1 (* 'df' command for virtual domains.
3 (C) Copyright 2007 Richard W.M. Jones, Red Hat Inc.
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 open Virt_df_gettext.Gettext
29 open Virt_df_lvm2_metadata
31 let plugin_name = "LVM2"
34 let sector_size64 = 512L
36 (*----------------------------------------------------------------------*)
37 (* Block device which can do linear maps, same as the kernel dm-linear.c *)
38 class linear_map_device name extent_size segments =
39 (* The segments are passed containing (start_extent, extent_count, ...)
40 * but it's easier to deal with (start_extent, end_extent, ...) so
43 let segments = List.map
44 (fun (start_extent, extent_count, dev, pvoffset) ->
45 (start_extent, start_extent +^ extent_count, dev, pvoffset)
48 (* Calculate the size of the device (in bytes). Note that because
49 * of the random nature of the mapping this doesn't imply that we can
50 * satisfy any read request up to the full size.
54 (List.map (fun (_, end_extent, _, _) -> end_extent) segments) in
55 let size = size_in_extents *^ extent_size in
61 (* Read method checks which segment the request lies inside and
62 * maps it to the underlying device. If there is no mapping then
63 * we have to return an error.
65 * The request must lie inside a single extent, otherwise this is
66 * also an error (XXX - should lift this restriction, however default
67 * extent size is 4 MB so we probably won't hit this very often).
69 method read offset len =
70 let offset_in_extents = offset /^ extent_size in
72 (* Check we don't cross an extent boundary. *)
73 if (offset +^ Int64.of_int (len-1)) /^ extent_size <> offset_in_extents
74 then invalid_arg "linear_map_device: request crosses extent boundary";
76 if offset_in_extents < 0L || offset_in_extents >= size_in_extents then
77 invalid_arg "linear_map_device: read outside device";
79 let rec loop = function
81 invalid_arg "linear_map_device: offset not mapped"
82 | (start_extent, end_extent, dev, pvoffset) :: rest ->
83 eprintf "pvoffset = %Ld\n" pvoffset;
84 if start_extent <= offset_in_extents &&
85 offset_in_extents < end_extent
86 then dev#read (offset +^ pvoffset *^ extent_size) len
92 (*----------------------------------------------------------------------*)
93 (* Probe to see if it's an LVM2 PV. *)
94 let rec probe_pv lvm_plugin_id dev =
96 let uuid, _ = read_pv_label dev in
98 eprintf "LVM2 detected PV UUID %s\n%!" uuid;
99 { lvm_plugin_id = lvm_plugin_id; pv_uuid = uuid }
101 if !debug then prerr_endline (Printexc.to_string exn);
104 and read_pv_label dev =
105 (* Load the first 8 sectors. I found by experimentation that
106 * the second sector contains the header ("LABELONE" etc) and
107 * the nineth sector contains some additional information about
108 * the location of the current metadata.
110 let bits = dev#read_bitstring 0L (9 * sector_size) in
112 (*Bitmatch.hexdump_bitstring stdout bits;*)
115 | sector0 : sector_size*8 : bitstring; (* sector 0 *)
116 labelone : 8*8 : bitstring; (* "LABELONE" *)
117 padding : 16*8 : bitstring; (* Seems to contain something. *)
118 lvm2_ver : 8*8 : bitstring; (* "LVM2 001" *)
119 uuid : 32*8 : bitstring; (* UUID *)
120 padding2 : (sector_size-64)*8 : bitstring; (* to end of second sector *)
121 sector234567 : sector_size*8 * 6 : bitstring; (* sectors 2-6 *)
122 padding3 : 0x28*8 : bitstring; (* start of sector 8 *)
123 metadata_offset : 32 : littleendian;(* metadata offset *)
124 padding4 : 4*8 : bitstring;
125 metadata_length : 32 : littleendian (* length of metadata (bytes) *)
126 when Bitmatch.string_of_bitstring labelone = "LABELONE" &&
127 Bitmatch.string_of_bitstring lvm2_ver = "LVM2 001" ->
129 (* Metadata offset is relative to end of PV label. *)
130 let metadata_offset = metadata_offset +* 0x1000_l in
131 (* Metadata length appears to include the trailing \000 which
134 let metadata_length = metadata_length -* 1_l in
136 let metadata = read_metadata dev metadata_offset metadata_length in
138 let uuid = Bitmatch.string_of_bitstring uuid in
144 (sprintf "LVM2: read_pv_label: %s: not an LVM2 physical volume" dev#name)
146 and read_metadata dev offset32 len32 =
148 eprintf "metadata: offset 0x%lx len %ld bytes\n%!" offset32 len32;
150 (* Check the offset and length are sensible. *)
152 if offset32 <= Int32.max_int then Int64.of_int32 offset32
153 else invalid_arg "LVM2: read_metadata: metadata offset too large" in
155 if len32 <= 2_147_483_647_l then Int64.of_int32 len32
156 else invalid_arg "LVM2: read_metadata: metadata length too large" in
158 if offset64 <= 0x1200L || offset64 >= dev#size
159 || len64 <= 0L || offset64 +^ len64 >= dev#size then
160 invalid_arg "LVM2: read_metadata: bad metadata offset or length";
162 (* If it is outside the disk boundaries, this will throw an exception,
163 * otherwise it will read and return the metadata string.
165 dev#read offset64 (Int64.to_int len64)
167 (*----------------------------------------------------------------------*)
168 (* We are passed a list of devices which we previously identified
169 * as PVs belonging to us. From these produce a list of all LVs
170 * (as devices) and return them. Note that we don't try to detect
171 * what is on these LVs - that will be done in the main code.
173 let rec list_lvs devs =
174 (* Read the UUID and metadata (again) from each device to end up with
175 * an assoc list of PVs, keyed on the UUID.
179 let uuid, metadata = read_pv_label dev in
180 (uuid, (metadata, dev))
183 (* Parse the metadata using the external lexer/parser. *)
185 fun (uuid, (metadata, dev)) ->
186 uuid, (Virt_df_lvm2_lexer.parse_lvm2_metadata_from_string metadata,
190 (* Print the parsed metadata. *)
193 fun (uuid, (metadata, dev)) ->
194 eprintf "metadata for PV UUID %s on %s:\n" uuid dev#name;
195 output_metadata stderr metadata
198 (* Scan for volume groups. The first entry in the metadata
199 * appears to be the volume group name. This gives us a
200 * list of VGs and the metadata for each underlying PV.
205 | pvuuid, (((vgname, Metadata vgmeta) :: _), dev) ->
206 Some (vgname, (pvuuid, vgmeta))
210 let cmp ((a:string),_) ((b:string),_) = compare a b in
211 let vgnames = List.sort ~cmp vgnames in
212 let vgs = group_by vgnames in
214 (* Note that the metadata is supposed to be duplicated
215 * identically across all PVs (for redundancy purposes).
216 * In theory we should check this and use the 'seqno'
217 * field to find the latest metadata if it doesn't match,
218 * but in fact we don't check this.
221 fun (vgname, metas) ->
222 let pvuuids = List.map fst metas in
223 let _, vgmeta = List.hd metas in (* just pick any metadata *)
224 vgname, (pvuuids, vgmeta)) vgs in
229 fun (vgname, (pvuuids, vgmeta)) ->
230 eprintf "VG %s is on PVs: %s\n%!" vgname (String.concat "," pvuuids)
233 (* Some useful getter functions. If these can't get a value
234 * from the metadata or if the type is wrong they raise Not_found.
236 let rec get_int64 field meta =
237 match List.assoc field meta with
239 | _ -> raise Not_found
240 and get_int field meta min max =
241 match List.assoc field meta with
242 | Int i when Int64.of_int min <= i && i <= Int64.of_int max ->
244 | _ -> raise Not_found
245 and get_string field meta =
246 match List.assoc field meta with
248 | _ -> raise Not_found
249 and get_meta field meta =
250 match List.assoc field meta with
252 | _ -> raise Not_found
253 and get_stripes field meta = (* List of (string,int) pairs. *)
254 match List.assoc field meta with
256 let rec loop = function
258 | String pvname :: Int offset :: xs ->
259 (pvname, offset) :: loop xs
260 | _ -> raise Not_found
263 | _ -> raise Not_found
266 (* The volume groups refer to the physical volumes using their
267 * own naming system ("pv0", "pv1", etc.) instead of PV UUIDs.
269 * Each PV also has a start (in sectors) & count (in extents)
270 * of the writable area (the bit after the superblock and metadata)
271 * which normally starts at sector 384.
273 * Create a PV device (simple offset + size) and a map from PV
274 * names to these devices.
277 fun (vgname, (pvuuids, vgmeta)) ->
278 let pvdevs, extent_size =
280 (* NB: extent_size is in sectors here - we convert to bytes. *)
281 let extent_size = get_int "extent_size" vgmeta 0 (1024*1024) in
282 let extent_size = Int64.of_int extent_size *^ sector_size64 in
284 (* Get the physical_volumes section of the metadata. *)
285 let pvdevs = get_meta "physical_volumes" vgmeta in
289 | (pvname, Metadata meta) ->
291 let pvuuid = get_string "id" meta in
292 let pvuuid = canonical_uuid pvuuid in
294 (* Get the underlying physical device. *)
295 let _, dev = List.assoc pvuuid pvs in
297 (* Construct a PV device. *)
298 let pe_start = get_int64 "pe_start" meta in
299 let pe_start = pe_start *^ sector_size64 in
300 let pe_count = get_int64 "pe_count" meta in
301 let pe_count = pe_count *^ extent_size in
302 let pvdev = new offset_device pvuuid pe_start pe_count dev in
307 ) pvdevs, extent_size
309 (* Something went wrong - just return an empty map. *)
310 Not_found -> [], 0L in
311 (vgname, (pvuuids, vgmeta, pvdevs, extent_size))
314 (* Scan for logical volumes. Each VG contains several LVs.
315 * This gives us a list of LVs within each VG (hence extends
319 fun (vgname, (pvuuids, vgmeta, pvdevs, extent_size)) ->
322 let lvs = get_meta "logical_volumes" vgmeta in
323 let lvs = List.filter_map (
325 | lvname, Metadata lvmeta ->
327 let segment_count = get_int "segment_count" lvmeta 0 1024 in
329 (* Get the segments for this LV. *)
330 let segments = range 1 (segment_count+1) in
333 (fun i -> get_meta ("segment" ^ string_of_int i) lvmeta)
340 get_int64 "start_extent" segmeta in
342 get_int64 "extent_count" segmeta in
343 let segtype = get_string "type" segmeta in
345 (* Can only handle striped segments at the
348 if segtype <> "striped" then raise Not_found;
351 get_int "stripe_count" segmeta 0 1024 in
352 let stripes = get_stripes "stripes" segmeta in
354 if List.length stripes <> stripe_count then
357 (* Can only handle linear striped segments at
360 if stripe_count <> 1 then raise Not_found;
361 let pvname, pvoffset = List.hd stripes in
363 (start_extent, extent_count, pvname, pvoffset)
366 Some (lvname, segments)
368 (* Something went wrong with segments - omit this LV. *)
376 (* Something went wrong - assume no LVs found. *)
378 (vgname, (pvuuids, vgmeta, pvdevs, extent_size, lvs))
384 fun (vgname, (pvuuids, vgmeta, pvdevs, extent_size, lvs)) ->
385 eprintf "VG %s: (extent_size = %Ld bytes)\n" vgname extent_size;
387 fun (lvname, segments) ->
388 eprintf " %s/%s:\n" vgname lvname;
390 fun (start_extent, extent_count, pvname, pvoffset) ->
391 eprintf " start %Ld count %Ld at %s:%Ld\n"
392 start_extent extent_count pvname pvoffset
399 (* Finally we can set up devices for the LVs. *)
402 fun (vgname, (pvuuid, vgmeta, pvdevs, extent_size, lvs)) ->
405 fun (lvname, segments) ->
406 let name = vgname ^ "/" ^ lvname in
407 let segments = List.map (
408 fun (start_extent, extent_count, pvname, pvoffset) ->
409 (* Get the PV device. *)
410 let pvdev = List.assoc pvname pvdevs in
412 (* Extents mapped to: *)
413 (start_extent, extent_count, pvdev, pvoffset)
416 (* Create a linear mapping device. *)
417 let lv_dev = new linear_map_device name extent_size segments in
424 let lvs = List.concat lvs in
426 (* Return the list of LV devices. *)
429 (*----------------------------------------------------------------------*)
430 (* Register with main code. *)
432 lvm_type_register plugin_name probe_pv list_lvs