/// Process the characters we receive as the user presses keys.
static void process_input(bool continuous_mode) {
bool first_char_seen = false;
double prev_tstamp = 0.0;
std::vector<wchar_t> bind_chars;
std::fwprintf(stderr, L"Press a key\n\n");
while (keep_running) {
char_event_t evt{0};
if (reader_test_and_clear_interrupted()) {
evt = char_event_t{shell_modes.c_cc[VINTR]};
} else {
evt = input_common_readch_timed(true);
}
if (!evt.is_char()) {
output_bind_command(bind_chars);
if (first_char_seen && !continuous_mode) {
return;
}
continue;
}
wchar_t wc = evt.get_char();
prev_tstamp = output_elapsed_time(prev_tstamp, first_char_seen);
add_char_to_bind_command(wc, bind_chars);
output_info_about_char(wc);
if (output_matching_key_name(wc)) {
output_bind_command(bind_chars);
}
if (should_exit(wc)) {
std::fwprintf(stderr, L"\nExiting at your request.\n");
break;
}
first_char_seen = true;
}
}
/// Process the characters we receive as the user presses keys.
void process_input(bool continuous_mode) {
bool first_char_seen = false;
while (true) {
wchar_t wc = input_common_readch(first_char_seen && !continuous_mode);
if (wc == WEOF) {
return;
}
if (wc > 255) {
printf("\nUnexpected wide character from input_common_readch(): %lld / 0x%llx\n",
(long long)wc, (long long)wc);
return;
}
long long int curr_tstamp, delta_tstamp;
timeval char_tstamp;
gettimeofday(&char_tstamp, NULL);
curr_tstamp = char_tstamp.tv_sec * 1000000 + char_tstamp.tv_usec;
delta_tstamp = curr_tstamp - prev_tstamp;
if (delta_tstamp >= 1000000) delta_tstamp = 999999;
if (delta_tstamp >= 200000 && continuous_mode) {
printf("\n");
printf("Type 'exit' or 'quit' to terminate this program.\n");
printf("\n");
}
prev_tstamp = curr_tstamp;
unsigned char c = wc;
if (c < 32) {
// Control characters.
if (ctrl_equivalents[c]) {
printf("%6lld usec dec: %3u hex: %2x char: %s (aka \\c%c)\n", delta_tstamp, c, c,
ctrl_equivalents[c], c + 64);
} else {
printf("%6lld usec dec: %3u hex: %2x char: \\c%c\n", delta_tstamp, c, c, c + 64);
}
} else if (c == 32) {
// The space character.
printf("%6lld usec dec: %3u hex: %2x char: <space>\n", delta_tstamp, c, c);
} else if (c == 127) {
// The "del" character.
printf("%6lld usec dec: %3u hex: %2x char: \\x7f (aka del)\n", delta_tstamp, c, c);
} else if (c >= 128) {
// Non-ASCII characters (i.e., those with bit 7 set).
printf("%6lld usec dec: %3u hex: %2x char: non-ASCII\n", delta_tstamp, c, c);
} else {
// ASCII characters that are not control characters.
printf("%6lld usec dec: %3u hex: %2x char: %c\n", delta_tstamp, c, c, c);
}
char *const name = key_name(c);
if (name) {
printf("FYI: Saw sequence for bind key name \"%s\"\n", name);
free(name);
}
if (should_exit(c)) {
printf("\nExiting at your request.\n");
break;
}
first_char_seen = true;
}
}
static int one_round_run(int round_no)
{
int ret = 0, fd = -1, j;
unsigned long i, chunk_no = 0;
struct write_unit wu;
MPI_Request request;
MPI_Status status;
/*
* Root rank creates working file in chunks.
*/
if (!rank) {
rank_printf("Prepare file of %lu bytes\n", file_size);
open_rw_flags |= O_DIRECT;
open_ro_flags |= O_DIRECT;
ret = prep_orig_file_in_chunks(workfile, file_size);
should_exit(ret);
}
MPI_Barrier_Sync();
if (!rank) {
fd = open_file(workfile, open_rw_flags);
should_exit(fd);
} else {
/*
* Verification at the very beginning doesn't do anything more
* than reading the file into pagecache on none-root nodes.
*/
open_rw_flags &= ~O_DIRECT;
open_ro_flags &= ~O_DIRECT;
ret = verify_file(1, NULL, remote_wus, workfile, file_size);
should_exit(fd);
}
MPI_Barrier_Sync();
/*
* Root ranks write chunks at random serially.
*/
for (i = 0; i < num_chunks; i++) {
MPI_Barrier_Sync();
/*
* Root rank generates random write unit, then sends it to
* rest of ranks in-memoery after O_DIRECT write into file.
*/
if (!rank) {
chunk_no = get_rand_ul(0, num_chunks - 1);
prep_rand_dest_write_unit(&wu, chunk_no);
rank_printf("Write #%lu chunk with char(%c)\n",
chunk_no, wu.wu_char);
ret = do_write_chunk(fd, wu);
should_exit(ret);
memcpy(&remote_wus[wu.wu_chunk_no], &wu, sizeof(wu));
for (j = 1; j < size; j++) {
if (verbose)
rank_printf("Send write unit #%lu chunk "
"char(%c) to rank %d\n",
wu.wu_chunk_no,
wu.wu_char, j);
ret = MPI_Isend(&wu, sizeof(wu), MPI_BYTE, j,
1, MPI_COMM_WORLD, &request);
if (ret != MPI_SUCCESS)
abort_printf("MPI_Isend failed: %d\n",
ret);
MPI_Wait(&request, &status);
}
} else {
MPI_Irecv(&wu, sizeof(wu), MPI_BYTE, 0, 1,
MPI_COMM_WORLD, &request);
MPI_Wait(&request, &status);
if (verbose)
rank_printf("Receive write unit #%lu chunk "
"char(%c)\n", wu.wu_chunk_no, wu.wu_char);
if (wu.wu_timestamp >=
remote_wus[wu.wu_chunk_no].wu_timestamp)
memcpy(&remote_wus[wu.wu_chunk_no],
&wu, sizeof(wu));
}
MPI_Barrier_Sync();
if (rank) {
/*
* All none-root ranks need to verify if O_DIRECT writes
* from remote root node can be seen locally.
*/
rank_printf("Try to verify whole file in chunks.\n");
ret = verify_file(1, NULL, remote_wus, workfile, file_size);
should_exit(ret);
}
}
MPI_Barrier_Sync();
if (!rank)
if (fd > 0)
close(fd);
return ret;
}
static int comp_test(void)
{
int ret;
char dest[PATH_MAX];
unsigned long i;
root_printf("Test %d: Multi-nodes comprehensive test.\n", testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_comp_refile",
workplace);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
if (!rank) {
ret = prep_orig_file(orig_path, file_size, 1);
should_exit(ret);
ret = do_reflinks(orig_path, orig_path, ref_counts, 0);
should_exit(ret);
ret = reflink(orig_path, dest, 1);
should_exit(ret);
}
MPI_Barrier_Sync();
if (rank == 1) {
/*also doing reflinks and unlinks*/
printf(" *Test Rank %d: Doing reflinks,cows and unlink.\n",
rank);
ret = do_reflinks(dest, dest, ref_counts, 0);
should_exit(ret);
ret = do_cows_on_write(dest, ref_counts, file_size, HUNK_SIZE);
should_exit(ret);
ret = do_unlinks(dest, ref_counts);
should_exit(ret);
}
if (rank % 6 == 2) {
/*Write former reflinks to cause cow*/
printf(" *Test Rank %d: Doing cows.\n", rank);
ret = do_cows_on_write(orig_path, ref_counts, file_size,
HUNK_SIZE);
should_exit(ret);
}
if (rank % 6 == 3) {
/*Read former reflinks*/
printf(" *Test Rank %d: Doing reads.\n", rank);
ret = do_reads_on_reflinks(orig_path, ref_counts, file_size,
HUNK_SIZE);
should_exit(ret);
}
if (rank % 6 == 4) {
/*Append to former reflinks*/
printf(" *Test Rank %d: Doing appends.\n", rank);
ret = do_appends(orig_path, ref_counts);
should_exit(ret);
}
if (rank % 6 == 5) {
/*Truncate former reflinks*/
printf(" *Test Rank %d: Doing truncates.\n", rank);
ret = do_cows_on_ftruncate(orig_path, ref_counts, file_size);
should_exit(ret);
}
if (!rank) {
printf(" *Test Rank %d: Doing verifications.\n", rank);
for (i = 0; i < size; i++) {
ret = verify_orig_file(orig_path);
should_exit(ret);
sleep(1);
}
}
MPI_Barrier_Sync();
if (!rank) {
printf(" *Test Rank %d: Doing unlinks.\n", rank);
ret = verify_orig_file(orig_path);
should_exit(ret);
ret = do_unlinks(orig_path, ref_counts);
should_exit(ret);
ret = do_unlink(orig_path);
should_exit(ret);
ret = do_unlink(dest);
should_exit(ret);
}
MPI_Barrier_Sync();
return 0;
}
static int directio_test(void)
{
int ret, fd;
char dest[PATH_MAX];
int sub_testno = 1;
int o_flags_rw, o_flags_ro;
unsigned long write_size = 0, read_size = 0;
unsigned long append_size = 0, truncate_size = 0;
unsigned long interval, offset = 0;
unsigned long align_slice = 512;
unsigned long align_filesz = align_slice;
o_flags_rw = open_rw_flags;
o_flags_ro = open_ro_flags;
open_rw_flags |= O_DIRECT;
open_ro_flags |= O_DIRECT;
while (align_filesz < file_size)
align_filesz += align_slice;
root_printf("Test %d: Multi-nodes O_DIRECT test.\n", testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_directio_refile",
workplace);
root_printf(" *SubTest %d:Prepare original inode %s.\n",
sub_testno++, orig_path);
if (!rank) {
ret = prep_orig_file_dio(orig_path, align_filesz);
should_exit(ret);
}
MPI_Barrier_Sync();
/*
* All ranks try to reflink the original to increment the
* refcount concurrently.
*/
root_printf(" *SubTest %d:Reflinking inode %s among nodes.\n",
sub_testno++, orig_path);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
ret = reflink(orig_path, dest, 1);
should_exit(ret);
}
MPI_Barrier_Sync();
/*
* All ranks try to do cow to decrement the
* refcount concurrently.
*/
root_printf(" *SubTest %d:Cowing reflinks by O_DIRECT writes among"
" nodes.\n", sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
interval = DIRECTIO_SLICE;
offset = 0;
while (offset < align_filesz) {
write_size = DIRECTIO_SLICE;
if (offset + write_size > align_filesz)
write_size = align_filesz - offset;
get_rand_buf(dio_buf, write_size);
ret = write_at_file(dest, dio_buf, write_size, offset);
should_exit(ret);
offset += write_size + interval;
}
}
MPI_Barrier_Sync();
/*
* All ranks try to read reflinks concurrently
*/
root_printf(" *SubTest %d:O_DIRECT reading reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
interval = DIRECTIO_SLICE;
offset = 0;
while (offset < align_filesz) {
read_size = DIRECTIO_SLICE;
if (offset + read_size > align_filesz)
read_size = align_filesz - offset;
ret = read_at_file(dest, dio_buf, read_size, offset);
should_exit(ret);
offset = offset + read_size + interval;
}
}
MPI_Barrier_Sync();
/*
* All ranks try to append reflinks concurrently
*/
root_printf(" *SubTest %d:Appending reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
fd = open64(dest, open_rw_flags | O_APPEND);
if (fd < 0) {
fd = errno;
abort_printf("open file %s failed:%d:%s\n",
dest, fd, strerror(fd));
}
append_size = DIRECTIO_SLICE;
get_rand_buf(dio_buf, append_size);
ret = write(fd, dio_buf, append_size);
if (ret < 0) {
ret = errno;
abort_printf("write file %s failed:%d:%s\n",
dest, ret, strerror(ret));
}
close(fd);
}
MPI_Barrier_Sync();
/*
* All ranks try to truncate reflinks concurrently
*/
root_printf(" *SubTest %d:Truncating reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
truncate_size = get_rand(0, align_filesz / DIRECTIO_SLICE) *
DIRECTIO_SLICE;
ret = truncate(dest, truncate_size);
if (ret < 0) {
ret = errno;
abort_printf("truncate file %s failed:%d:%s\n",
dest, ret, strerror(ret));
}
}
MPI_Barrier_Sync();
if (rank) {
ret = do_unlink(dest);
should_exit(ret);
} else {
ret = do_unlink(orig_path);
should_exit(ret);
}
open_rw_flags = o_flags_rw;
open_ro_flags = o_flags_ro;
MPI_Barrier_Sync();
return 0;
}
static int basic_test(void)
{
int ret = 0, fd;
char dest[PATH_MAX];
int sub_testno = 1;
char *write_buf = NULL, *read_buf = NULL;
unsigned long write_size = 0, read_size = 0;
unsigned long append_size = 0, truncate_size = 0;
unsigned long interval, offset = 0;
write_buf = (char *)malloc(HUNK_SIZE * 2);
read_buf = (char *)malloc(HUNK_SIZE * 2);
root_printf("Test %d: Multi-nodes basic refcount test.\n", testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_basic_refile",
workplace);
root_printf(" *SubTest %d:Prepare original inode %s.\n",
sub_testno++, orig_path);
if (!rank) {
ret = prep_orig_file(orig_path, file_size, 1);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
/*
* All ranks try to reflink the original to increment the
* refcount concurrently.
*/
root_printf(" *SubTest %d:Reflinking inode %s among nodes.\n",
sub_testno++, orig_path);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
ret = reflink(orig_path, dest, 1);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
/*
* All ranks try to do cow to decrement the
* refcount concurrently.
*/
root_printf(" *SubTest %d:Cowing reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
interval = file_size / 100;
offset = 0;
while (offset < file_size) {
if (test_flags & RAND_TEST)
write_size = get_rand(1, M_SIZE * 2);
else
write_size = 1;
if (offset + write_size > file_size)
write_size = file_size - offset;
get_rand_buf(write_buf, write_size);
if (test_flags & MMAP_TEST)
ret = mmap_write_at_file(dest, write_buf,
write_size, offset);
else
ret = write_at_file(dest, write_buf, write_size,
offset);
if (ret)
goto bail_free;
if (test_flags & RAND_TEST)
offset += write_size + get_rand(1, interval);
else
offset += write_size + interval;
}
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
/*
* All ranks try to read reflinks concurrently
*/
root_printf(" *SubTest %d:Reading reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
interval = file_size / 100;
offset = 0;
while (offset < file_size) {
if (test_flags & RAND_TEST)
read_size = get_rand(1, M_SIZE * 2);
else
read_size = 1;
if (offset + read_size > file_size)
read_size = file_size - offset;
if (test_flags & MMAP_TEST)
ret = mmap_read_at_file(dest, read_buf,
read_size, offset);
else
ret = read_at_file(dest, read_buf, read_size,
offset);
if (ret)
goto bail_free;
if (test_flags & RAND_TEST)
offset = offset + read_size +
get_rand(1, interval);
else
offset = offset + read_size + interval;
}
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
if (test_flags & MMAP_TEST)
goto bail;
/*
* All ranks try to append reflinks concurrently
*/
root_printf(" *SubTest %d:Appending reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
fd = open64(dest, open_rw_flags | O_APPEND);
if (fd < 0) {
if (write_buf)
free(write_buf);
if (read_buf)
free(read_buf);
fd = errno;
abort_printf("open file %s failed:%d:%s\n",
dest, fd, strerror(fd));
}
if (test_flags & RAND_TEST)
append_size = get_rand(1, HUNK_SIZE);
else
append_size = HUNK_SIZE;
get_rand_buf(write_buf, append_size);
ret = write(fd, write_buf, append_size);
if (ret < 0) {
if (write_buf)
free(write_buf);
if (read_buf)
free(read_buf);
ret = errno;
abort_printf("write file %s failed:%d:%s\n",
dest, ret, strerror(ret));
}
close(fd);
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
/*
* All ranks try to truncate reflinks concurrently
*/
root_printf(" *SubTest %d:Truncating reflinks among nodes.\n",
sub_testno++);
if (rank) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path, hostname, rank);
if (test_flags & RAND_TEST)
truncate_size = get_rand(0, file_size);
else
truncate_size = file_size / (rank + 1);
ret = truncate(dest, truncate_size);
if (ret < 0) {
if (write_buf)
free(write_buf);
if (read_buf)
free(read_buf);
ret = errno;
abort_printf("truncate file %s failed:%d:%s\n",
dest, ret, strerror(ret));
}
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
bail:
if (rank) {
ret = do_unlink(dest);
if (ret)
goto bail_free;
} else {
ret = do_unlink(orig_path);
if (ret)
goto bail_free;
}
MPI_Barrier_Sync();
bail_free:
if (write_buf)
free(write_buf);
if (read_buf)
free(read_buf);
should_exit(ret);
return ret;
}
static int dest_test(void)
{
int ret;
char dest[PATH_MAX];
int sub_testno = 1;
/*
* Muti-nodes to reflink the original from a node forever,
* Then manually or automatically crash the target node
* where reflinking is performing, we next check the validation
* of original file and restore its correct state by fsck.
* For target nodes, the reflinked inode should not exist
* in orphan directory anyway!
*/
root_printf("Test %d: Multi-nodes destructive refcount test.\n",
testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_basic_refile",
workplace);
/*
* Note that we use somewhat large files with very separated extents
* to let the reflinking process relatively slow. which therefore make
* it easy for us to interrupt.
*/
if (!rank) {
ret = prep_orig_file(orig_path, file_size, 0);
should_exit(ret);
}
MPI_Barrier_Sync();
if (rank) {
while (1) {
snprintf(dest, PATH_MAX, "%s-%s-%d", orig_path,
hostname, rank);
printf(" *SubTest %d:Reflinking to inode %s on %s.\n",
sub_testno++, dest, hostname);
ret = reflink(orig_path, dest, 1);
should_exit(ret);
sleep(5);
printf(" *SubTest %d:Unlinking inode %s from %s.\n",
sub_testno++, dest, hostname);
ret = do_unlink(dest);
should_exit(ret);
}
}
MPI_Barrier_Sync();
if (!rank) {
ret = verify_orig_file(orig_path);
should_exit(ret);
ret = do_unlink(orig_path);
should_exit(ret);
}
return 0;
}
static int stress_test(void)
{
unsigned long i, j;
int ret = 0, sub_testno = 1;
char *write_buf = NULL, dest[PATH_MAX];
char tmp_dest[PATH_MAX], tmp_orig[PATH_MAX];
char *pattern_buf = NULL, *verify_buf = NULL;
unsigned long offset = 0, write_size = 0, interval = 0;
unsigned long verify_size = 0, verify_offset = 0;
write_buf = (char *)malloc(HUNK_SIZE * 2);
pattern_buf = (char *)malloc(HUNK_SIZE * 2);
verify_buf = (char *)malloc(HUNK_SIZE * 2);
root_printf("Test %d: Multi-nodes stress refcount test.\n", testno++);
root_printf(" *SubTest %d: Stress test with tremendous refcount "
"trees.\n", sub_testno++);
for (i = 0; i < ref_trees; i++) {
snprintf(orig_path, PATH_MAX, "%s/multi_original_stress_"
"refile_rank%d_%ld", workplace, rank, i);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
ret = prep_orig_file(orig_path, 32 * 1024, 1);
if (ret)
goto bail;
ret = reflink(orig_path, dest, 1);
if (ret)
goto bail;
}
for (i = 0; i < ref_trees; i++) {
snprintf(orig_path, PATH_MAX, "%s/multi_original_stress_refile_"
"rank%d_%ld", workplace, rank, i);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
offset = get_rand(0, 32 * 1024 - 1);
write_size = 1;
get_rand_buf(write_buf, write_size);
ret = write_at_file(dest, write_buf, write_size, offset);
if (ret)
goto bail;
}
for (i = 0; i < ref_trees; i++) {
snprintf(orig_path, PATH_MAX, "%s/multi_original_stress_refile_"
"rank%d_%ld", workplace, rank, i);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
ret = do_unlink(orig_path);
if (ret)
goto bail;
ret = do_unlink(dest);
if (ret)
goto bail;
}
MPI_Barrier_Sync();
root_printf(" *SubTest %d: Stress test with tremendous shared inodes "
"on one refcount tree.\n", sub_testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_stress_refile",
workplace);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
if (!rank) {
ret = prep_orig_file(orig_path, 10 * HUNK_SIZE, 1);
if (ret)
goto bail;
for (i = 1; i < size; i++) {
snprintf(ref_path, PATH_MAX, "%s_%ld", dest, i);
ret = reflink(orig_path, ref_path, 1);
if (ret)
goto bail;
}
}
MPI_Barrier_Sync();
if (rank) {
snprintf(ref_path, PATH_MAX, "%s_%d", dest, rank);
ret = do_reflinks(ref_path, ref_path, ref_counts, 1);
if (ret)
goto bail;
for (i = 0; i < ref_counts; i++) {
if (get_rand(0, 1))
continue;
snprintf(ref_path, PATH_MAX, "%s_%dr%ld", dest, rank,
i);
offset = get_rand(0, 10 * HUNK_SIZE - 1);
write_size = get_rand(1, HUNK_SIZE);
if (offset + write_size > 10 * HUNK_SIZE)
write_size = 10 * HUNK_SIZE - offset;
get_rand_buf(write_buf, write_size);
ret = write_at_file(ref_path, write_buf, write_size,
offset);
if (ret)
goto bail;
}
}
MPI_Barrier_Sync();
if (rank) {
for (i = 0; i < ref_counts; i++) {
if (get_rand(0, 1))
continue;
snprintf(ref_path, PATH_MAX, "%s_%dr%ld", orig_path,
rank, i);
truncate(ref_path, 0);
}
}
MPI_Barrier_Sync();
if (rank) {
snprintf(ref_path, PATH_MAX, "%s_%d", dest, rank);
ret = do_unlinks(ref_path, ref_counts);
if (ret)
goto bail;
} else {
for (i = 1; i < size; i++) {
snprintf(ref_path, PATH_MAX, "%s_%ld", dest, i);
ret = do_unlink(ref_path);
if (ret)
goto bail;
}
ret = do_unlink(orig_path);
if (ret)
goto bail;
}
MPI_Barrier_Sync();
root_printf(" *SubTest %d: Stress test with HUGEFILE reflinked.\n",
sub_testno++);
snprintf(orig_path, PATH_MAX, "%s/multi_original_stress_huge_refile",
workplace);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
strcpy(tmp_dest, dest);
strcpy(tmp_orig, orig_path);
if (!rank) {
ret = prep_orig_file(orig_path, file_size, 1);
if (ret)
goto bail;
ret = reflink(orig_path, dest, 1);
if (ret)
goto bail;
}
MPI_Barrier_Sync();
if (rank) {
offset = 0;
interval = file_size / 1000;
i = 0;
while (offset < file_size) {
snprintf(dest, PATH_MAX, "%s_%d_%ld", tmp_dest,
rank, i);
ret = reflink(orig_path, dest, 1);
if (ret)
goto bail;
write_size = get_rand(1, M_SIZE);
get_rand_buf(write_buf, write_size);
verify_size = get_rand(M_SIZE, 2 * M_SIZE);
if (offset < (verify_size - write_size) / 2)
verify_offset = 0;
else
verify_offset = offset -
(verify_size - write_size) / 2;
if (verify_offset + verify_size > file_size)
verify_size = file_size - verify_offset;
ret = read_at_file(orig_path, pattern_buf, verify_size,
verify_offset);
if (ret)
goto bail;
ret = write_at_file(dest, write_buf, write_size,
offset);
if (ret)
goto bail;
ret = read_at_file(orig_path, verify_buf, verify_size,
verify_offset);
if (ret)
goto bail;
if (memcmp(pattern_buf, verify_buf, verify_size)) {
abort_printf("Verify original file date failed"
" after writting to snapshot!\n");
ret = -1;
goto bail;
}
offset = offset + write_size + interval;
strcpy(orig_path, dest);
i++;
}
for (j = 0; j < i; j++) {
snprintf(dest, PATH_MAX, "%s_%d_%ld", tmp_dest,
rank, j);
ret = do_unlink(dest);
if (ret)
goto bail;
}
strcpy(dest, tmp_dest);
offset = 0;
interval = file_size / 1000;
while (offset < file_size) {
write_size = get_rand(1, M_SIZE * 2);
get_rand_buf(write_buf, write_size);
ret = write_at_file(dest, write_buf,
write_size, offset);
if (ret)
goto bail;
offset = offset + write_size + interval;
}
}
MPI_Barrier_Sync();
if (!rank) {
ret = do_unlink(dest);
if (ret)
goto bail;
ret = do_unlink(orig_path);
if (ret)
goto bail;
}
bail:
if (write_buf)
free(write_buf);
if (pattern_buf)
free(pattern_buf);
if (verify_buf)
free(verify_buf);
should_exit(ret);
return ret;
}
static int xattr_basic_test(int ea_name_size, int ea_value_size)
{
int ret, fd = -1;
int sub_testno = 1;
char dest[PATH_MAX];
MPI_Request request;
MPI_Status status;
unsigned long i, j;
xattr_name_sz = ea_name_size;
xattr_value_sz = ea_value_size;
snprintf(orig_path, PATH_MAX, "%s/multi_original_xattr_refile",
workplace);
snprintf(dest, PATH_MAX, "%s_target", orig_path);
root_printf(" *SubTest %d: Prep original inode.\n", sub_testno++);
if (!rank) {
ret = prep_orig_file(orig_path, file_size, 1);
should_exit(ret);
}
MPI_Barrier_Sync();
root_printf(" *SubTest %d: Prep %ld xattr name list among nodes.\n",
sub_testno++, xattr_nums);
for (i = 0; i < xattr_nums; i++) {
memset(xattr_name, 0, xattr_name_sz + 1);
memset(xattr_value, 0, xattr_value_sz);
memset(xattr_value_get, 0, xattr_value_sz);
if (!rank) {
xattr_name_generator(i, USER, xattr_name_sz,
xattr_name_sz);
strcpy(xattr_name_list_set[i], xattr_name);
for (j = 1; j < size; j++) {
ret = MPI_Isend(xattr_name, xattr_name_sz + 1,
MPI_BYTE, j, 1, MPI_COMM_WORLD,
&request);
if (ret != MPI_SUCCESS)
abort_printf("MPI_Isend failed: %d\n",
ret);
MPI_Wait(&request, &status);
}
} else {
ret = MPI_Irecv(xattr_name, xattr_name_sz + 1, MPI_BYTE,
0, 1, MPI_COMM_WORLD, &request);
if (ret != MPI_SUCCESS)
abort_printf("MPI_Irecv failed: %d\n", ret);
MPI_Wait(&request, &status);
strcpy(xattr_name_list_set[i], xattr_name);
}
}
MPI_Barrier_Sync();
root_printf(" *SubTest %d: Prep original inode with %ld EAs.\n",
sub_testno++, xattr_nums);
if (!rank) {
fd = open64(orig_path, open_rw_flags);
for (i = 0; i < xattr_nums; i++) {
strcpy(xattr_name, xattr_name_list_set[i]);
xattr_value_constructor(i);
ret = add_or_update_ea(NORMAL, fd, XATTR_CREATE, "add");
should_exit(ret);
ret = read_ea(NORMAL, fd);
should_exit(ret);
ret = xattr_value_validator(i);
should_exit(ret);
}
ret = do_reflinks(orig_path, orig_path, ref_counts, 0);
should_exit(ret);
ret = reflink(orig_path, dest, 1);
should_exit(ret);
}
MPI_Barrier_Sync();
if (rank % 6 == 1) {
/*also doing reflinks and unlinks*/
printf(" *SubTest Rank %d: Do reflinks and cows on %ld EAs.\n",
rank, xattr_nums);
ret = do_reflinks(dest, dest, ref_counts, 0);
should_exit(ret);
ret = do_xattr_cows(dest, ref_counts, xattr_nums);
should_exit(ret);
ret = do_unlinks(dest, ref_counts);
should_exit(ret);
}
if (rank % 6 == 2) {
printf(" *SubTest Rank %d: Do cows on %ld EAs.\n", rank,
xattr_nums);
ret = do_xattr_cows(orig_path, ref_counts, xattr_nums);
should_exit(ret);
}
if (rank % 6 == 3) {
printf(" *SubTest Rank %d: Do data&ea cows on %ld EAs.\n",
rank, xattr_nums);
ret = do_xattr_data_cows(orig_path, ref_counts, xattr_nums);
should_exit(ret);
}
if (rank % 6 == 4) {
printf(" *SubTest Rank %d: Do reads on %ld EAs.\n", rank,
xattr_nums);
xattr_value_sz = XATTR_VALUE_MAX_SZ;
ret = do_xattr_reads(orig_path, ref_counts, xattr_nums);
should_exit(ret);
}
if (rank % 6 == 5) {
printf(" *SubTest Rank %d: Do lists on %ld EAs.\n", rank,
xattr_nums);
if (list_sz < XATTR_LIST_MAX_SZ) {
ret = do_xattr_lists(orig_path, ref_counts);
should_exit(ret);
}
}
MPI_Barrier_Sync();
if (!rank) {
printf(" *SubTest Rank %d: Do EA removal.\n", rank);
ret = do_xattr_removes(orig_path, ref_counts, xattr_nums);
should_exit(ret);
}
MPI_Barrier_Sync();
if (!rank) {
close(fd);
ret = do_unlinks(orig_path, ref_counts);
should_exit(ret);
ret = do_unlink(dest);
should_exit(ret);
ret = do_unlink(orig_path);
should_exit(ret);
}
return 0;
}
static void *am_log_worker(void *arg) {
struct am_log *log = AM_LOG();
int i, level, is_audit;
unsigned int index;
char *data;
size_t data_sz;
unsigned long instance_id;
struct stat st;
struct log_files *f;
if (log == NULL) {
return NULL;
}
#ifdef _WIN32
WaitForSingleObject(am_log_lck.lock, INFINITE);
#else
pthread_mutex_lock(&log->lock);
#endif
for (;;) {
index = log->out;
#ifdef _WIN32
while (log->read_count == 0 || !log->bucket[index].ready_to_read) {
ReleaseMutex(am_log_lck.lock);
if (WaitForSingleObject(am_log_lck.new_data_cond, 1000) == WAIT_TIMEOUT) {
if (WaitForSingleObject(am_log_lck.exit, 0) == WAIT_OBJECT_0) {
return NULL;
}
}
WaitForSingleObject(am_log_lck.lock, INFINITE);
}
#else
while (log->read_count == 0 || !log->bucket[index].ready_to_read) {
struct timeval now = {0, 0};
struct timespec ts = {0, 0};
gettimeofday(&now, NULL);
ts.tv_sec = now.tv_sec + 1;
ts.tv_nsec = now.tv_usec * 1000;
if (pthread_cond_timedwait(&log->new_data_cond, &log->lock, &ts) == ETIMEDOUT) {
if (should_exit(&log->exit)) {
pthread_mutex_unlock(&log->lock);
return NULL;
}
}
}
#endif /* _WIN32 */
log->bucket[index].ready_to_read = AM_FALSE;
#ifdef _WIN32
ReleaseMutex(am_log_lck.lock);
#else
pthread_mutex_unlock(&log->lock);
#endif /* _WIN32 */
data = log->bucket[index].data;
data_sz = log->bucket[index].size;
level = log->bucket[index].level;
is_audit = (level & AM_LOG_LEVEL_AUDIT) != 0;
instance_id = log->bucket[index].instance_id;
f = NULL;
for (i = 0; i < AM_MAX_INSTANCES; i++) {
f = &log->files[i];
if (f->used && f->instance_id == instance_id) {
break;
}
}
if (f != NULL) {
if (ISINVALID(f->name_debug)) {
fprintf(stderr, "am_log_worker(): the debug file name is invalid (i.e. empty or null)\n");
f->fd_debug = -1;
f->fd_audit = -1;
return NULL;
}
if (ISINVALID(f->name_audit)) {
fprintf(stderr, "am_log_worker(): the audit file name is invalid (i.e. empty or null)\n");
f->fd_debug = -1;
f->fd_audit = -1;
return NULL;
}
/* log files are not opened yet, do it now */
if (f->fd_audit == -1 && f->fd_debug == -1) {
#ifdef _WIN32
f->fd_debug = _open(f->name_debug, _O_CREAT | _O_WRONLY | _O_APPEND | _O_BINARY,
_S_IREAD | _S_IWRITE);
f->fd_audit = _open(f->name_audit, _O_CREAT | _O_WRONLY | _O_APPEND | _O_BINARY,
_S_IREAD | _S_IWRITE);
if (f->fd_debug != -1 && stat(f->name_debug, &st) == 0) {
f->created_debug = st.st_ctime;
f->owner = getpid();
}
if (f->fd_audit != -1 && stat(f->name_audit, &st) == 0) {
f->created_audit = st.st_ctime;
f->owner = getpid();
}
#else
f->fd_debug = open(f->name_debug, O_CREAT | O_WRONLY | O_APPEND, S_IWUSR | S_IRUSR);
f->fd_audit = open(f->name_audit, O_CREAT | O_WRONLY | O_APPEND, S_IWUSR | S_IRUSR);
if (f->fd_debug != -1 && stat(f->name_debug, &st) == 0) {
f->node_debug = st.st_ino;
f->created_debug = st.st_ctime;
f->owner = getpid();
}
if (f->fd_audit != -1 && stat(f->name_audit, &st) == 0) {
f->node_audit = st.st_ino;
f->created_audit = st.st_ctime;
f->owner = getpid();
}
#endif
}
if (f->fd_debug == -1) {
fprintf(stderr, "am_log_worker() failed to open log file %s: error: %d", f->name_debug, errno);
f->fd_debug = f->fd_audit = -1;
} else if (f->fd_audit == -1) {
fprintf(stderr, "am_log_worker() failed to open audit file %s: error: %d", f->name_audit, errno);
f->fd_debug = f->fd_audit = -1;
} else {
int file_handle = is_audit ? f->fd_audit : f->fd_debug;
char *file_name = is_audit ? f->name_audit : f->name_debug;
int max_size = is_audit ? f->max_size_audit : f->max_size_debug;
time_t file_created = is_audit ? f->created_audit : f->created_debug;
#ifdef _WIN32
int wrote;
#else
ssize_t wrote;
ino_t file_inode = is_audit ? f->node_audit : f->node_debug;
#endif
wrote = write(file_handle, data, (unsigned int) data_sz);
#ifdef _WIN32
wrote = write(file_handle, "\r\n", 2);
_commit(file_handle);
/* check file timestamp; rotate by date if set so */
if (max_size == -1 && should_rotate_time(file_created)) {
HANDLE fh = (HANDLE) _get_osfhandle(file_handle);
unsigned int idx = 1;
static char tmp[AM_PATH_SIZE];
do {
snprintf(tmp, sizeof (tmp), "%s.%d", file_name, idx);
idx++;
} while (_access(tmp, 0) == 0);
if (CopyFileA(file_name, tmp, FALSE)) {
SetFilePointer(fh, 0, NULL, FILE_BEGIN);
SetEndOfFile(fh);
if (is_audit) {
f->created_audit = time(NULL);
} else {
f->created_debug = time(NULL);
}
} else {
fprintf(stderr, "could not rotate log file %s (error: %d)\n",
file_name, GetLastError());
}
}
/* check file size; rotate by size if set so */
if (max_size > 0) {
BY_HANDLE_FILE_INFORMATION info;
uint64_t fsz = 0;
HANDLE fh = (HANDLE) _get_osfhandle(file_handle);
if (GetFileInformationByHandle(fh, &info)) {
fsz = ((DWORDLONG) (((DWORD) (info.nFileSizeLow)) |
(((DWORDLONG) ((DWORD) (info.nFileSizeHigh))) << 32)));
}
if ((fsz + 1024) > max_size) {
unsigned int idx = 1;
static char tmp[AM_PATH_SIZE];
do {
snprintf(tmp, sizeof (tmp), "%s.%d", file_name, idx);
idx++;
} while (_access(tmp, 0) == 0);
if (CopyFileA(file_name, tmp, FALSE)) {
SetFilePointer(fh, 0, NULL, FILE_BEGIN);
SetEndOfFile(fh);
if (is_audit) {
f->created_audit = time(NULL);
} else {
f->created_debug = time(NULL);
}
} else {
fprintf(stderr, "could not rotate log file %s (error: %d)\n",
file_name, GetLastError());
}
}
}
#else
wrote = write(file_handle, "\n", 1);
fsync(file_handle);
/* check file timestamp; rotate by date if set so */
if (max_size == -1 && should_rotate_time(file_created)) {
rename_file(file_name);
}
/* check file size; rotate by size if set so */
if (max_size > 0 && stat(file_name, &st) == 0 && (st.st_size + 1024) > max_size) {
rename_file(file_name);
}
/* reset file inode number (in case it has changed as a result of rename_file) */
if (stat(file_name, &st) != 0 || st.st_ino != file_inode) {
close(file_handle);
if (is_audit) {
f->fd_audit = open(f->name_audit, O_CREAT | O_WRONLY | O_APPEND, S_IWUSR | S_IRUSR);
f->node_audit = st.st_ino;
f->created_audit = st.st_ctime;
f->owner = getpid();
} else {
f->fd_debug = open(f->name_debug, O_CREAT | O_WRONLY | O_APPEND, S_IWUSR | S_IRUSR);
f->node_debug = st.st_ino;
f->created_debug = st.st_ctime;
f->owner = getpid();
}
if (f->fd_debug == -1 || f->fd_audit == -1) {
fprintf(stderr, "am_log_worker() log file re-open failed with error: %d", errno);
f->fd_debug = f->fd_audit = -1;
}
}
#endif
}
}
log->out = AM_LOG_BUFFER_MASK(log->out + 1, log->bucket_count);
#ifdef _WIN32
WaitForSingleObject(am_log_lck.lock, INFINITE);
#else
pthread_mutex_lock(&log->lock);
#endif
log->read_count--;
#ifdef _WIN32
SetEvent(am_log_lck.new_space_cond);
#else
pthread_cond_broadcast(&log->new_space_cond);
#endif
}
return NULL;
}
