/* returns negative on error; 0 if success; 1 if success & log destroyed */
int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
int index)
{
struct llog_log_hdr *llh = loghandle->lgh_hdr;
int rc = 0;
ENTRY;
CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n",
index, POSTID(&loghandle->lgh_id.lgl_oi));
if (index == 0) {
CERROR("Can't cancel index 0 which is header\n");
RETURN(-EINVAL);
}
spin_lock(&loghandle->lgh_hdr_lock);
if (!ext2_clear_bit(index, llh->llh_bitmap)) {
spin_unlock(&loghandle->lgh_hdr_lock);
CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
RETURN(-ENOENT);
}
llh->llh_count--;
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
(llh->llh_count == 1) &&
(loghandle->lgh_last_idx == (LLOG_BITMAP_BYTES * 8) - 1)) {
spin_unlock(&loghandle->lgh_hdr_lock);
rc = llog_destroy(env, loghandle);
if (rc < 0) {
CERROR("%s: can't destroy empty llog #"DOSTID
"#%08x: rc = %d\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, rc);
GOTO(out_err, rc);
}
RETURN(1);
}
spin_unlock(&loghandle->lgh_hdr_lock);
rc = llog_write(env, loghandle, &llh->llh_hdr, NULL, 0, NULL, 0);
if (rc < 0) {
CERROR("%s: fail to write header for llog #"DOSTID
"#%08x: rc = %d\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, rc);
GOTO(out_err, rc);
}
RETURN(0);
out_err:
spin_lock(&loghandle->lgh_hdr_lock);
ext2_set_bit(index, llh->llh_bitmap);
llh->llh_count++;
spin_unlock(&loghandle->lgh_hdr_lock);
return rc;
}
/* returns negative on error; 0 if success; 1 if success & log destroyed */
int mlog_cancel_rec(struct mlog_handle *loghandle, int index)
{
struct mlog_log_hdr *mlh = loghandle->mgh_hdr;
int ret = 0;
MENTRY();
MDEBUG("Canceling %d in log %llx\n",
index, loghandle->mgh_id.mgl_oid);
if (index == 0) {
MERROR("Can't cancel index 0 which is header\n");
ret = -EINVAL;
goto out;
}
if (!ext2_clear_bit(index, mlh->mlh_bitmap)) {
MDEBUG("Catalog index %u already clear?\n", index);
ret = -ENOENT;
goto out;
}
mlh->mlh_count--;
if ((mlh->mlh_flags & MLOG_F_ZAP_WHEN_EMPTY) &&
(mlh->mlh_count == 1) &&
(loghandle->mgh_last_idx == (MLOG_BITMAP_BYTES * 8) - 1)) {
ret = mlog_destroy(loghandle);
if (ret) {
MERROR("Failure destroying log after last cancel: %d\n",
ret);
ext2_set_bit(index, mlh->mlh_bitmap);
mlh->mlh_count++;
} else {
ret = 1;
}
goto out;
}
ret = mlog_write_rec(loghandle, &mlh->mlh_hdr, NULL, 0, NULL, 0);
if (ret) {
MERROR("Failure re-writing header %d\n", ret);
ext2_set_bit(index, mlh->mlh_bitmap);
mlh->mlh_count++;
}
out:
MRETURN(ret);
}
/* free the goal block, clean it */
void ext2_free_block(unsigned int block)
{
unsigned int block_group;
unsigned int bit;
struct ext2_group_desc *desc;
struct ext2_sb_info * sbi = EXT2_SBI();
void * bitmap;
block_group = ext2_get_group_num (block, BLOCK);
bit = ext2_get_group_offset (block, BLOCK);
desc = ext2_get_group_desc (block_group);
bitmap = ext2_read_block_bitmap (block_group);
if ( !ext2_clear_bit(bitmap, bit) )
ext2_error("bit %d (%d) alread cleard", bit, block_group);
desc->bg_free_blocks_count ++;
sbi->s_free_blocks_count ++;
}
/*
* NOTE! When we get the inode, we're the only people
* that have access to it, and as such there are no
* race conditions we have to worry about. The inode
* is not on the hash-lists, and it cannot be reached
* through the filesystem because the directory entry
* has been deleted earlier.
*
* HOWEVER: we must make sure that we get no aliases,
* which means that we have to call "clear_inode()"
* _before_ we mark the inode not in use in the inode
* bitmaps. Otherwise a newly created file might use
* the same inode number (not actually the same pointer
* though), and then we'd have two inodes sharing the
* same inode number and space on the harddisk.
*/
void ext2_free_inode (struct inode * inode)
{
struct super_block * sb = inode->i_sb;
int is_directory;
unsigned long ino;
struct buffer_head * bh;
struct buffer_head * bh2;
unsigned long block_group;
unsigned long bit;
struct ext2_group_desc * desc;
struct ext2_super_block * es;
ino = inode->i_ino;
ext2_debug ("freeing inode %lu\n", ino);
/*
* Note: we must free any quota before locking the superblock,
* as writing the quota to disk may need the lock as well.
*/
if (!is_bad_inode(inode)) {
/* Quota is already initialized in iput() */
DQUOT_FREE_INODE(inode);
DQUOT_DROP(inode);
}
lock_super (sb);
es = sb->u.ext2_sb.s_es;
is_directory = S_ISDIR(inode->i_mode);
/* Do this BEFORE marking the inode not in use or returning an error */
clear_inode (inode);
if (ino < EXT2_FIRST_INO(sb) ||
ino > le32_to_cpu(es->s_inodes_count)) {
ext2_error (sb, "ext2_free_inode",
"reserved or nonexistent inode %lu", ino);
goto error_return;
}
block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
bh = load_inode_bitmap (sb, block_group);
if (IS_ERR(bh))
goto error_return;
/* Ok, now we can actually update the inode bitmaps.. */
if (!ext2_clear_bit (bit, bh->b_data))
ext2_error (sb, "ext2_free_inode",
"bit already cleared for inode %lu", ino);
else {
desc = ext2_get_group_desc (sb, block_group, &bh2);
if (desc) {
desc->bg_free_inodes_count =
cpu_to_le16(le16_to_cpu(desc->bg_free_inodes_count) + 1);
if (is_directory)
desc->bg_used_dirs_count =
cpu_to_le16(le16_to_cpu(desc->bg_used_dirs_count) - 1);
}
mark_buffer_dirty(bh2);
es->s_free_inodes_count =
cpu_to_le32(le32_to_cpu(es->s_free_inodes_count) + 1);
mark_buffer_dirty(sb->u.ext2_sb.s_sbh);
}
mark_buffer_dirty(bh);
if (sb->s_flags & MS_SYNCHRONOUS) {
ll_rw_block (WRITE, 1, &bh);
wait_on_buffer (bh);
}
sb->s_dirt = 1;
error_return:
unlock_super (sb);
}
/**
* 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 inline void log_clear_bit(struct log_c *l,
uint32_t *bs, unsigned bit)
{
ext2_clear_bit(bit, (unsigned long *) bs);
l->touched_dirtied = 1;
}
/*
* NOTE! When we get the inode, we're the only people
* that have access to it, and as such there are no
* race conditions we have to worry about. The inode
* is not on the hash-lists, and it cannot be reached
* through the filesystem because the directory entry
* has been deleted earlier.
*
* HOWEVER: we must make sure that we get no aliases,
* which means that we have to call "clear_inode()"
* _before_ we mark the inode not in use in the inode
* bitmaps. Otherwise a newly created file might use
* the same inode number (not actually the same pointer
* though), and then we'd have two inodes sharing the
* same inode number and space on the harddisk.
*/
void ext2_free_inode (struct inode * inode)
{
struct super_block * sb = inode->i_sb;
int is_directory;
unsigned long ino;
struct buffer_head * bh;
struct buffer_head * bh2;
unsigned long block_group;
unsigned long bit;
int bitmap_nr;
struct ext2_group_desc * gdp;
struct ext2_super_block * es;
if (!inode->i_dev) {
printk ("ext2_free_inode: inode has no device\n");
return;
}
if (inode->i_count > 1) {
printk ("ext2_free_inode: inode has count=%d\n", inode->i_count);
return;
}
if (inode->i_nlink) {
printk ("ext2_free_inode: inode has nlink=%d\n",
(int) inode->i_nlink);
return;
}
if (!sb) {
printk("ext2_free_inode: inode on nonexistent device\n");
return;
}
ino = inode->i_ino;
ext2_debug ("freeing inode %lu\n", ino);
/*
* Note: we must free any quota before locking the superblock,
* as writing the quota to disk may need the lock as well.
*/
DQUOT_FREE_INODE(sb, inode);
DQUOT_DROP(inode);
lock_super (sb);
es = sb->u.ext2_sb.s_es;
if (ino < EXT2_FIRST_INO(sb) ||
ino > le32_to_cpu(es->s_inodes_count)) {
ext2_error (sb, "free_inode",
"reserved inode or nonexistent inode");
goto error_return;
}
block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
bitmap_nr = load_inode_bitmap (sb, block_group);
if (bitmap_nr < 0)
goto error_return;
bh = sb->u.ext2_sb.s_inode_bitmap[bitmap_nr];
is_directory = S_ISDIR(inode->i_mode);
/* Do this BEFORE marking the inode not in use */
clear_inode (inode);
/* Ok, now we can actually update the inode bitmaps.. */
if (!ext2_clear_bit (bit, bh->b_data))
ext2_warning (sb, "ext2_free_inode",
"bit already cleared for inode %lu", ino);
else {
gdp = ext2_get_group_desc (sb, block_group, &bh2);
if (gdp) {
gdp->bg_free_inodes_count =
cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) + 1);
if (is_directory)
gdp->bg_used_dirs_count =
cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) - 1);
}
mark_buffer_dirty(bh2, 1);
es->s_free_inodes_count =
cpu_to_le32(le32_to_cpu(es->s_free_inodes_count) + 1);
mark_buffer_dirty(sb->u.ext2_sb.s_sbh, 1);
}
mark_buffer_dirty(bh, 1);
if (sb->s_flags & MS_SYNCHRONOUS) {
ll_rw_block (WRITE, 1, &bh);
wait_on_buffer (bh);
}
sb->s_dirt = 1;
error_return:
unlock_super (sb);
}
/* returns negative on error; 0 if success; 1 if success & log destroyed */
int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
int index)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_device *dt;
struct llog_log_hdr *llh = loghandle->lgh_hdr;
struct thandle *th;
int rc;
int rc1;
bool subtract_count = false;
ENTRY;
CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n", index,
POSTID(&loghandle->lgh_id.lgl_oi));
if (index == 0) {
CERROR("Can't cancel index 0 which is header\n");
RETURN(-EINVAL);
}
LASSERT(loghandle != NULL);
LASSERT(loghandle->lgh_ctxt != NULL);
LASSERT(loghandle->lgh_obj != NULL);
dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = llog_declare_write_rec(env, loghandle, &llh->llh_hdr, index, th);
if (rc < 0)
GOTO(out_trans, rc);
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY))
rc = llog_declare_destroy(env, loghandle, th);
th->th_wait_submit = 1;
rc = dt_trans_start_local(env, dt, th);
if (rc < 0)
GOTO(out_trans, rc);
down_write(&loghandle->lgh_lock);
/* clear bitmap */
mutex_lock(&loghandle->lgh_hdr_mutex);
if (!ext2_clear_bit(index, LLOG_HDR_BITMAP(llh))) {
CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
GOTO(out_unlock, rc);
}
loghandle->lgh_hdr->llh_count--;
subtract_count = true;
/* Pass this index to llog_osd_write_rec(), which will use the index
* to only update the necesary bitmap. */
lgi->lgi_cookie.lgc_index = index;
/* update header */
rc = llog_write_rec(env, loghandle, &llh->llh_hdr, &lgi->lgi_cookie,
LLOG_HEADER_IDX, th);
if (rc != 0)
GOTO(out_unlock, rc);
if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
(llh->llh_count == 1) &&
((loghandle->lgh_last_idx == LLOG_HDR_BITMAP_SIZE(llh) - 1) ||
(loghandle->u.phd.phd_cat_handle != NULL &&
loghandle->u.phd.phd_cat_handle->u.chd.chd_current_log !=
loghandle))) {
/* never try to destroy it again */
llh->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY;
rc = llog_trans_destroy(env, loghandle, th);
if (rc < 0) {
/* Sigh, can not destroy the final plain llog, but
* the bitmap has been clearly, so the record can not
* be accessed anymore, let's return 0 for now, and
* the orphan will be handled by LFSCK. */
CERROR("%s: can't destroy empty llog #"DOSTID
"#%08x: rc = %d\n",
loghandle->lgh_ctxt->loc_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, rc);
GOTO(out_unlock, rc);
}
rc = LLOG_DEL_PLAIN;
}
out_unlock:
mutex_unlock(&loghandle->lgh_hdr_mutex);
up_write(&loghandle->lgh_lock);
out_trans:
rc1 = dt_trans_stop(env, dt, th);
if (rc == 0)
rc = rc1;
if (rc < 0 && subtract_count) {
mutex_lock(&loghandle->lgh_hdr_mutex);
loghandle->lgh_hdr->llh_count++;
ext2_set_bit(index, LLOG_HDR_BITMAP(llh));
mutex_unlock(&loghandle->lgh_hdr_mutex);
}
RETURN(rc);
}
/* 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, old_tail_idx;
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 "DOSTID": idx:%d/%u/%d, len:%u "
"offset %llu\n",
POSTID(&loghandle->lgh_id.lgl_oi), 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);
old_tail_idx = llh->llh_tail.lrt_index;
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)
GOTO(out, rc);
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;
rc = llog_osd_write_blob(env, o, rec, buf, &lgi->lgi_off, th);
out:
/* cleanup llog for error case */
if (rc) {
spin_lock(&loghandle->lgh_hdr_lock);
ext2_clear_bit(index, llh->llh_bitmap);
llh->llh_count--;
spin_unlock(&loghandle->lgh_hdr_lock);
/* restore the header */
loghandle->lgh_last_idx--;
llh->llh_tail.lrt_index = old_tail_idx;
lgi->lgi_off = 0;
llog_osd_write_blob(env, o, &llh->llh_hdr, NULL,
&lgi->lgi_off, th);
}
CDEBUG(D_RPCTRACE, "added record "DOSTID": idx: %u, %u\n",
POSTID(&loghandle->lgh_id.lgl_oi), 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);
}
void ext2_free_inode (struct inode * inode)
{
struct super_block * sb;
struct buffer_head * bh;
struct buffer_head * bh2;
unsigned long block_group;
unsigned long bit;
int bitmap_nr;
int bs;
struct ext2_group_desc * gdp;
struct ext2_super_block * es;
if (!inode)
return;
if (!inode->i_dev) {
printk ("ext2_free_inode: inode has no device\n");
return;
}
if (inode->i_count > 1) {
printk ("ext2_free_inode: inode has count=%ld\n",
inode->i_count);
return;
}
if (inode->i_nlink) {
printk ("ext2_free_inode: inode has nlink=%d\n",
inode->i_nlink);
return;
}
sb = inode->i_sb;
if (!sb) {
printk("ext2_free_inode: inode on nonexistent device\n");
return;
}
ext2_debug ("freeing inode %lu\n", inode->i_ino);
/* We need to kill quota references now, before grabbing the
* superblock lock because writing the quota out to disk
* may need to lock the superblock as well.
*
* It is safe to do this early instead of the original
* places because we cannot be here in ext2_free_inode
* if any other references to this inode exist at all.
*
* Based upon a 2.1.x fix by Bill Hawes. --DaveM
*/
if (sb->dq_op) {
sb->dq_op->free_inode (inode, 1);
if (IS_WRITABLE (inode))
sb->dq_op->drop(inode);
}
lock_super (sb);
bs = BYTE_SWAP(inode->i_sb->u.ext2_sb.s_byte_swapped);
if (inode->i_ino < EXT2_FIRST_INO(sb) ||
inode->i_ino > e_swab (bs, sb->u.ext2_sb.s_es->s_inodes_count)) {
ext2_error (sb, "free_inode",
"reserved inode or nonexistent inode");
unlock_super (sb);
return;
}
es = sb->u.ext2_sb.s_es;
block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(sb);
bit = (inode->i_ino - 1) % EXT2_INODES_PER_GROUP(sb);
bitmap_nr = load_inode_bitmap (sb, block_group);
if (bitmap_nr < 0) {
unlock_super (sb);
return;
}
bh = sb->u.ext2_sb.s_inode_bitmap[bitmap_nr];
if (!ext2_clear_bit (bit, bh->b_data))
ext2_warning (sb, "ext2_free_inode",
"bit already cleared for inode %lu", inode->i_ino);
else {
gdp = get_group_desc (sb, block_group, &bh2);
e_set_swab (bs, gdp->bg_free_inodes_count,
e_swab (bs, gdp->bg_free_inodes_count) + 1);
if (S_ISDIR(inode->i_mode))
e_set_swab (bs, gdp->bg_used_dirs_count,
e_swab (bs, gdp->bg_used_dirs_count) - 1);
mark_buffer_dirty(bh2, 1);
e_set_swab (bs, es->s_free_inodes_count,
e_swab (bs, es->s_free_inodes_count) + 1);
mark_buffer_dirty(sb->u.ext2_sb.s_sbh, 1);
inode->i_dirt = 0;
}
mark_buffer_dirty(bh, 1);
if (sb->s_flags & MS_SYNCHRONOUS) {
ll_rw_block (WRITE, 1, &bh);
wait_on_buffer (bh);
}
sb->s_dirt = 1;
clear_inode (inode);
unlock_super (sb);
}
