).active[queue] = true;
2013-07-22 07:15:53 +00:00
if (rw) {
atomic_inc(&brick->copy_write_flight);
} else {
atomic_inc(&brick->copy_read_flight);
}
2011-02-23 20:48:06 +00:00
GENERIC_INPUT_CALL(input, mref_io, mref);
done:
return status;
}
2011-03-07 05:55:10 +00:00
static
void _update_percent(struct copy_brick *brick)
{
2011-06-17 11:32:38 +00:00
if (brick->copy_last > brick->copy_start + 8 * 1024 * 1024
2011-03-07 05:55:10 +00:00
|| (long long)jiffies > brick->last_jiffies + 5 * HZ
2011-06-17 11:32:38 +00:00
|| (brick->copy_last == brick->copy_end && brick->copy_end > 0)) {
brick->copy_start = brick->copy_last;
2011-03-07 05:55:10 +00:00
brick->last_jiffies = jiffies;
brick->power.percent_done = brick->copy_end > 0 ? brick->copy_start * 100 / brick->copy_end : 0;
MARS_INF("'%s' copied %lld / %lld bytes (%d%%)\n", brick->brick_path, brick->copy_last, brick->copy_end, brick->power.percent_done);
2011-03-07 05:55:10 +00:00
}
}
2012-08-02 08:16:55 +00:00
/* The heart of this brick.
* State transition function of the finite automaton.
* In case no progress is possible (e.g. preconditions not
* yet true), the state is left as is (idempotence property:
* calling this too often does no harm, just costs performance).
*/
2011-02-23 20:48:06 +00:00
static
2011-06-10 13:57:52 +00:00
int _next_state(struct copy_brick *brick, int index, loff_t pos)
2011-02-23 20:48:06 +00:00
{
2011-06-17 11:32:38 +00:00
struct mref_object *mref0;
2011-02-23 20:48:06 +00:00
struct mref_object *mref1;
2011-06-10 13:57:52 +00:00
struct copy_state *st;
2011-02-23 20:48:06 +00:00
char state;
char next_state;
2013-08-06 08:29:12 +00:00
bool do_restart = false;
2012-08-22 12:37:34 +00:00
int progress = 0;
2011-02-23 20:48:06 +00:00
int status;
st = &GET_STATE(brick, index);
2012-08-22 12:37:34 +00:00
next_state = st->state;
restart:
state = next_state;
2011-02-23 20:48:06 +00:00
2013-08-06 08:29:12 +00:00
MARS_IO("ENTER index=%d state=%d pos=%lld table[0]=%p table[1]=%p active[0]=%d active[1]=%d writeout=%d prev=%d len=%d error=%d do_restart=%d\n",
index,
state,
pos,
st->table[0],
st->table[1],
st->active[0],
st->active[1],
st->writeout,
st->prev,
st->len,
st->error,
do_restart);
do_restart = false;
2011-02-23 20:48:06 +00:00
switch (state) {
case COPY_STATE_RESET:
/* This state is only entered after errors or
* in restarting situations.
*/
_clear_mref(brick, index, 1);
_clear_mref(brick, index, 0);
next_state = COPY_STATE_START;
/* fallthrough */
2011-02-23 20:48:06 +00:00
case COPY_STATE_START:
/* This is the relgular starting state.
* It must be zero, automatically entered via memset()
*/
2011-06-10 13:57:52 +00:00
if (st->table[0] || st->table[1]) {
2011-02-23 20:48:06 +00:00
MARS_ERR("index %d not startable\n", index);
2012-08-22 12:37:34 +00:00
progress = -EPROTO;
goto idle;
2011-02-23 20:48:06 +00:00
}
2012-02-22 13:34:04 +00:00
2012-01-23 12:39:08 +00:00
_clear_mref(brick, index, 1);
_clear_mref(brick, index, 0);
2012-12-20 16:54:22 +00:00
st->writeout = false;
st->error = 0;
2012-01-23 12:39:08 +00:00
2013-07-22 07:15:53 +00:00
if (brick->is_aborting ||
is_read_limited(brick))
goto idle;
2012-08-02 08:16:55 +00:00
status = _make_mref(brick, index, 0, NULL, pos, brick->copy_end, READ, brick->verify_mode ? 2 : 0);
2012-02-22 13:34:04 +00:00
if (unlikely(status < 0)) {
MARS_WRN("status = %d\n", status);
2012-08-22 12:37:34 +00:00
progress = status;
break;
2012-02-22 13:34:04 +00:00
}
2011-02-23 20:48:06 +00:00
next_state = COPY_STATE_READ1;
2012-08-02 08:16:55 +00:00
if (!brick->verify_mode) {
break;
}
next_state = COPY_STATE_START2;
/* fallthrough */
case COPY_STATE_START2:
status = _make_mref(brick, index, 1, NULL, pos, brick->copy_end, READ, 2);
2012-08-02 08:16:55 +00:00
if (unlikely(status < 0)) {
MARS_WRN("status = %d\n", status);
2012-08-22 12:37:34 +00:00
progress = status;
break;
2011-02-23 20:48:06 +00:00
}
next_state = COPY_STATE_READ2;
/* fallthrough */
2011-02-23 20:48:06 +00:00
case COPY_STATE_READ2:
2011-06-17 11:32:38 +00:00
mref1 = st->table[1];
2012-08-02 08:16:55 +00:00
if (!mref1) { // idempotence: wait by unchanged state
goto idle;
2011-02-23 20:48:06 +00:00
}
/* fallthrough => wait for both mrefs to appear */
2011-02-23 20:48:06 +00:00
case COPY_STATE_READ1:
2012-08-02 08:16:55 +00:00
case COPY_STATE_READ3:
2011-06-17 11:32:38 +00:00
mref0 = st->table[0];
2012-08-02 08:16:55 +00:00
if (!mref0) { // idempotence: wait by unchanged state
goto idle;
2011-02-23 20:48:06 +00:00
}
2013-07-15 10:12:54 +00:00
if (brick->copy_limiter) {
int amount = (mref0->ref_len - 1) / 1024 + 1;
mars_limit_sleep(brick->copy_limiter, amount);
}
2011-06-17 11:32:38 +00:00
// on append mode: increase the end pointer dynamically
if (brick->append_mode > 0 && mref0->ref_total_size && mref0->ref_total_size > brick->copy_end) {
brick->copy_end = mref0->ref_total_size;
2011-05-26 14:32:32 +00:00
}
2011-06-17 11:32:38 +00:00
// do verify (when applicable)
mref1 = st->table[1];
2012-08-02 08:16:55 +00:00
if (mref1 && state != COPY_STATE_READ3) {
2011-06-17 11:32:38 +00:00
int len = mref0->ref_len;
2012-08-02 08:16:55 +00:00
bool ok;
if (len != mref1->ref_len) {
ok = false;
} else if (mref0->ref_cs_mode) {
static unsigned char null[sizeof(mref0->ref_checksum)];
ok = !memcmp(mref0->ref_checksum, mref1->ref_checksum, sizeof(mref0->ref_checksum));
if (ok)
ok = memcmp(mref0->ref_checksum, null, sizeof(mref0->ref_checksum)) != 0;
} else if (!mref0->ref_data || !mref1->ref_data) {
ok = false;
} else {
ok = !memcmp(mref0->ref_data, mref1->ref_data, len);
}
2011-06-17 11:32:38 +00:00
_clear_mref(brick, index, 1);
2012-08-02 08:16:55 +00:00
if (ok)
brick->verify_ok_count++;
else
brick->verify_error_count++;
if (ok || !brick->repair_mode) {
2011-02-23 20:48:06 +00:00
/* skip start of writing, goto final treatment of writeout */
2012-08-02 08:16:55 +00:00
next_state = COPY_STATE_CLEANUP;
break;
2011-02-23 20:48:06 +00:00
}
}
2012-08-02 08:16:55 +00:00
if (mref0->ref_cs_mode > 1) { // re-read, this time with data
2012-08-02 08:16:55 +00:00
_clear_mref(brick, index, 0);
status = _make_mref(brick, index, 0, NULL, pos, brick->copy_end, READ, 0);
if (unlikely(status < 0)) {
MARS_WRN("status = %d\n", status);
2012-08-22 12:37:34 +00:00
progress = status;
next_state = COPY_STATE_RESET;
break;
2012-08-02 08:16:55 +00:00
}
next_state = COPY_STATE_READ3;
break;
}
2011-02-23 20:48:06 +00:00
next_state = COPY_STATE_WRITE;
2011-06-17 11:32:38 +00:00
/* fallthrough */
2011-02-23 20:48:06 +00:00
case COPY_STATE_WRITE:
2013-07-22 07:15:53 +00:00
if (is_write_limited(brick))
goto idle;
2011-06-17 11:32:38 +00:00
/* Obey ordering to get a strict "append" behaviour.
* We assume that we don't need to wait for completion
* of the previous write to avoid a sparse result file
* under all circumstances, i.e. we only assure that
* _starting_ the writes is in order.
* This is only correct when all lower bricks obey the
* order of ref_io() operations.
* Currenty, bio and aio are obeying this. Be careful when
* implementing new IO bricks!
*/
if (st->prev >= 0 && !GET_STATE(brick, st->prev).writeout) {
goto idle;
2011-02-23 20:48:06 +00:00
}
2011-06-17 11:32:38 +00:00
mref0 = st->table[0];
2012-08-02 08:16:55 +00:00
if (unlikely(!mref0 || !mref0->ref_data)) {
2012-01-31 13:05:18 +00:00
MARS_ERR("src buffer for write does not exist, state %d at index %d\n", state, index);
2012-08-22 12:37:34 +00:00
progress = -EILSEQ;
break;
2011-06-17 11:32:38 +00:00
}
2012-08-22 12:37:34 +00:00
if (unlikely(brick->is_aborting)) {
progress = -EINTR;
break;
2012-08-22 12:37:34 +00:00
}
2011-06-17 11:32:38 +00:00
/* start writeout */
2012-08-02 08:16:55 +00:00
status = _make_mref(brick, index, 1, mref0->ref_data, pos, pos + mref0->ref_len, WRITE, 0);
2012-08-22 12:37:34 +00:00
if (unlikely(status < 0)) {
MARS_WRN("status = %d\n", status);
progress = status;
next_state = COPY_STATE_RESET;
2012-08-22 12:37:34 +00:00
break;
}
/* Attention! overlapped IO behind EOF could
* lead to temporary inconsistent state of the
* file, because the write order may be different from
* strict O_APPEND behaviour.
*/
if (mars_copy_overlap)
st->writeout = true;
2011-06-17 11:32:38 +00:00
next_state = COPY_STATE_WRITTEN;
/* fallthrough */
2011-06-17 11:32:38 +00:00
case COPY_STATE_WRITTEN:
mref1 = st->table[1];
2012-08-02 08:16:55 +00:00
if (!mref1) { // idempotence: wait by unchanged state
2011-06-17 11:32:38 +00:00
MARS_IO("irrelevant\n");
goto idle;
}
st->writeout = true;
/* rechecking means to start over again.
* ATTENTIION! this may lead to infinite request
* submission loops, intentionally.
* TODO: implement some timeout means.
*/
if (brick->recheck_mode && brick->repair_mode) {
next_state = COPY_STATE_RESET;
break;
2011-02-23 20:48:06 +00:00
}
next_state = COPY_STATE_CLEANUP;
/* fallthrough */
case COPY_STATE_CLEANUP:
_clear_mref(brick, index, 1);
2011-06-17 11:32:38 +00:00
_clear_mref(brick, index, 0);
next_state = COPY_STATE_FINISHED;
/* fallthrough */
2011-06-17 11:32:38 +00:00
case COPY_STATE_FINISHED:
2012-08-02 08:16:55 +00:00
/* Indicate successful completion by remaining in this state.
* Restart of the finite automaton must be done externally.
*/
goto idle;
2011-02-23 20:48:06 +00:00
default:
MARS_ERR("illegal state %d at index %d\n", state, index);
_clash(brick);
2012-08-22 12:37:34 +00:00
progress = -EILSEQ;
2011-02-23 20:48:06 +00:00
}
2012-08-22 12:37:34 +00:00
do_restart = (state != next_state);
idle:
2012-08-22 12:37:34 +00:00
if (unlikely(progress < 0)) {
st->error = progress;
MARS_WRN("progress = %d\n", progress);
progress = 0;
2011-02-23 20:48:06 +00:00
_clash(brick);
} else if (do_restart) {
goto restart;
2012-08-22 12:37:34 +00:00
} else if (st->state != next_state) {
progress++;
2011-02-23 20:48:06 +00:00
}
2013-08-06 08:29:12 +00:00
MARS_IO("LEAVE index=%d state=%d next_state=%d table[0]=%p table[1]=%p active[0]=%d active[1]=%d writeout=%d prev=%d len=%d error=%d progress=%d\n",
index,
st->state,
next_state,
st->table[0],
st->table[1],
st->active[0],
st->active[1],
st->writeout,
st->prev,
st->len,
st->error,
progress);
2012-08-22 12:37:34 +00:00
// save the resulting state
st->state = next_state;
return progress;
2011-02-23 20:48:06 +00:00
}
static
int _run_copy(struct copy_brick *brick)
2011-02-23 20:48:06 +00:00
{
int max;
loff_t pos;
2012-02-22 13:34:04 +00:00
loff_t limit = -1;
2011-06-17 11:32:38 +00:00
short prev;
2012-08-22 12:37:34 +00:00
int progress;
2011-02-23 20:48:06 +00:00
2011-06-17 11:32:38 +00:00
if (unlikely(_clear_clash(brick))) {
2011-02-23 20:48:06 +00:00
MARS_DBG("clash\n");
2013-07-22 07:15:53 +00:00
if (atomic_read(&brick->copy_read_flight) + atomic_read(&brick->copy_write_flight) > 0) {
2011-02-23 20:48:06 +00:00
/* wait until all pending copy IO has finished
*/
_clash(brick);
MARS_DBG("re-clash\n");
brick_msleep(100);
return 0;
2011-02-23 20:48:06 +00:00
}
_clear_all_mref(brick);
_clear_state_table(brick);
2011-02-23 20:48:06 +00:00
}
2011-06-10 13:57:52 +00:00
/* Do at most max iterations in the below loop
*/
max = NR_COPY_REQUESTS - atomic_read(&brick->io_flight) * 2;
2011-02-23 20:48:06 +00:00
MARS_IO("max = %d\n", max);
2011-06-17 11:32:38 +00:00
prev = -1;
2012-08-22 12:37:34 +00:00
progress = 0;
2011-06-17 11:32:38 +00:00
for (pos = brick->copy_last; pos < brick->copy_end || brick->append_mode > 1; pos = ((pos / COPY_CHUNK) + 1) * COPY_CHUNK) {
2011-06-10 13:57:52 +00:00
int index = GET_INDEX(pos);
struct copy_state *st = &GET_STATE(brick, index);
if (max-- <= 0) {
2011-02-23 20:48:06 +00:00
break;
2011-05-26 14:32:32 +00:00
}
2011-06-17 11:32:38 +00:00
st->prev = prev;
prev = index;
// call the finite state automaton
2012-08-22 12:37:34 +00:00
if (!(st->active[0] | st->active[1])) {
progress += _next_state(brick, index, pos);
2011-06-17 11:32:38 +00:00
limit = pos;
2011-06-10 13:57:52 +00:00
}
}
2011-06-17 11:32:38 +00:00
// check the resulting state: can we advance the copy_last pointer?
2012-08-22 12:37:34 +00:00
if (likely(progress && !brick->clash)) {
2011-06-10 13:57:52 +00:00
int count = 0;
2011-06-17 11:32:38 +00:00
for (pos = brick->copy_last; pos <= limit; pos = ((pos / COPY_CHUNK) + 1) * COPY_CHUNK) {
2011-06-10 13:57:52 +00:00
int index = GET_INDEX(pos);
struct copy_state *st = &GET_STATE(brick, index);
2011-06-17 11:32:38 +00:00
if (st->state != COPY_STATE_FINISHED) {
2011-06-10 13:57:52 +00:00
break;
2011-06-17 11:32:38 +00:00
}
if (unlikely(st->error < 0)) {
2012-08-22 12:37:34 +00:00
if (!brick->copy_error) {
brick->copy_error = st->error;
MARS_WRN("IO error = %d\n", st->error);
}
if (brick->abort_mode) {
brick->is_aborting = true;
}
2011-06-17 11:32:38 +00:00
break;
}
// rollover
st->state = COPY_STATE_START;
2011-06-17 11:32:38 +00:00
count += st->len;
// check contiguity
2012-02-22 13:34:04 +00:00
if (unlikely(GET_OFFSET(pos) + st->len != COPY_CHUNK)) {
2011-06-17 11:32:38 +00:00
break;
}
2011-06-10 13:57:52 +00:00
}
if (count > 0) {
2011-06-17 11:32:38 +00:00
brick->copy_last += count;
MARS_IO("new copy_last += %d => %lld\n", count, brick->copy_last);
2011-06-10 13:57:52 +00:00
_update_percent(brick);
}
2011-02-23 20:48:06 +00:00
}
2012-08-22 12:37:34 +00:00
return progress;
2011-02-23 20:48:06 +00:00
}
static
bool _is_done(struct copy_brick *brick)
{
if (brick_thread_should_stop())
2012-08-22 12:37:34 +00:00
brick->is_aborting = true;
return brick->is_aborting &&
2013-07-22 07:15:53 +00:00
atomic_read(&brick->copy_read_flight) + atomic_read(&brick->copy_write_flight) <= 0;
}
2011-02-23 20:48:06 +00:00
static int _copy_thread(void *data)
{
struct copy_brick *brick = data;
2012-08-22 12:37:34 +00:00
int rounds = 0;
2011-02-23 20:48:06 +00:00
MARS_DBG("--------------- copy_thread %p starting\n", brick);
brick->copy_error = 0;
2012-08-22 12:37:34 +00:00
brick->copy_error_count = 0;
2012-08-02 08:16:55 +00:00
brick->verify_ok_count = 0;
brick->verify_error_count = 0;
2011-02-23 20:48:06 +00:00
mars_power_led_on((void*)brick, true);
brick->trigger = true;
while (!_is_done(brick)) {
2011-02-23 20:48:06 +00:00
loff_t old_start = brick->copy_start;
loff_t old_end = brick->copy_end;
2012-08-22 12:37:34 +00:00
int progress = 0;
2011-06-10 13:57:52 +00:00
if (old_end > 0) {
2012-08-22 12:37:34 +00:00
progress = _run_copy(brick);
if (!progress || ++rounds > 1000) {
rounds = 0;
}
2011-06-10 13:57:52 +00:00
}
2011-02-23 20:48:06 +00:00
wait_event_interruptible_timeout(brick->event,
2012-08-22 12:37:34 +00:00
progress > 0 ||
brick->trigger ||
brick->copy_start != old_start ||
brick->copy_end != old_end ||
_is_done(brick),
1 * HZ);
2011-02-23 20:48:06 +00:00
brick->trigger = false;
}
2013-07-22 07:15:53 +00:00
MARS_DBG("--------------- copy_thread terminating (%d read requests / %d write requests flying, copy_start = %lld copy_end = %lld)\n",
atomic_read(&brick->copy_read_flight),
atomic_read(&brick->copy_write_flight),
brick->copy_start,
brick->copy_end);
_clear_all_mref(brick);
2011-02-23 20:48:06 +00:00
mars_power_led_off((void*)brick, true);
MARS_DBG("--------------- copy_thread done.\n");
return 0;
}
////////////////// own brick / input / output operations //////////////////
static int copy_get_info(struct copy_output *output, struct mars_info *info)
{
struct copy_input *input = output->brick->inputs[INPUT_B_IO];
return GENERIC_INPUT_CALL(input, mars_get_info, info);
}
static int copy_ref_get(struct copy_output *output, struct mref_object *mref)
{
struct copy_input *input;
int index;
int status;
index = _determine_input(output->brick, mref);
input = output->brick->inputs[index];
status = GENERIC_INPUT_CALL(input, mref_get, mref);
if (status >= 0) {
atomic_inc(&output->brick->io_flight);
}
return status;
}
static void copy_ref_put(struct copy_output *output, struct mref_object *mref)
{
struct copy_input *input;
int index;
index = _determine_input(output->brick, mref);
input = output->brick->inputs[index];
GENERIC_INPUT_CALL(input, mref_put, mref);
if (atomic_dec_and_test(&output->brick->io_flight)) {
output->brick->trigger = true;
wake_up_interruptible(&output->brick->event);
}
}
static void copy_ref_io(struct copy_output *output, struct mref_object *mref)
{
struct copy_input *input;
int index;
index = _determine_input(output->brick, mref);
input = output->brick->inputs[index];
GENERIC_INPUT_CALL(input, mref_io, mref);
}
static int copy_switch(struct copy_brick *brick)
{
static int version = 0;
MARS_DBG("power.button = %d\n", brick->power.button);
if (brick->power.button) {
if (brick->power.led_on)
goto done;
2011-02-23 20:48:06 +00:00
mars_power_led_off((void*)brick, false);
brick->is_aborting = false;
2011-02-23 20:48:06 +00:00
if (!brick->thread) {
2011-02-25 11:46:38 +00:00
brick->copy_last = brick->copy_start;
brick->thread = brick_thread_create(_copy_thread, brick, "mars_copy%d", version++);
2011-02-23 20:48:06 +00:00
if (brick->thread) {
brick->trigger = true;
} else {
mars_power_led_off((void*)brick, true);
MARS_ERR("could not start copy thread\n");
}
}
} else {
if (brick->power.led_off)
goto done;
2011-02-23 20:48:06 +00:00
mars_power_led_on((void*)brick, false);
if (brick->thread) {
2011-08-25 10:16:32 +00:00
MARS_INF("stopping thread...\n");
brick_thread_stop(brick->thread);
2011-02-23 20:48:06 +00:00
}
}
2011-03-07 05:55:10 +00:00
_update_percent(brick);
done:
2011-02-23 20:48:06 +00:00
return 0;
}
2011-06-17 11:32:38 +00:00
//////////////// informational / statistics ///////////////
static
char *copy_statistics(struct copy_brick *brick, int verbose)
{
char *res = brick_string_alloc(1024);
2011-06-17 11:32:38 +00:00
if (!res)
return NULL;
snprintf(res, 1024,
2012-02-12 11:19:57 +00:00
"copy_start = %lld "
"copy_last = %lld "
"copy_end = %lld "
"copy_error = %d "
2012-08-22 12:37:34 +00:00
"copy_error_count = %d "
2012-08-02 08:16:55 +00:00
"verify_ok_count = %d "
"verify_error_count = %d "
2012-02-12 11:19:57 +00:00
"low_dirty = %d "
"is_aborting = %d "
"clash = %lu | "
2012-12-20 10:22:41 +00:00
"total clash_count = %d | "
2012-02-12 11:19:57 +00:00
"io_flight = %d "
2013-07-22 07:15:53 +00:00
"copy_read_flight = %d "
"copy_write_flight = %d\n",
2012-02-12 11:19:57 +00:00
brick->copy_start,
brick->copy_last,
brick->copy_end,
brick->copy_error,
2012-08-22 12:37:34 +00:00
brick->copy_error_count,
2012-08-02 08:16:55 +00:00
brick->verify_ok_count,
brick->verify_error_count,
2012-02-12 11:19:57 +00:00
brick->low_dirty,
brick->is_aborting,
brick->clash,
2012-12-20 10:22:41 +00:00
atomic_read(&brick->total_clash_count),
2012-02-12 11:19:57 +00:00
atomic_read(&brick->io_flight),
2013-07-22 07:15:53 +00:00
atomic_read(&brick->copy_read_flight),
atomic_read(&brick->copy_write_flight));
2011-06-17 11:32:38 +00:00
return res;
}
static
void copy_reset_statistics(struct copy_brick *brick)
{
2012-12-20 10:22:41 +00:00
atomic_set(&brick->total_clash_count, 0);
2011-06-17 11:32:38 +00:00
}
2011-02-23 20:48:06 +00:00
//////////////// object / aspect constructors / destructors ///////////////
static int copy_mref_aspect_init_fn(struct generic_aspect *_ini)
2011-02-23 20:48:06 +00:00
{
struct copy_mref_aspect *ini = (void*)_ini;
(void)ini;
return 0;
}
static void copy_mref_aspect_exit_fn(struct generic_aspect *_ini)
2011-02-23 20:48:06 +00:00
{
struct copy_mref_aspect *ini = (void*)_ini;
(void)ini;
}
MARS_MAKE_STATICS(copy);
////////////////////// brick constructors / destructors ////////////////////
static
void _free_pages(struct copy_brick *brick)
{
int i;
for (i = 0; i < MAX_SUB_TABLES; i++) {
struct copy_state *sub_table = brick->st[i];
if (!sub_table) {
continue;
}
brick_block_free(sub_table, PAGE_SIZE);
}
brick_block_free(brick->st, PAGE_SIZE);
}
2011-02-23 20:48:06 +00:00
static int copy_brick_construct(struct copy_brick *brick)
{
int i;
brick->st = brick_block_alloc(0, PAGE_SIZE);
if (unlikely(!brick->st)) {
MARS_ERR("cannot allocate state directory table.\n");
return -ENOMEM;
}
memset(brick->st, 0, PAGE_SIZE);
for (i = 0; i < MAX_SUB_TABLES; i++) {
struct copy_state *sub_table;
// this should be usually optimized away as dead code
if (unlikely(i >= MAX_SUB_TABLES)) {
MARS_ERR("sorry, subtable index %d is too large.\n", i);
_free_pages(brick);
return -EINVAL;
}
sub_table = brick_block_alloc(0, PAGE_SIZE);
brick->st[i] = sub_table;
if (unlikely(!sub_table)) {
MARS_ERR("cannot allocate state subtable %d.\n", i);
_free_pages(brick);
return -ENOMEM;
}
memset(sub_table, 0, PAGE_SIZE);
}
2011-02-23 20:48:06 +00:00
init_waitqueue_head(&brick->event);
sema_init(&brick->mutex, 1);
return 0;
}
static int copy_brick_destruct(struct copy_brick *brick)
{
_free_pages(brick);
2011-02-23 20:48:06 +00:00
return 0;
}
static int copy_output_construct(struct copy_output *output)
{
return 0;
}
static int copy_output_destruct(struct copy_output *output)
{
return 0;
}
///////////////////////// static structs ////////////////////////
static struct copy_brick_ops copy_brick_ops = {
.brick_switch = copy_switch,
2011-06-17 11:32:38 +00:00
.brick_statistics = copy_statistics,
.reset_statistics = copy_reset_statistics,
2011-02-23 20:48:06 +00:00
};
static struct copy_output_ops copy_output_ops = {
.mars_get_info = copy_get_info,
.mref_get = copy_ref_get,
.mref_put = copy_ref_put,
.mref_io = copy_ref_io,
};
const struct copy_input_type copy_input_type = {
.type_name = "copy_input",
.input_size = sizeof(struct copy_input),
};
static const struct copy_input_type *copy_input_types[] = {
&copy_input_type,
&copy_input_type,
&copy_input_type,
&copy_input_type,
};
const struct copy_output_type copy_output_type = {
.type_name = "copy_output",
.output_size = sizeof(struct copy_output),
.master_ops = &copy_output_ops,
.output_construct = &copy_output_construct,
.output_destruct = &copy_output_destruct,
};
static const struct copy_output_type *copy_output_types[] = {
&copy_output_type,
};
const struct copy_brick_type copy_brick_type = {
.type_name = "copy_brick",
.brick_size = sizeof(struct copy_brick),
.max_inputs = 4,
.max_outputs = 1,
.master_ops = &copy_brick_ops,
.aspect_types = copy_aspect_types,
2011-02-23 20:48:06 +00:00
.default_input_types = copy_input_types,
.default_output_types = copy_output_types,
.brick_construct = &copy_brick_construct,
.brick_destruct = &copy_brick_destruct,
};
EXPORT_SYMBOL_GPL(copy_brick_type);
////////////////// module init stuff /////////////////////////
2011-08-25 10:16:32 +00:00
int __init init_mars_copy(void)
2011-02-23 20:48:06 +00:00
{
MARS_INF("init_copy()\n");
return copy_register_brick_type();
}
2011-08-25 10:16:32 +00:00
void __exit exit_mars_copy(void)
2011-02-23 20:48:06 +00:00
{
MARS_INF("exit_copy()\n");
copy_unregister_brick_type();
}
2011-08-25 10:16:32 +00:00
#ifndef CONFIG_MARS_HAVE_BIGMODULE
2011-02-23 20:48:06 +00:00
MODULE_DESCRIPTION("MARS copy brick");
MODULE_AUTHOR("Thomas Schoebel-Theuer <tst@1und1.de>");
MODULE_LICENSE("GPL");
2011-08-25 10:16:32 +00:00
module_init(init_mars_copy);
module_exit(exit_mars_copy);
#endif