static int verify_handle(char *test, struct llog_handle *llh, int num_recs)
{
int i;
int last_idx = 0;
int active_recs = 0;
for (i = 0; i < LLOG_BITMAP_BYTES * 8; i++) {
if (ext2_test_bit(i, llh->lgh_hdr->llh_bitmap)) {
last_idx = i;
active_recs++;
}
}
if (active_recs != num_recs) {
CERROR("%s: expected %d active recs after write, found %d\n",
test, num_recs, active_recs);
return -ERANGE;
}
if (llh->lgh_hdr->llh_count != num_recs) {
CERROR("%s: handle->count is %d, expected %d after write\n",
test, llh->lgh_hdr->llh_count, num_recs);
return -ERANGE;
}
if (llh->lgh_last_idx < last_idx) {
CERROR("%s: handle->last_idx is %d, expected %d after write\n",
test, llh->lgh_last_idx, last_idx);
return -ERANGE;
}
return 0;
}
static int llog_catinfo_cb(struct llog_handle *cat,
struct llog_rec_hdr *rec, void *data)
{
static char *out = NULL;
static int remains = 0;
struct llog_ctxt *ctxt = NULL;
struct llog_handle *handle = NULL;
struct llog_logid *logid;
struct llog_logid_rec *lir;
int l, rc, index, count = 0;
struct cb_data *cbd = (struct cb_data*)data;
ENTRY;
if (cbd->init) {
out = cbd->out;
remains = cbd->remains;
cbd->init = 0;
}
if (!(cat->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
RETURN(-EINVAL);
if (!cbd->ctxt)
RETURN(-ENODEV);
lir = (struct llog_logid_rec *)rec;
logid = &lir->lid_id;
rc = llog_create(ctxt, &handle, logid, NULL);
if (rc)
RETURN(-EINVAL);
rc = llog_init_handle(handle, 0, NULL);
if (rc)
GOTO(out_close, rc);
for (index = 1; index < (LLOG_BITMAP_BYTES * 8); index++) {
if (ext2_test_bit(index, handle->lgh_hdr->llh_bitmap))
count++;
}
l = snprintf(out, remains, "\t[Log ID]: #"LPX64"#"LPX64"#%08x\n"
"\tLog Size: %llu\n\tLast Index: %d\n"
"\tUncanceled Records: %d\n",
logid->lgl_oid, logid->lgl_oseq, logid->lgl_ogen,
i_size_read(handle->lgh_file->f_dentry->d_inode),
handle->lgh_last_idx, count);
out += l;
remains -= l;
cbd->out = out;
cbd->remains = remains;
if (remains <= 0) {
CWARN("Not enough memory\n");
rc = -ENOMEM;
}
EXIT;
out_close:
llog_close(handle);
return rc;
}
static int llog_catinfo_config(struct obd_device *obd, char *buf, int buf_len,
char *client)
{
struct mds_obd *mds = &obd->u.mds;
struct llog_ctxt *ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
struct lvfs_run_ctxt saved;
struct llog_handle *handle = NULL;
char name[4][64];
int rc, i, l, remains = buf_len;
char *out = buf;
ENTRY;
if (ctxt == NULL || mds == NULL)
GOTO(release_ctxt, rc = -ENODEV);
push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
sprintf(name[0], "%s", mds->mds_profile);
sprintf(name[1], "%s-clean", mds->mds_profile);
sprintf(name[2], "%s", client);
sprintf(name[3], "%s-clean", client);
for (i = 0; i < 4; i++) {
int index, uncanceled = 0;
rc = llog_create(ctxt, &handle, NULL, name[i]);
if (rc)
GOTO(out_pop, rc);
rc = llog_init_handle(handle, 0, NULL);
if (rc) {
llog_close(handle);
GOTO(out_pop, rc = -ENOENT);
}
for (index = 1; index < (LLOG_BITMAP_BYTES * 8); index ++) {
if (ext2_test_bit(index, handle->lgh_hdr->llh_bitmap))
uncanceled++;
}
l = snprintf(out, remains, "[Log Name]: %s\nLog Size: %llu\n"
"Last Index: %d\nUncanceled Records: %d\n\n",
name[i],
i_size_read(handle->lgh_file->f_dentry->d_inode),
handle->lgh_last_idx, uncanceled);
out += l;
remains -= l;
llog_close(handle);
if (remains <= 0)
break;
}
GOTO(out_pop, rc);
out_pop:
pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
release_ctxt:
llog_ctxt_put(ctxt);
return rc;
}
/*
* get the specified goal block
* if alread alloced find it by looking forward
*/
int ext2_grab_block(char *bitmap, unsigned int goal)
{
if ( !ext2_test_bit(bitmap, goal) )
goto got_block;
/* else looking forward */
goal = find_first_zero (bitmap, goal + 1, EXT2_BLOCKS_PER_GROUP);
got_block:
return goal;
}
static int ext2_valid_block_bitmap(struct super_block *sb,
struct ext2_group_desc *desc,
unsigned int block_group,
struct buffer_head *bh)
{
ext2_grpblk_t offset;
ext2_grpblk_t next_zero_bit;
ext2_fsblk_t bitmap_blk;
ext2_fsblk_t group_first_block;
group_first_block = ext2_group_first_block_no(sb, block_group);
/* check whether block bitmap block number is set */
bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
offset = bitmap_blk - group_first_block;
if (!ext2_test_bit(offset, bh->b_data))
/* bad block bitmap */
goto err_out;
/* check whether the inode bitmap block number is set */
bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
offset = bitmap_blk - group_first_block;
if (!ext2_test_bit(offset, bh->b_data))
/* bad block bitmap */
goto err_out;
/* check whether the inode table block number is set */
bitmap_blk = le32_to_cpu(desc->bg_inode_table);
offset = bitmap_blk - group_first_block;
next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
offset + EXT2_SB(sb)->s_itb_per_group,
offset);
if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
/* good bitmap for inode tables */
return 1;
err_out:
ext2_error(sb, __func__,
"Invalid block bitmap - "
"block_group = %d, block = %lu",
block_group, bitmap_blk);
return 0;
}
int llog_reverse_process(const struct lu_env *env,
struct llog_handle *loghandle, llog_cb_t cb,
void *data, void *catdata)
{
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = catdata;
void *buf;
int rc = 0, first_index = 1, index, idx;
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf)
return -ENOMEM;
if (cd != NULL)
first_index = cd->lpcd_first_idx + 1;
if (cd != NULL && cd->lpcd_last_idx)
index = cd->lpcd_last_idx;
else
index = LLOG_BITMAP_BYTES * 8 - 1;
while (rc == 0) {
struct llog_rec_hdr *rec;
struct llog_rec_tail *tail;
/* skip records not set in bitmap */
while (index >= first_index &&
!ext2_test_bit(index, llh->llh_bitmap))
--index;
LASSERT(index >= first_index - 1);
if (index == first_index - 1)
break;
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, LLOG_CHUNK_SIZE);
rc = llog_prev_block(env, loghandle, index, buf,
LLOG_CHUNK_SIZE);
if (rc)
GOTO(out, rc);
rec = buf;
idx = rec->lrh_index;
CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
while (idx < index) {
rec = (void *)rec + rec->lrh_len;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
idx ++;
}
LASSERT(idx == index);
tail = (void *)rec + rec->lrh_len - sizeof(*tail);
/* process records in buffer, starting where we found one */
while ((void *)tail > buf) {
if (tail->lrt_index == 0)
GOTO(out, rc = 0); /* no more records */
/* if set, process the callback on this record */
if (ext2_test_bit(index, llh->llh_bitmap)) {
rec = (void *)tail - tail->lrt_len +
sizeof(*tail);
rc = cb(env, loghandle, rec, data);
if (rc == LLOG_PROC_BREAK) {
GOTO(out, rc);
} else if (rc == LLOG_DEL_RECORD) {
llog_cancel_rec(env, loghandle,
tail->lrt_index);
rc = 0;
}
if (rc)
GOTO(out, rc);
}
/* previous record, still in buffer? */
--index;
if (index < first_index)
GOTO(out, rc = 0);
tail = (void *)tail - tail->lrt_len;
}
}
out:
if (buf)
OBD_FREE(buf, LLOG_CHUNK_SIZE);
return rc;
}
static int llog_process_thread(void *arg)
{
struct llog_process_info *lpi = arg;
struct llog_handle *loghandle = lpi->lpi_loghandle;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = lpi->lpi_catdata;
char *buf;
__u64 cur_offset = LLOG_CHUNK_SIZE;
__u64 last_offset;
int rc = 0, index = 1, last_index;
int saved_index = 0;
int last_called_index = 0;
LASSERT(llh);
OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
if (!buf) {
lpi->lpi_rc = -ENOMEM;
return 0;
}
if (cd != NULL) {
last_called_index = cd->lpcd_first_idx;
index = cd->lpcd_first_idx + 1;
}
if (cd != NULL && cd->lpcd_last_idx)
last_index = cd->lpcd_last_idx;
else
last_index = LLOG_BITMAP_BYTES * 8 - 1;
while (rc == 0) {
struct llog_rec_hdr *rec;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, llh->llh_bitmap))
++index;
LASSERT(index <= last_index + 1);
if (index == last_index + 1)
break;
repeat:
CDEBUG(D_OTHER, "index: %d last_index %d\n",
index, last_index);
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, LLOG_CHUNK_SIZE);
last_offset = cur_offset;
rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
index, &cur_offset, buf, LLOG_CHUNK_SIZE);
if (rc)
GOTO(out, rc);
/* NB: when rec->lrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop. */
for (rec = (struct llog_rec_hdr *)buf;
(char *)rec < buf + LLOG_CHUNK_SIZE;
rec = (struct llog_rec_hdr *)((char *)rec + rec->lrh_len)){
CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
rec, rec->lrh_type);
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
rec->lrh_type, rec->lrh_index);
if (rec->lrh_index == 0) {
/* probably another rec just got added? */
if (index <= loghandle->lgh_last_idx)
GOTO(repeat, rc = 0);
GOTO(out, rc = 0); /* no more records */
}
if (rec->lrh_len == 0 ||
rec->lrh_len > LLOG_CHUNK_SIZE) {
CWARN("invalid length %d in llog record for "
"index %d/%d\n", rec->lrh_len,
rec->lrh_index, index);
GOTO(out, rc = -EINVAL);
}
if (rec->lrh_index < index) {
CDEBUG(D_OTHER, "skipping lrh_index %d\n",
rec->lrh_index);
continue;
}
CDEBUG(D_OTHER,
"lrh_index: %d lrh_len: %d (%d remains)\n",
rec->lrh_index, rec->lrh_len,
(int)(buf + LLOG_CHUNK_SIZE - (char *)rec));
loghandle->lgh_cur_idx = rec->lrh_index;
loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
last_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, llh->llh_bitmap)) {
rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
lpi->lpi_cbdata);
last_called_index = index;
if (rc == LLOG_PROC_BREAK) {
GOTO(out, rc);
} else if (rc == LLOG_DEL_RECORD) {
llog_cancel_rec(lpi->lpi_env,
loghandle,
rec->lrh_index);
rc = 0;
}
if (rc)
GOTO(out, rc);
} else {
CDEBUG(D_OTHER, "Skipped index %d\n", index);
}
/* next record, still in buffer? */
++index;
if (index > last_index)
GOTO(out, rc = 0);
}
}
out:
if (cd != NULL)
cd->lpcd_last_idx = last_called_index;
OBD_FREE(buf, LLOG_CHUNK_SIZE);
lpi->lpi_rc = rc;
return 0;
}
/*
* The touched member needs to be updated every time we access
* one of the bitsets.
*/
static inline int log_test_bit(uint32_t *bs, unsigned bit)
{
return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
}
int llog_pack_buffer(int fd, struct llog_log_hdr **llog,
struct llog_rec_hdr ***recs,
int *recs_number)
{
int rc = 0, recs_num,rd;
off_t file_size;
struct stat st;
char *file_buf=NULL, *recs_buf=NULL;
struct llog_rec_hdr **recs_pr=NULL;
char *ptr=NULL;
int i;
rc = fstat(fd,&st);
if (rc < 0){
printf("Get file stat error.\n");
goto out;
}
file_size = st.st_size;
file_buf = malloc(file_size);
if (file_buf == NULL){
printf("Memory Alloc for file_buf error.\n");
rc = -ENOMEM;
goto out;
}
*llog = (struct llog_log_hdr*)file_buf;
rd = read(fd,file_buf,file_size);
if (rd < file_size){
printf("Read file error.\n");
rc = -EIO; /*FIXME*/
goto clear_file_buf;
}
/* the llog header not countable here.*/
recs_num = le32_to_cpu((*llog)->llh_count)-1;
recs_buf = malloc(recs_num * sizeof(struct llog_rec_hdr *));
if (recs_buf == NULL){
printf("Memory Alloc for recs_buf error.\n");
rc = -ENOMEM;
goto clear_file_buf;
}
recs_pr = (struct llog_rec_hdr **)recs_buf;
ptr = file_buf + le32_to_cpu((*llog)->llh_hdr.lrh_len);
i = 0;
while (i < recs_num){
struct llog_rec_hdr *cur_rec = (struct llog_rec_hdr*)ptr;
int idx = le32_to_cpu(cur_rec->lrh_index);
recs_pr[i] = cur_rec;
if (ext2_test_bit(idx, (*llog)->llh_bitmap)) {
if (le32_to_cpu(cur_rec->lrh_type) != OBD_CFG_REC)
printf("rec #%d type=%x len=%u\n", idx,
cur_rec->lrh_type, cur_rec->lrh_len);
} else {
printf("Bit %d of %d not set\n", idx, recs_num);
cur_rec->lrh_id = CANCELLED;
/* The header counts only set records */
i--;
}
ptr += le32_to_cpu(cur_rec->lrh_len);
if ((ptr - file_buf) > file_size) {
printf("The log is corrupt (too big at %d)\n", i);
rc = -EINVAL;
goto clear_recs_buf;
}
i++;
}
*recs = recs_pr;
*recs_number = recs_num;
out:
return rc;
clear_recs_buf:
free(recs_buf);
clear_file_buf:
free(file_buf);
*llog=NULL;
goto out;
}
int mlog_reverse_process(struct mlog_handle *loghandle, mlog_cb_t cb,
void *data, void *catdata)
{
struct mlog_log_hdr *mlh = loghandle->mgh_hdr;
struct mlog_process_cat_data *cd = catdata;
void *buf = NULL;
int ret = 0, first_index = 1, index, idx;
MENTRY();
MTFS_ALLOC(buf, MLOG_CHUNK_SIZE);
if (buf == NULL) {
ret = -ENOMEM;
goto out;
}
if (cd != NULL)
first_index = cd->mpcd_first_idx + 1;
if (cd != NULL && cd->mpcd_last_idx)
index = cd->mpcd_last_idx;
else
index = MLOG_BITMAP_BYTES * 8 - 1;
while (ret == 0) {
struct mlog_rec_hdr *rec;
struct mlog_rec_tail *tail;
/* skip records not set in bitmap */
while (index >= first_index &&
!ext2_test_bit(index, mlh->mlh_bitmap))
--index;
MASSERT(index >= first_index - 1);
if (index == first_index - 1)
break;
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, MLOG_CHUNK_SIZE);
ret = mlog_prev_block(loghandle, index, buf, MLOG_CHUNK_SIZE);
if (ret) {
goto out_free;
}
rec = buf;
idx = le32_to_cpu(rec->mrh_index);
if (idx < index)
MDEBUG("index %u : idx %u\n", index, idx);
while (idx < index) {
rec = ((void *)rec + le32_to_cpu(rec->mrh_len));
idx ++;
}
tail = (void *)rec + le32_to_cpu(rec->mrh_len) - sizeof(*tail);
/* process records in buffer, starting where we found one */
while ((void *)tail > buf) {
rec = (void *)tail - le32_to_cpu(tail->mrt_len) +
sizeof(*tail);
if (rec->mrh_index == 0) {
/* no more records */
ret = 0;
goto out_free;
}
/* if set, process the callback on this record */
if (ext2_test_bit(index, mlh->mlh_bitmap)) {
ret = cb(loghandle, rec, data);
if (ret == MLOG_PROC_BREAK) {
MPRINT("recovery from log: %llx:%x"
" stopped\n",
loghandle->mgh_id.mgl_oid,
loghandle->mgh_id.mgl_ogen);
goto out_free;
}
if (ret) {
goto out_free;
}
}
/* previous record, still in buffer? */
--index;
if (index < first_index) {
ret = 0;
goto out_free;
}
tail = (void *)rec - sizeof(*tail);
}
}
out_free:
MTFS_FREE(buf, MLOG_CHUNK_SIZE);
out:
MRETURN(ret);
}
/* * bitmap_init_from_disk -- called at bitmap_create time to initialize
* the in-memory bitmap from the on-disk bitmap -- also, sets up the
* memory mapping of the bitmap file
* Special cases:
* if there's no bitmap file, or if the bitmap file had been
* previously kicked from the array, we mark all the bits as
* 1's in order to cause a full resync.
*
* We ignore all bits for sectors that end earlier than 'start'.
* This is used when reading an out-of-date bitmap...
*/
static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
{
unsigned long i, chunks, index, oldindex, bit;
struct page *page = NULL, *oldpage = NULL;
unsigned long num_pages, bit_cnt = 0;
struct file *file;
unsigned long bytes, offset;
int outofdate;
int ret = -ENOSPC;
void *paddr;
chunks = bitmap->chunks;
file = bitmap->file;
BUG_ON(!file && !bitmap->offset);
#ifdef INJECT_FAULTS_3
outofdate = 1;
#else
outofdate = bitmap->flags & BITMAP_STALE;
#endif
if (outofdate)
printk(KERN_INFO "%s: bitmap file is out of date, doing full "
"recovery\n", bmname(bitmap));
bytes = (chunks + 7) / 8;
num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE;
if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) {
printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
bmname(bitmap),
(unsigned long) i_size_read(file->f_mapping->host),
bytes + sizeof(bitmap_super_t));
goto out;
}
ret = -ENOMEM;
bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
if (!bitmap->filemap)
goto out;
/* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */
bitmap->filemap_attr = kzalloc(
roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
GFP_KERNEL);
if (!bitmap->filemap_attr)
goto out;
oldindex = ~0L;
for (i = 0; i < chunks; i++) {
int b;
index = file_page_index(i);
bit = file_page_offset(i);
if (index != oldindex) { /* this is a new page, read it in */
int count;
/* unmap the old page, we're done with it */
if (index == num_pages-1)
count = bytes + sizeof(bitmap_super_t)
- index * PAGE_SIZE;
else
count = PAGE_SIZE;
if (index == 0) {
/*
* if we're here then the superblock page
* contains some bits (PAGE_SIZE != sizeof sb)
* we've already read it in, so just use it
*/
page = bitmap->sb_page;
offset = sizeof(bitmap_super_t);
} else if (file) {
page = read_page(file, index, bitmap, count);
offset = 0;
} else {
page = read_sb_page(bitmap->mddev, bitmap->offset, index);
offset = 0;
}
if (IS_ERR(page)) { /* read error */
ret = PTR_ERR(page);
goto out;
}
oldindex = index;
oldpage = page;
if (outofdate) {
/*
* if bitmap is out of date, dirty the
* whole page and write it out
*/
paddr = kmap_atomic(page, KM_USER0);
memset(paddr + offset, 0xff,
PAGE_SIZE - offset);
kunmap_atomic(paddr, KM_USER0);
ret = write_page(bitmap, page, 1);
if (ret) {
/* release, page not in filemap yet */
put_page(page);
goto out;
}
}
bitmap->filemap[bitmap->file_pages++] = page;
bitmap->last_page_size = count;
}
paddr = kmap_atomic(page, KM_USER0);
if (bitmap->flags & BITMAP_HOSTENDIAN)
b = test_bit(bit, paddr);
else
b = ext2_test_bit(bit, paddr);
kunmap_atomic(paddr, KM_USER0);
if (b) {
/* if the disk bit is set, set the memory bit */
bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap),
((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start)
);
bit_cnt++;
set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN);
}
}
/* everything went OK */
ret = 0;
bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET);
if (bit_cnt) { /* Kick recovery if any bits were set */
set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
md_wakeup_thread(bitmap->mddev->thread);
}
out:
printk(KERN_INFO "%s: bitmap initialized from disk: "
"read %lu/%lu pages, set %lu bits, status: %d\n",
bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret);
return ret;
}
static int llog_process_thread(void *arg)
{
struct llog_process_info *lpi = arg;
struct llog_handle *loghandle = lpi->lpi_loghandle;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = lpi->lpi_catdata;
char *buf;
u64 cur_offset, tmp_offset;
int chunk_size;
int rc = 0, index = 1, last_index;
int saved_index = 0;
int last_called_index = 0;
if (!llh)
return -EINVAL;
cur_offset = llh->llh_hdr.lrh_len;
chunk_size = llh->llh_hdr.lrh_len;
/* expect chunk_size to be power of two */
LASSERT(is_power_of_2(chunk_size));
buf = libcfs_kvzalloc(chunk_size, GFP_NOFS);
if (!buf) {
lpi->lpi_rc = -ENOMEM;
return 0;
}
if (cd) {
last_called_index = cd->lpcd_first_idx;
index = cd->lpcd_first_idx + 1;
}
if (cd && cd->lpcd_last_idx)
last_index = cd->lpcd_last_idx;
else
last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
while (rc == 0) {
unsigned int buf_offset = 0;
struct llog_rec_hdr *rec;
bool partial_chunk;
off_t chunk_offset;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
++index;
if (index > last_index)
break;
CDEBUG(D_OTHER, "index: %d last_index %d\n",
index, last_index);
repeat:
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, chunk_size);
rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
index, &cur_offset, buf, chunk_size);
if (rc)
goto out;
/*
* NB: after llog_next_block() call the cur_offset is the
* offset of the next block after read one.
* The absolute offset of the current chunk is calculated
* from cur_offset value and stored in chunk_offset variable.
*/
tmp_offset = cur_offset;
if (do_div(tmp_offset, chunk_size)) {
partial_chunk = true;
chunk_offset = cur_offset & ~(chunk_size - 1);
} else {
partial_chunk = false;
chunk_offset = cur_offset - chunk_size;
}
/* NB: when rec->lrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop.
*/
for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
(char *)rec < buf + chunk_size;
rec = llog_rec_hdr_next(rec)) {
CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
rec, rec->lrh_type);
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
rec->lrh_type, rec->lrh_index);
/*
* for partial chunk the end of it is zeroed, check
* for index 0 to distinguish it.
*/
if (partial_chunk && !rec->lrh_index) {
/* concurrent llog_add() might add new records
* while llog_processing, check this is not
* the case and re-read the current chunk
* otherwise.
*/
if (index > loghandle->lgh_last_idx) {
rc = 0;
goto out;
}
CDEBUG(D_OTHER, "Re-read last llog buffer for new records, index %u, last %u\n",
index, loghandle->lgh_last_idx);
/* save offset inside buffer for the re-read */
buf_offset = (char *)rec - (char *)buf;
cur_offset = chunk_offset;
goto repeat;
}
if (!rec->lrh_len || rec->lrh_len > chunk_size) {
CWARN("invalid length %d in llog record for index %d/%d\n",
rec->lrh_len,
rec->lrh_index, index);
rc = -EINVAL;
goto out;
}
if (rec->lrh_index < index) {
CDEBUG(D_OTHER, "skipping lrh_index %d\n",
rec->lrh_index);
continue;
}
if (rec->lrh_index != index) {
CERROR("%s: Invalid record: index %u but expected %u\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
rec->lrh_index, index);
rc = -ERANGE;
goto out;
}
CDEBUG(D_OTHER,
"lrh_index: %d lrh_len: %d (%d remains)\n",
rec->lrh_index, rec->lrh_len,
(int)(buf + chunk_size - (char *)rec));
loghandle->lgh_cur_idx = rec->lrh_index;
loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
chunk_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
lpi->lpi_cbdata);
last_called_index = index;
if (rc)
goto out;
}
/* exit if the last index is reached */
if (index >= last_index) {
rc = 0;
goto out;
}
index++;
}
}
out:
if (cd)
cd->lpcd_last_idx = last_called_index;
kfree(buf);
lpi->lpi_rc = rc;
return 0;
}
/* appends if idx == -1, otherwise overwrites record idx. */
static int llog_lvfs_write_rec(struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
struct llog_cookie *reccookie, int cookiecount,
void *buf, int idx)
{
struct llog_log_hdr *llh;
int reclen = rec->lrh_len, index, rc;
struct llog_rec_tail *lrt;
struct obd_device *obd;
struct file *file;
size_t left;
ENTRY;
llh = loghandle->lgh_hdr;
file = loghandle->lgh_file;
obd = loghandle->lgh_ctxt->loc_exp->exp_obd;
/* record length should not bigger than LLOG_CHUNK_SIZE */
if (buf)
rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
else
rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
if (rc)
RETURN(rc);
if (buf)
/* write_blob adds header and tail to lrh_len. */
reclen = sizeof(*rec) + rec->lrh_len +
sizeof(struct llog_rec_tail);
if (idx != -1) {
loff_t saved_offset;
/* no header: only allowed to insert record 1 */
if (idx != 1 && !i_size_read(file->f_dentry->d_inode)) {
CERROR("idx != -1 in empty log\n");
LBUG();
}
if (idx && llh->llh_size && llh->llh_size != rec->lrh_len)
RETURN(-EINVAL);
if (!ext2_test_bit(idx, llh->llh_bitmap))
CERROR("Modify unset record %u\n", idx);
if (idx != rec->lrh_index)
CERROR("Index mismatch %d %u\n", idx, rec->lrh_index);
rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
/* we are done if we only write the header or on error */
if (rc || idx == 0)
RETURN(rc);
/* Assumes constant lrh_len */
saved_offset = sizeof(*llh) + (idx - 1) * reclen;
if (buf) {
struct llog_rec_hdr check;
/* We assume that caller has set lgh_cur_* */
saved_offset = loghandle->lgh_cur_offset;
CDEBUG(D_OTHER,
"modify record "LPX64": idx:%d/%u/%d, len:%u "
"offset %llu\n",
loghandle->lgh_id.lgl_oid, idx, rec->lrh_index,
loghandle->lgh_cur_idx, rec->lrh_len,
(long long)(saved_offset - sizeof(*llh)));
if (rec->lrh_index != loghandle->lgh_cur_idx) {
CERROR("modify idx mismatch %u/%d\n",
idx, loghandle->lgh_cur_idx);
RETURN(-EFAULT);
}
#if 1 /* FIXME remove this safety check at some point */
/* Verify that the record we're modifying is the
right one. */
rc = llog_lvfs_read_blob(obd, file, &check,
sizeof(check), saved_offset);
if (check.lrh_index != idx || check.lrh_len != reclen) {
CERROR("Bad modify idx %u/%u size %u/%u (%d)\n",
idx, check.lrh_index, reclen,
check.lrh_len, rc);
RETURN(-EFAULT);
}
#endif
}
rc = llog_lvfs_write_blob(obd, file, rec, buf, saved_offset);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
RETURN(rc);
}
/* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*
* We know that llog_current_log() will return a loghandle that is
* big enough to hold reclen, so all we care about is padding here.
*/
left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + LLOG_MIN_REC_SIZE)) {
index = loghandle->lgh_last_idx + 1;
rc = llog_lvfs_pad(obd, file, left, index);
if (rc)
RETURN(rc);
loghandle->lgh_last_idx++; /*for pad rec*/
}
/* if it's the last idx in log file, then return -ENOSPC */
if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
RETURN(-ENOSPC);
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
LASSERT(index < LLOG_BITMAP_SIZE(llh));
rec->lrh_index = index;
if (buf == NULL) {
lrt = (struct llog_rec_tail *)
((char *)rec + rec->lrh_len - sizeof(*lrt));
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
}
/*The caller should make sure only 1 process access the lgh_last_idx,
*Otherwise it might hit the assert.*/
LASSERT(index < LLOG_BITMAP_SIZE(llh));
if (ext2_set_bit(index, llh->llh_bitmap)) {
CERROR("argh, index %u already set in log bitmap?\n", index);
LBUG(); /* should never happen */
}
llh->llh_count++;
llh->llh_tail.lrt_index = index;
rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
if (rc)
RETURN(rc);
rc = llog_lvfs_write_blob(obd, file, rec, buf, file->f_pos);
if (rc)
RETURN(rc);
CDEBUG(D_RPCTRACE, "added record "LPX64": idx: %u, %u \n",
loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = index;
if ((rec->lrh_type == MDS_UNLINK_REC) ||
(rec->lrh_type == MDS_SETATTR_REC) ||
(rec->lrh_type == MDS_SETATTR64_REC))
reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
else if (rec->lrh_type == OST_SZ_REC)
reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
else if (rec->lrh_type == OST_RAID1_REC)
reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
else
reccookie->lgc_subsys = -1;
rc = 1;
}
if (rc == 0 && rec->lrh_type == LLOG_GEN_REC)
rc = 1;
RETURN(rc);
}
/* appends if idx == -1, otherwise overwrites record idx. */
static int mlog_vfs_write_rec(struct mlog_handle *loghandle,
struct mlog_rec_hdr *rec,
struct mlog_cookie *reccookie,
int cookiecount,
void *buf, int idx)
{
struct mlog_log_hdr *mlh;
int reclen = rec->mrh_len;
int index = 0;
int ret = 0;
struct mlog_rec_tail *mrt;
struct file *file;
size_t left;
struct mtfs_lowerfs *lowerfs = NULL;
MENTRY();
mlh = loghandle->mgh_hdr;
file = loghandle->mgh_file;
lowerfs = loghandle->mgh_ctxt->moc_lowerfs;
/* record length should not bigger than MLOG_CHUNK_SIZE */
if (buf){
ret = (reclen > MLOG_CHUNK_SIZE - sizeof(struct mlog_rec_hdr) -
sizeof(struct mlog_rec_tail)) ? -E2BIG : 0;
} else {
ret = (reclen > MLOG_CHUNK_SIZE) ? -E2BIG : 0;
}
if (ret) {
goto out;
}
if (buf) {
/* write_blob adds header and tail to mrh_len. */
reclen = sizeof(*rec) + rec->mrh_len +
sizeof(struct mlog_rec_tail);
}
if (idx != -1) {
loff_t saved_offset;
/* no header: only allowed to insert record 1 */
if (idx != 1 && !i_size_read(file->f_dentry->d_inode)) {
MERROR("idx != -1 in empty log\n");
MBUG();
}
if (idx && mlh->mlh_size && mlh->mlh_size != rec->mrh_len) {
ret = -EINVAL;
goto out;
}
if (!ext2_test_bit(idx, mlh->mlh_bitmap)) {
MERROR("Modify unset record %u\n", idx);
}
if (idx != rec->mrh_index) {
MERROR("Index mismatch %d %u\n", idx, rec->mrh_index);
}
ret = mlog_vfs_write_blob(lowerfs, file, &mlh->mlh_hdr, NULL, 0);
/* we are done if we only write the header or on error */
if (ret || idx == 0) {
goto out;
}
/* Assumes constant mrh_len */
saved_offset = sizeof(*mlh) + (idx - 1) * reclen;
if (buf) {
struct mlog_rec_hdr check;
/* We assume that caller has set mgh_cur_* */
saved_offset = loghandle->mgh_cur_offset;
MDEBUG("modify record %I64x: idx:%d/%u/%d, len:%u "
"offset %llu\n",
loghandle->mgh_id.mgl_oid, idx, rec->mrh_index,
loghandle->mgh_cur_idx, rec->mrh_len,
(long long)(saved_offset - sizeof(*mlh)));
if (rec->mrh_index != loghandle->mgh_cur_idx) {
MERROR("modify idx mismatch %u/%d\n",
idx, loghandle->mgh_cur_idx);
ret = -EFAULT;
goto out;
}
#if 1 /* FIXME remove this safety check at some point */
/* Verify that the record we're modifying is the
right one. */
ret = mlog_vfs_read_blob(lowerfs, file, &check,
sizeof(check), saved_offset);
if (check.mrh_index != idx || check.mrh_len != reclen) {
MERROR("bad modify idx %u/%u size %u/%u (%d)\n",
idx, check.mrh_index, reclen,
check.mrh_len, ret);
ret = -EFAULT;
goto out;
}
#endif
}
ret = mlog_vfs_write_blob(lowerfs, file, rec, buf, saved_offset);
if (ret == 0 && reccookie) {
reccookie->mgc_mgl = loghandle->mgh_id;
reccookie->mgc_index = idx;
ret = 1;
}
goto out;
}
/* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*
* We know that mlog_current_log() will return a loghandle that is
* big enough to hold reclen, so all we care about is padding here.
*/
left = MLOG_CHUNK_SIZE - (file->f_pos & (MLOG_CHUNK_SIZE - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + MLOG_MIN_REC_SIZE)) {
index = loghandle->mgh_last_idx + 1;
ret = mlog_vfs_pad(lowerfs, file, left, index);
if (ret) {
goto out;
}
loghandle->mgh_last_idx++; /*for pad rec*/
}
/* if it's the last idx in log file, then return -ENOSPC */
if (loghandle->mgh_last_idx >= MLOG_BITMAP_SIZE(mlh) - 1) {
ret = -ENOSPC;
goto out;
}
index = ++loghandle->mgh_last_idx;
rec->mrh_index = index;
if (buf == NULL) {
mrt = (struct mlog_rec_tail *)
((char *)rec + rec->mrh_len - sizeof(*mrt));
mrt->mrt_len = rec->mrh_len;
mrt->mrt_index = rec->mrh_index;
}
/*The caller should make sure only 1 process access the mgh_last_idx,
*Otherwise it might hit the assert.*/
MASSERT(index < MLOG_BITMAP_SIZE(mlh));
if (ext2_set_bit(index, mlh->mlh_bitmap)) {
MERROR("argh, index %u already set in log bitmap?\n", index);
MBUG(); /* should never happen */
}
mlh->mlh_count++;
mlh->mlh_tail.mrt_index = index;
ret = mlog_vfs_write_blob(lowerfs, file, &mlh->mlh_hdr, NULL, 0);
if (ret) {
goto out;
}
ret = mlog_vfs_write_blob(lowerfs, file, rec, buf, file->f_pos);
if (ret) {
goto out;
}
MDEBUG("added record %I64x: idx: %u, %u bytes\n",
loghandle->mgh_id.mgl_oid, index, rec->mrh_len);
if (ret == 0 && reccookie) {
reccookie->mgc_mgl = loghandle->mgh_id;
reccookie->mgc_index = index;
ret = 1;
}
if (ret == 0 && rec->mrh_type == MLOG_GEN_REC) {
ret = 1;
}
out:
MRETURN(ret);
}
/* appends if idx == -1, otherwise overwrites record idx. */
static int llog_osd_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
struct llog_cookie *reccookie, int cookiecount,
void *buf, int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_log_hdr *llh;
int reclen = rec->lrh_len;
int index, rc;
struct llog_rec_tail *lrt;
struct dt_object *o;
size_t left;
ENTRY;
LASSERT(env);
llh = loghandle->lgh_hdr;
LASSERT(llh);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(th);
CDEBUG(D_OTHER, "new record %x to "DFID"\n",
rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
/* record length should not bigger than LLOG_CHUNK_SIZE */
if (buf)
rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
else
rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
if (rc)
RETURN(rc);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
if (buf)
/* write_blob adds header and tail to lrh_len. */
reclen = sizeof(*rec) + rec->lrh_len +
sizeof(struct llog_rec_tail);
if (idx != -1) {
/* no header: only allowed to insert record 1 */
if (idx != 1 && lgi->lgi_attr.la_size == 0)
LBUG();
if (idx && llh->llh_size && llh->llh_size != rec->lrh_len)
RETURN(-EINVAL);
if (!ext2_test_bit(idx, llh->llh_bitmap))
CERROR("%s: modify unset record %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx);
if (idx != rec->lrh_index)
CERROR("%s: index mismatch %d %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
rec->lrh_index);
lgi->lgi_off = 0;
rc = llog_osd_write_blob(env, o, &llh->llh_hdr, NULL,
&lgi->lgi_off, th);
/* we are done if we only write the header or on error */
if (rc || idx == 0)
RETURN(rc);
if (buf) {
/* We assume that caller has set lgh_cur_* */
lgi->lgi_off = loghandle->lgh_cur_offset;
CDEBUG(D_OTHER,
"modify record "LPX64": idx:%d/%u/%d, len:%u "
"offset %llu\n",
loghandle->lgh_id.lgl_oid, idx, rec->lrh_index,
loghandle->lgh_cur_idx, rec->lrh_len,
(long long)(lgi->lgi_off - sizeof(*llh)));
if (rec->lrh_index != loghandle->lgh_cur_idx) {
CERROR("%s: modify idx mismatch %u/%d\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
loghandle->lgh_cur_idx);
RETURN(-EFAULT);
}
} else {
/* Assumes constant lrh_len */
lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
}
rc = llog_osd_write_blob(env, o, rec, buf, &lgi->lgi_off, th);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
RETURN(rc);
}
/* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*
* We know that llog_current_log() will return a loghandle that is
* big enough to hold reclen, so all we care about is padding here.
*/
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
left = LLOG_CHUNK_SIZE - (lgi->lgi_off & (LLOG_CHUNK_SIZE - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + LLOG_MIN_REC_SIZE)) {
index = loghandle->lgh_last_idx + 1;
rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
if (rc)
RETURN(rc);
loghandle->lgh_last_idx++; /*for pad rec*/
}
/* if it's the last idx in log file, then return -ENOSPC */
if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
RETURN(-ENOSPC);
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
LASSERT(index < LLOG_BITMAP_SIZE(llh));
rec->lrh_index = index;
if (buf == NULL) {
lrt = (struct llog_rec_tail *)((char *)rec + rec->lrh_len -
sizeof(*lrt));
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
}
/* The caller should make sure only 1 process access the lgh_last_idx,
* Otherwise it might hit the assert.*/
LASSERT(index < LLOG_BITMAP_SIZE(llh));
spin_lock(&loghandle->lgh_hdr_lock);
if (ext2_set_bit(index, llh->llh_bitmap)) {
CERROR("%s: index %u already set in log bitmap\n",
o->do_lu.lo_dev->ld_obd->obd_name, index);
spin_unlock(&loghandle->lgh_hdr_lock);
LBUG(); /* should never happen */
}
llh->llh_count++;
spin_unlock(&loghandle->lgh_hdr_lock);
llh->llh_tail.lrt_index = index;
lgi->lgi_off = 0;
rc = llog_osd_write_blob(env, o, &llh->llh_hdr, NULL, &lgi->lgi_off,
th);
if (rc)
RETURN(rc);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
rc = llog_osd_write_blob(env, o, rec, buf, &lgi->lgi_off, th);
if (rc)
RETURN(rc);
CDEBUG(D_RPCTRACE, "added record "LPX64": idx: %u, %u\n",
loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = index;
if ((rec->lrh_type == MDS_UNLINK_REC) ||
(rec->lrh_type == MDS_SETATTR64_REC))
reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
else if (rec->lrh_type == OST_SZ_REC)
reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
else
reccookie->lgc_subsys = -1;
rc = 1;
}
RETURN(rc);
}
static int llog_catinfo_deletions(struct obd_device *obd, char *buf,
int buf_len)
{
struct mds_obd *mds = &obd->u.mds;
struct llog_handle *handle;
struct lvfs_run_ctxt saved;
int size, i, count;
struct llog_catid *idarray;
char name[32] = CATLIST;
struct cb_data data;
struct llog_ctxt *ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
int rc;
ENTRY;
if (ctxt == NULL || mds == NULL)
GOTO(release_ctxt, rc = -ENODEV);
count = mds->mds_lov_desc.ld_tgt_count;
size = sizeof(*idarray) * count;
OBD_ALLOC_LARGE(idarray, size);
if (!idarray)
GOTO(release_ctxt, rc = -ENOMEM);
cfs_mutex_lock(&obd->obd_olg.olg_cat_processing);
rc = llog_get_cat_list(obd, name, 0, count, idarray);
if (rc)
GOTO(out_free, rc);
push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
data.ctxt = ctxt;
data.out = buf;
data.remains = buf_len;
for (i = 0; i < count; i++) {
int l, index, uncanceled = 0;
rc = llog_create(ctxt, &handle, &idarray[i].lci_logid, NULL);
if (rc)
GOTO(out_pop, rc);
rc = llog_init_handle(handle, 0, NULL);
if (rc) {
llog_close(handle);
GOTO(out_pop, rc = -ENOENT);
}
for (index = 1; index < (LLOG_BITMAP_BYTES * 8); index++) {
if (ext2_test_bit(index, handle->lgh_hdr->llh_bitmap))
uncanceled++;
}
l = snprintf(data.out, data.remains,
"\n[Catlog ID]: #"LPX64"#"LPX64"#%08x "
"[Log Count]: %d\n",
idarray[i].lci_logid.lgl_oid,
idarray[i].lci_logid.lgl_oseq,
idarray[i].lci_logid.lgl_ogen, uncanceled);
data.out += l;
data.remains -= l;
data.init = 1;
llog_process(handle, llog_catinfo_cb, &data, NULL);
llog_close(handle);
if (data.remains <= 0)
break;
}
EXIT;
out_pop:
pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_lvfs_ctxt, NULL);
out_free:
cfs_mutex_unlock(&obd->obd_olg.olg_cat_processing);
OBD_FREE_LARGE(idarray, size);
release_ctxt:
llog_ctxt_put(ctxt);
return rc;
}
static int llog_process_thread(void *arg)
{
struct llog_process_info *lpi = arg;
struct llog_handle *loghandle = lpi->lpi_loghandle;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct llog_process_cat_data *cd = lpi->lpi_catdata;
char *buf;
size_t chunk_size;
__u64 cur_offset, tmp_offset;
int rc = 0, index = 1, last_index;
int saved_index = 0;
int last_called_index = 0;
ENTRY;
if (llh == NULL)
RETURN(-EINVAL);
cur_offset = chunk_size = llh->llh_hdr.lrh_len;
/* expect chunk_size to be power of two */
LASSERT(is_power_of_2(chunk_size));
OBD_ALLOC_LARGE(buf, chunk_size);
if (buf == NULL) {
lpi->lpi_rc = -ENOMEM;
RETURN(0);
}
if (cd != NULL) {
last_called_index = cd->lpcd_first_idx;
index = cd->lpcd_first_idx + 1;
}
if (cd != NULL && cd->lpcd_last_idx)
last_index = cd->lpcd_last_idx;
else
last_index = LLOG_HDR_BITMAP_SIZE(llh) - 1;
while (rc == 0) {
struct llog_rec_hdr *rec;
off_t chunk_offset;
unsigned int buf_offset = 0;
bool partial_chunk;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, LLOG_HDR_BITMAP(llh)))
++index;
/* There are no indices prior the last_index */
if (index > last_index)
break;
CDEBUG(D_OTHER, "index: %d last_index %d\n", index,
last_index);
repeat:
/* get the buf with our target record; avoid old garbage */
memset(buf, 0, chunk_size);
rc = llog_next_block(lpi->lpi_env, loghandle, &saved_index,
index, &cur_offset, buf, chunk_size);
if (rc != 0)
GOTO(out, rc);
/* NB: after llog_next_block() call the cur_offset is the
* offset of the next block after read one.
* The absolute offset of the current chunk is calculated
* from cur_offset value and stored in chunk_offset variable.
*/
tmp_offset = cur_offset;
if (do_div(tmp_offset, chunk_size) != 0) {
partial_chunk = true;
chunk_offset = cur_offset & ~(chunk_size - 1);
} else {
partial_chunk = false;
chunk_offset = cur_offset - chunk_size;
}
/* NB: when rec->lrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop. */
for (rec = (struct llog_rec_hdr *)(buf + buf_offset);
(char *)rec < buf + chunk_size;
rec = llog_rec_hdr_next(rec)) {
CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
rec, rec->lrh_type);
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
rec->lrh_type, rec->lrh_index);
/* for partial chunk the end of it is zeroed, check
* for index 0 to distinguish it. */
if (partial_chunk && rec->lrh_index == 0) {
/* concurrent llog_add() might add new records
* while llog_processing, check this is not
* the case and re-read the current chunk
* otherwise. */
if (index > loghandle->lgh_last_idx)
GOTO(out, rc = 0);
CDEBUG(D_OTHER, "Re-read last llog buffer for "
"new records, index %u, last %u\n",
index, loghandle->lgh_last_idx);
/* save offset inside buffer for the re-read */
buf_offset = (char *)rec - (char *)buf;
cur_offset = chunk_offset;
goto repeat;
}
if (rec->lrh_len == 0 || rec->lrh_len > chunk_size) {
CWARN("invalid length %d in llog record for "
"index %d/%d\n", rec->lrh_len,
rec->lrh_index, index);
GOTO(out, rc = -EINVAL);
}
if (rec->lrh_index < index) {
CDEBUG(D_OTHER, "skipping lrh_index %d\n",
rec->lrh_index);
continue;
}
if (rec->lrh_index != index) {
CERROR("%s: Invalid record: index %u but "
"expected %u\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
rec->lrh_index, index);
GOTO(out, rc = -ERANGE);
}
CDEBUG(D_OTHER,
"lrh_index: %d lrh_len: %d (%d remains)\n",
rec->lrh_index, rec->lrh_len,
(int)(buf + chunk_size - (char *)rec));
loghandle->lgh_cur_idx = rec->lrh_index;
loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
chunk_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, LLOG_HDR_BITMAP(llh))) {
rc = lpi->lpi_cb(lpi->lpi_env, loghandle, rec,
lpi->lpi_cbdata);
last_called_index = index;
if (rc == LLOG_PROC_BREAK) {
GOTO(out, rc);
} else if (rc == LLOG_DEL_RECORD) {
rc = llog_cancel_rec(lpi->lpi_env,
loghandle,
rec->lrh_index);
}
if (rc)
GOTO(out, rc);
}
/* exit if the last index is reached */
if (index >= last_index)
GOTO(out, rc = 0);
++index;
}
}
out:
if (cd != NULL)
cd->lpcd_last_idx = last_called_index;
if (unlikely(rc == -EIO && loghandle->lgh_obj != NULL)) {
if (dt_object_remote(loghandle->lgh_obj)) {
/* If it is remote object, then -EIO might means
* disconnection or eviction, let's return -EAGAIN,
* so for update recovery log processing, it will
* retry until the umount or abort recovery, see
* lod_sub_recovery_thread() */
CERROR("%s retry remote llog process\n",
loghandle->lgh_ctxt->loc_obd->obd_name);
rc = -EAGAIN;
} else {
/* something bad happened to the processing of a local
* llog file, probably I/O error or the log got
* corrupted to be able to finally release the log we
* discard any remaining bits in the header */
CERROR("Local llog found corrupted\n");
while (index <= last_index) {
if (ext2_test_bit(index,
LLOG_HDR_BITMAP(llh)) != 0)
llog_cancel_rec(lpi->lpi_env, loghandle,
index);
index++;
}
rc = 0;
}
}
OBD_FREE_LARGE(buf, chunk_size);
lpi->lpi_rc = rc;
return 0;
}
/**
* Implementation of the llog_operations::lop_write
*
* This function writes the new record in the llog or modify the existed one.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in] rec llog record header. This is a real header of
* the full llog record to write. This is
* the beginning of buffer to write, the length
* of buffer is stored in \a rec::lrh_len
* \param[out] reccookie pointer to the cookie to return back if needed.
* It is used for further cancel of this llog
* record.
* \param[in] idx index of the llog record. If \a idx == -1 then
* this is append case, otherwise \a idx is
* the index of record to modify
* \param[in] th current transaction handle
*
* \retval 0 on successful write && \a reccookie == NULL
* 1 on successful write && \a reccookie != NULL
* \retval negative error if write failed
*/
static int llog_osd_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
struct llog_cookie *reccookie,
int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_log_hdr *llh;
int reclen = rec->lrh_len;
int index, rc;
struct llog_rec_tail *lrt;
struct dt_object *o;
size_t left;
bool header_is_updated = false;
ENTRY;
LASSERT(env);
llh = loghandle->lgh_hdr;
LASSERT(llh);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(th);
CDEBUG(D_OTHER, "new record %x to "DFID"\n",
rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
/* record length should not bigger than LLOG_CHUNK_SIZE */
if (reclen > LLOG_CHUNK_SIZE)
RETURN(-E2BIG);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
/**
* The modification case.
* If idx set then the record with that index must be modified.
* There are three cases possible:
* 1) the common case is the llog header update (idx == 0)
* 2) the llog record modification during llog process.
* This is indicated by the \a loghandle::lgh_cur_idx > 0.
* In that case the \a loghandle::lgh_cur_offset
* 3) otherwise this is assumed that llog consist of records of
* fixed size, i.e. catalog. The llog header must has llh_size
* field equal to record size. The record offset is calculated
* just by /a idx value
*
* During modification we don't need extra header update because
* the bitmap and record count are not changed. The record header
* and tail remains the same too.
*/
if (idx != LLOG_NEXT_IDX) {
/* llog can be empty only when first record is being written */
LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
if (!ext2_test_bit(idx, llh->llh_bitmap)) {
CERROR("%s: modify unset record %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx);
RETURN(-ENOENT);
}
if (idx != rec->lrh_index) {
CERROR("%s: modify index mismatch %d %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
rec->lrh_index);
RETURN(-EFAULT);
}
if (idx == LLOG_HEADER_IDX) {
/* llog header update */
LASSERT(reclen == sizeof(struct llog_log_hdr));
LASSERT(rec == &llh->llh_hdr);
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf,
&lgi->lgi_off, th);
RETURN(rc);
} else if (loghandle->lgh_cur_idx > 0) {
/**
* The lgh_cur_offset can be used only if index is
* the same.
*/
if (idx != loghandle->lgh_cur_idx) {
CERROR("%s: modify index mismatch %d %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
loghandle->lgh_cur_idx);
RETURN(-EFAULT);
}
lgi->lgi_off = loghandle->lgh_cur_offset;
CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
"len:%u offset %llu\n",
POSTID(&loghandle->lgh_id.lgl_oi), idx,
rec->lrh_len, (long long)lgi->lgi_off);
} else if (llh->llh_size > 0) {
if (llh->llh_size != rec->lrh_len) {
CERROR("%s: wrong record size, llh_size is %u"
" but record size is %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
llh->llh_size, rec->lrh_len);
RETURN(-EINVAL);
}
lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
} else {
/* This can be result of lgh_cur_idx is not set during
* llog processing or llh_size is not set to proper
* record size for fixed records llog. Therefore it is
* impossible to get record offset. */
CERROR("%s: can't get record offset, idx:%d, "
"len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
idx, rec->lrh_len);
RETURN(-EFAULT);
}
/* update only data, header and tail remain the same */
lgi->lgi_off += sizeof(struct llog_rec_hdr);
lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
lgi->lgi_buf.lb_buf = REC_DATA(rec);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
RETURN(rc);
}
/**
* The append case.
* The most common case of using llog. The new index is assigned to
* the new record, new bit is set in llog bitmap and llog count is
* incremented.
*
* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*/
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
left = LLOG_CHUNK_SIZE - (lgi->lgi_off & (LLOG_CHUNK_SIZE - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + LLOG_MIN_REC_SIZE)) {
index = loghandle->lgh_last_idx + 1;
rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
if (rc)
RETURN(rc);
loghandle->lgh_last_idx++; /* for pad rec */
}
/* if it's the last idx in log file, then return -ENOSPC */
if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
RETURN(-ENOSPC);
/* increment the last_idx along with llh_tail index, they should
* be equal for a llog lifetime */
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
llh->llh_tail.lrt_index = index;
/**
* NB: the caller should make sure only 1 process access
* the lgh_last_idx, e.g. append should be exclusive.
* Otherwise it might hit the assert.
*/
LASSERT(index < LLOG_BITMAP_SIZE(llh));
rec->lrh_index = index;
lrt = rec_tail(rec);
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
/* the lgh_hdr_lock protects llog header data from concurrent
* update/cancel, the llh_count and llh_bitmap are protected */
spin_lock(&loghandle->lgh_hdr_lock);
if (ext2_set_bit(index, llh->llh_bitmap)) {
CERROR("%s: index %u already set in log bitmap\n",
o->do_lu.lo_dev->ld_obd->obd_name, index);
spin_unlock(&loghandle->lgh_hdr_lock);
LBUG(); /* should never happen */
}
llh->llh_count++;
spin_unlock(&loghandle->lgh_hdr_lock);
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc)
GOTO(out, rc);
header_is_updated = true;
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
GOTO(out, rc);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc < 0)
GOTO(out, rc);
CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u\n",
POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len);
if (reccookie != NULL) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = index;
if ((rec->lrh_type == MDS_UNLINK_REC) ||
(rec->lrh_type == MDS_SETATTR64_REC))
reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
else if (rec->lrh_type == OST_SZ_REC)
reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
else
reccookie->lgc_subsys = -1;
rc = 1;
}
RETURN(rc);
out:
/* cleanup llog for error case */
spin_lock(&loghandle->lgh_hdr_lock);
ext2_clear_bit(index, llh->llh_bitmap);
llh->llh_count--;
spin_unlock(&loghandle->lgh_hdr_lock);
/* restore llog last_idx */
loghandle->lgh_last_idx--;
llh->llh_tail.lrt_index = loghandle->lgh_last_idx;
/* restore the header on disk if it was written */
if (header_is_updated) {
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
}
RETURN(rc);
}
static int mlog_process_thread(void *arg)
{
struct mlog_process_info *mpi = (struct mlog_process_info *)arg;
struct mlog_handle *loghandle = mpi->mpi_loghandle;
struct mlog_log_hdr *mlh = loghandle->mgh_hdr;
struct mlog_process_cat_data *cd = mpi->mpi_catdata;
char *buf;
__u64 cur_offset = MLOG_CHUNK_SIZE;
__u64 last_offset;
int ret = 0, index = 1, last_index;
int saved_index = 0, last_called_index = 0;
MASSERT(mlh);
MTFS_ALLOC(buf, MLOG_CHUNK_SIZE);
if (!buf) {
mpi->mpi_ret = -ENOMEM;
#ifdef __KERNEL__
complete(&mpi->mpi_completion);
#endif
return 0;
}
mtfs_daemonize_ctxt("mlog_process_thread");
if (cd != NULL) {
last_called_index = cd->mpcd_first_idx;
index = cd->mpcd_first_idx + 1;
}
if (cd != NULL && cd->mpcd_last_idx)
last_index = cd->mpcd_last_idx;
else
last_index = MLOG_BITMAP_BYTES * 8 - 1;
while (ret == 0) {
struct mlog_rec_hdr *rec;
/* skip records not set in bitmap */
while (index <= last_index &&
!ext2_test_bit(index, mlh->mlh_bitmap))
++index;
MASSERT(index <= last_index + 1);
if (index == last_index + 1)
break;
MDEBUG("index: %d last_index %d\n",
index, last_index);
/* get the buf with our target record; avoid old garbage */
last_offset = cur_offset;
ret = mlog_next_block(loghandle, &saved_index, index,
&cur_offset, buf, MLOG_CHUNK_SIZE);
if (ret) {
goto out;
}
/* NB: when rec->mrh_len is accessed it is already swabbed
* since it is used at the "end" of the loop and the rec
* swabbing is done at the beginning of the loop. */
for (rec = (struct mlog_rec_hdr *)buf;
(char *)rec < buf + MLOG_CHUNK_SIZE;
rec = (struct mlog_rec_hdr *)((char *)rec + rec->mrh_len)){
MDEBUG("processing rec 0x%p type %#x\n",
rec, rec->mrh_type);
if (MLOG_REC_HDR_NEEDS_SWABBING(rec))
mlog_swab_rec(rec, NULL);
MDEBUG("after swabbing, type=%#x idx=%d\n",
rec->mrh_type, rec->mrh_index);
if (rec->mrh_index == 0) {
/* no more records */
goto out;
}
if (rec->mrh_len == 0 || rec->mrh_len > MLOG_CHUNK_SIZE){
MERROR("invalid length %d in mlog record for "
"index %d/%d\n", rec->mrh_len,
rec->mrh_index, index);
ret = -EINVAL;
goto out;
}
if (rec->mrh_index < index) {
MDEBUG("skipping mrh_index %d\n",
rec->mrh_index);
continue;
}
MDEBUG("mrh_index: %d mrh_len: %d (%d remains)\n",
rec->mrh_index, rec->mrh_len,
(int)(buf + MLOG_CHUNK_SIZE - (char *)rec));
loghandle->mgh_cur_idx = rec->mrh_index;
loghandle->mgh_cur_offset = (char *)rec - (char *)buf +
last_offset;
/* if set, process the callback on this record */
if (ext2_test_bit(index, mlh->mlh_bitmap)) {
ret = mpi->mpi_cb(loghandle, rec,
mpi->mpi_cbdata);
last_called_index = index;
if (ret == MLOG_PROC_BREAK) {
MDEBUG("recovery from log: %llx:%x stopped\n",
loghandle->mgh_id.mgl_oid,
loghandle->mgh_id.mgl_ogen);
goto out;
} else if (ret == MLOG_DEL_RECORD) {
mlog_cancel_rec(loghandle,
rec->mrh_index);
ret = 0;
}
if (ret) {
goto out;
}
} else {
MDEBUG("Skipped index %d\n", index);
}
/* next record, still in buffer? */
++index;
if (index > last_index) {
goto out;
}
}
}
out:
if (cd != NULL)
cd->mpcd_last_idx = last_called_index;
if (buf)
MTFS_FREE(buf, MLOG_CHUNK_SIZE);
mpi->mpi_ret = ret;
#ifdef __KERNEL__
complete(&mpi->mpi_completion);
#endif
MRETURN(ret);
}
