std::string const &command = args[0];
std::string data_pool_name;
+ std::set<std::string> extra_data_pool_names;
std::string pg_files_path;
std::set<pg_t> pg_files_pgs;
continue;
}
+ // Trailing positional arguments
+ if (command == "scan_extents") {
+ if (data_pool_name.empty()) {
+ data_pool_name = *i;
+ } else if (*i != data_pool_name) {
+ extra_data_pool_names.insert(*i);
+ }
+ continue;
+ }
+
// Trailing positional argument
if (i + 1 == args.end() &&
(command == "scan_inodes"
- || command == "scan_extents"
|| command == "cleanup")) {
data_pool_name = *i;
continue;
<< "' has ID " << data_pool_id << dendl;
}
- if (!fs->mds_map.is_data_pool(data_pool_id)) {
- std::cerr << "Warning: pool '" << data_pool_name << "' is not a "
- "CephFS data pool!" << std::endl;
+ if (data_pool_id != fs->mds_map.get_first_data_pool()) {
+ std::cerr << "Warning: pool '" << data_pool_name << "' is not the "
+ "main CephFS data pool!" << std::endl;
if (!force_pool) {
std::cerr << "Use --force-pool to continue" << std::endl;
return -EINVAL;
}
}
+ // Initialize extra data_ios for those commands that need it
+ if (command == "scan_extents") {
+ for (auto &data_pool_name: extra_data_pool_names) {
+ int64_t pool_id = rados.pool_lookup(data_pool_name.c_str());
+ if (data_pool_id < 0) {
+ std::cerr << "Data pool '" << data_pool_name << "' not found!" << std::endl;
+ return -ENOENT;
+ } else {
+ dout(4) << "data pool '" << data_pool_name << "' has ID " << pool_id
+ << dendl;
+ }
+
+ if (!fs->mds_map.is_data_pool(pool_id)) {
+ std::cerr << "Warning: pool '" << data_pool_name << "' is not a "
+ "CephFS data pool!" << std::endl;
+ if (!force_pool) {
+ std::cerr << "Use --force-pool to continue" << std::endl;
+ return -EINVAL;
+ }
+ }
+
+ dout(4) << "opening data pool '" << data_pool_name << "'" << dendl;
+ extra_data_ios.push_back({});
+ r = rados.ioctx_create(data_pool_name.c_str(), extra_data_ios.back());
+ if (r != 0) {
+ return r;
+ }
+ }
+ }
+
// Initialize metadata_io from MDSMap for scan_frags
if (command == "scan_frags" || command == "scan_links") {
const auto fs = fsmap->get_filesystem(fscid);
int DataScan::scan_extents()
{
- return forall_objects(data_io, false, [this](
+ std::vector<librados::IoCtx *> data_ios;
+ data_ios.push_back(&data_io);
+ for (auto &extra_data_io : extra_data_ios) {
+ data_ios.push_back(&extra_data_io);
+ }
+
+ for (auto ioctx : data_ios) {
+ int r = forall_objects(*ioctx, false, [this, ioctx](
std::string const &oid,
uint64_t obj_name_ino,
uint64_t obj_name_offset) -> int
- {
- // Read size
- uint64_t size;
- time_t mtime;
- int r = data_io.stat(oid, &size, &mtime);
- dout(10) << "handling object " << obj_name_ino
- << "." << obj_name_offset << dendl;
- if (r != 0) {
- dout(4) << "Cannot stat '" << oid << "': skipping" << dendl;
- return r;
- }
+ {
+ // Read size
+ uint64_t size;
+ time_t mtime;
+ int r = ioctx->stat(oid, &size, &mtime);
+ dout(10) << "handling object " << obj_name_ino
+ << "." << obj_name_offset << dendl;
+ if (r != 0) {
+ dout(4) << "Cannot stat '" << oid << "': skipping" << dendl;
+ return r;
+ }
+ int64_t obj_pool_id = data_io.get_id() != ioctx->get_id() ?
+ ioctx->get_id() : -1;
+
+ // I need to keep track of
+ // * The highest object ID seen
+ // * The size of the highest object ID seen
+ // * The largest object seen
+ // * The pool of the objects seen (if it is not the main data pool)
+ //
+ // Given those things, I can later infer the object chunking
+ // size, the offset of the last object (chunk size * highest ID seen),
+ // the actual size (offset of last object + size of highest ID seen),
+ // and the layout pool id.
+ //
+ // This logic doesn't take account of striping.
+ r = ClsCephFSClient::accumulate_inode_metadata(
+ data_io,
+ obj_name_ino,
+ obj_name_offset,
+ size,
+ obj_pool_id,
+ mtime);
+ if (r < 0) {
+ derr << "Failed to accumulate metadata data from '"
+ << oid << "': " << cpp_strerror(r) << dendl;
+ return r;
+ }
- // I need to keep track of
- // * The highest object ID seen
- // * The size of the highest object ID seen
- // * The largest object seen
- //
- // Given those things, I can later infer the object chunking
- // size, the offset of the last object (chunk size * highest ID seen)
- // and the actual size (offset of last object + size of highest ID seen)
- //
- // This logic doesn't take account of striping.
- r = ClsCephFSClient::accumulate_inode_metadata(
- data_io,
- obj_name_ino,
- obj_name_offset,
- size,
- mtime);
+ return r;
+ });
if (r < 0) {
- derr << "Failed to accumulate metadata data from '"
- << oid << "': " << cpp_strerror(r) << dendl;
return r;
}
+ }
- return r;
- });
+ return 0;
}
int DataScan::probe_filter(librados::IoCtx &ioctx)
// This is the layout we will use for injection, populated either
// from loaded_layout or from best guesses
file_layout_t guessed_layout;
- guessed_layout.pool_id = data_pool_id;
+ if (accum_res.obj_pool_id == -1) {
+ guessed_layout.pool_id = data_pool_id;
+ } else {
+ guessed_layout.pool_id = accum_res.obj_pool_id;
+
+ librados::IoCtx ioctx;
+ r = librados::Rados(data_io).ioctx_create2(guessed_layout.pool_id, ioctx);
+ if (r != 0) {
+ derr << "Unexpected error opening file data pool id="
+ << guessed_layout.pool_id << ": " << cpp_strerror(r) << dendl;
+ return r;
+ }
+
+ bufferlist bl;
+ int r = ioctx.getxattr(oid, "layout", bl);
+ if (r < 0) {
+ if (r != -ENODATA) {
+ derr << "Unexpected error reading layout for " << oid << ": "
+ << cpp_strerror(r) << dendl;
+ return r;
+ }
+ } else {
+ try {
+ auto q = bl.cbegin();
+ decode(loaded_layout, q);
+ } catch (ceph::buffer::error &e) {
+ derr << "Unexpected error decoding layout for " << oid << dendl;
+ return -EINVAL;
+ }
+ }
+ }
// Calculate file_size, guess the layout
if (accum_res.ceiling_obj_index > 0) {
// We have a stashed layout that we can't disprove, so apply it
guessed_layout = loaded_layout;
dout(20) << "loaded layout from xattr:"
+ << " pi: " << guessed_layout.pool_id
<< " os: " << guessed_layout.object_size
<< " sc: " << guessed_layout.stripe_count
<< " su: " << guessed_layout.stripe_unit
<< dendl;
// User might have transplanted files from a pool with a different
- // ID, so whatever the loaded_layout says, we'll force the injected
- // layout to point to the pool we really read from
- guessed_layout.pool_id = data_pool_id;
+ // ID, so if the pool from loaded_layout is not found in the list of
+ // the data pools, we'll force the injected layout to point to the
+ // pool we read from.
+ if (!fsmap->get_filesystem(fscid)->mds_map.is_data_pool(
+ guessed_layout.pool_id)) {
+ dout(20) << "overwriting layout pool_id " << data_pool_id << dendl;
+ guessed_layout.pool_id = data_pool_id;
+ }
}
if (guessed_layout.stripe_count == 1) {
// Striped file: need to examine the last stripe_count objects
// in the file to determine the size.
+ librados::IoCtx ioctx;
+ if (guessed_layout.pool_id == data_io.get_id()) {
+ ioctx.dup(data_io);
+ } else {
+ r = librados::Rados(data_io).ioctx_create2(guessed_layout.pool_id,
+ ioctx);
+ if (r != 0) {
+ derr << "Unexpected error opening file data pool id="
+ << guessed_layout.pool_id << ": " << cpp_strerror(r) << dendl;
+ return r;
+ }
+ }
+
// How many complete (i.e. not last stripe) objects?
uint64_t complete_objs = 0;
if (accum_res.ceiling_obj_index > guessed_layout.stripe_count - 1) {
uint64_t osize(0);
time_t omtime(0);
- r = data_io.stat(std::string(buf), &osize, &omtime);
+ r = ioctx.stat(std::string(buf), &osize, &omtime);
if (r == 0) {
if (osize > 0) {
// Upper bound within this object
|| loaded_layout.object_size < accum_res.max_obj_size) {
// No layout loaded, or inconsistent layout, use default
guessed_layout = file_layout_t::get_default();
- guessed_layout.pool_id = data_pool_id;
+ guessed_layout.pool_id = accum_res.obj_pool_id != -1 ?
+ accum_res.obj_pool_id : data_pool_id;
} else {
guessed_layout = loaded_layout;
}
projects / ceph.git / commitdiff