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 if start_extent <= offset_in_extents &&
84 offset_in_extents < end_extent
85 then dev#read (offset +^ pvoffset *^ extent_size) len
91 (*----------------------------------------------------------------------*)
92 (* Probe to see if it's an LVM2 PV. *)
93 let rec probe_pv lvm_plugin_id dev =
95 let uuid, _ = read_pv_label dev in
97 eprintf "LVM2 detected PV UUID %s\n%!" uuid;
98 { lvm_plugin_id = lvm_plugin_id; pv_uuid = uuid }
100 if !debug then prerr_endline (Printexc.to_string exn);
103 and read_pv_label dev =
104 (* Load the first 8 sectors. I found by experimentation that
105 * the second sector contains the header ("LABELONE" etc) and
106 * the nineth sector contains some additional information about
107 * the location of the current metadata.
109 let bits = dev#read_bitstring 0L (9 * sector_size) in
111 (*Bitmatch.hexdump_bitstring stdout bits;*)
114 | sector0 : sector_size*8 : bitstring; (* sector 0 *)
115 labelone : 8*8 : bitstring; (* "LABELONE" *)
116 padding : 16*8 : bitstring; (* Seems to contain something. *)
117 lvm2_ver : 8*8 : bitstring; (* "LVM2 001" *)
118 uuid : 32*8 : bitstring; (* UUID *)
119 padding2 : (sector_size-64)*8 : bitstring; (* to end of second sector *)
120 sector234567 : sector_size*8 * 6 : bitstring; (* sectors 2-6 *)
121 padding3 : 0x28*8 : bitstring; (* start of sector 8 *)
122 metadata_offset : 32 : littleendian;(* metadata offset *)
123 padding4 : 4*8 : bitstring;
124 metadata_length : 32 : littleendian (* length of metadata (bytes) *)
125 when Bitmatch.string_of_bitstring labelone = "LABELONE" &&
126 Bitmatch.string_of_bitstring lvm2_ver = "LVM2 001" ->
128 (* Metadata offset is relative to end of PV label. *)
129 let metadata_offset = metadata_offset +* 0x1000_l in
130 (* Metadata length appears to include the trailing \000 which
133 let metadata_length = metadata_length -* 1_l in
135 let metadata = read_metadata dev metadata_offset metadata_length in
137 let uuid = Bitmatch.string_of_bitstring uuid in
143 (sprintf "LVM2: read_pv_label: %s: not an LVM2 physical volume" dev#name)
145 and read_metadata dev offset32 len32 =
147 eprintf "metadata: offset 0x%lx len %ld bytes\n%!" offset32 len32;
149 (* Check the offset and length are sensible. *)
151 if offset32 <= Int32.max_int then Int64.of_int32 offset32
152 else invalid_arg "LVM2: read_metadata: metadata offset too large" in
154 if len32 <= 2_147_483_647_l then Int64.of_int32 len32
155 else invalid_arg "LVM2: read_metadata: metadata length too large" in
157 if offset64 <= 0x1200L || offset64 >= dev#size
158 || len64 <= 0L || offset64 +^ len64 >= dev#size then
159 invalid_arg "LVM2: read_metadata: bad metadata offset or length";
161 (* If it is outside the disk boundaries, this will throw an exception,
162 * otherwise it will read and return the metadata string.
164 dev#read offset64 (Int64.to_int len64)
166 (*----------------------------------------------------------------------*)
167 (* We are passed a list of devices which we previously identified
168 * as PVs belonging to us. From these produce a list of all LVs
169 * (as devices) and return them. Note that we don't try to detect
170 * what is on these LVs - that will be done in the main code.
172 let rec list_lvs devs =
173 (* Read the UUID and metadata (again) from each device to end up with
174 * an assoc list of PVs, keyed on the UUID.
178 let uuid, metadata = read_pv_label dev in
179 (uuid, (metadata, dev))
182 (* Parse the metadata using the external lexer/parser. *)
184 fun (uuid, (metadata, dev)) ->
185 uuid, (Virt_df_lvm2_lexer.parse_lvm2_metadata_from_string metadata,
189 (* Print the parsed metadata. *)
192 fun (uuid, (metadata, dev)) ->
193 eprintf "metadata for PV UUID %s on %s:\n" uuid dev#name;
194 output_metadata stderr metadata
197 (* Scan for volume groups. The first entry in the metadata
198 * appears to be the volume group name. This gives us a
199 * list of VGs and the metadata for each underlying PV.
204 | pvuuid, (((vgname, Metadata vgmeta) :: _), dev) ->
205 Some (vgname, (pvuuid, vgmeta))
209 let cmp ((a:string),_) ((b:string),_) = compare a b in
210 let vgnames = List.sort ~cmp vgnames in
211 let vgs = group_by vgnames in
213 (* Note that the metadata is supposed to be duplicated
214 * identically across all PVs (for redundancy purposes).
215 * In theory we should check this and use the 'seqno'
216 * field to find the latest metadata if it doesn't match,
217 * but in fact we don't check this.
220 fun (vgname, metas) ->
221 let pvuuids = List.map fst metas in
222 let _, vgmeta = List.hd metas in (* just pick any metadata *)
223 vgname, (pvuuids, vgmeta)) vgs in
228 fun (vgname, (pvuuids, vgmeta)) ->
229 eprintf "VG %s is on PVs: %s\n%!" vgname (String.concat "," pvuuids)
232 (* Some useful getter functions. If these can't get a value
233 * from the metadata or if the type is wrong they raise Not_found.
235 let rec get_int64 field meta =
236 match List.assoc field meta with
238 | _ -> raise Not_found
239 and get_int field meta min max =
240 match List.assoc field meta with
241 | Int i when Int64.of_int min <= i && i <= Int64.of_int max ->
243 | _ -> raise Not_found
244 and get_string field meta =
245 match List.assoc field meta with
247 | _ -> raise Not_found
248 and get_meta field meta =
249 match List.assoc field meta with
251 | _ -> raise Not_found
252 and get_stripes field meta = (* List of (string,int) pairs. *)
253 match List.assoc field meta with
255 let rec loop = function
257 | String pvname :: Int offset :: xs ->
258 (pvname, offset) :: loop xs
259 | _ -> raise Not_found
262 | _ -> raise Not_found
265 (* The volume groups refer to the physical volumes using their
266 * own naming system ("pv0", "pv1", etc.) instead of PV UUIDs.
268 * Each PV also has a start (in sectors) & count (in extents)
269 * of the writable area (the bit after the superblock and metadata)
270 * which normally starts at sector 384.
272 * Create a PV device (simple offset + size) and a map from PV
273 * names to these devices.
276 fun (vgname, (pvuuids, vgmeta)) ->
277 let pvdevs, extent_size =
279 (* NB: extent_size is in sectors here - we convert to bytes. *)
280 let extent_size = get_int "extent_size" vgmeta 0 (1024*1024) in
281 let extent_size = Int64.of_int extent_size *^ sector_size64 in
283 (* Get the physical_volumes section of the metadata. *)
284 let pvdevs = get_meta "physical_volumes" vgmeta in
288 | (pvname, Metadata meta) ->
290 let pvuuid = get_string "id" meta in
291 let pvuuid = canonical_uuid pvuuid in
293 (* Get the underlying physical device. *)
294 let _, dev = List.assoc pvuuid pvs in
296 (* Construct a PV device. *)
297 let pe_start = get_int64 "pe_start" meta in
298 let pe_start = pe_start *^ sector_size64 in
299 let pe_count = get_int64 "pe_count" meta in
300 let pe_count = pe_count *^ extent_size in
301 let pvdev = new offset_device pvuuid pe_start pe_count dev in
306 ) pvdevs, extent_size
308 (* Something went wrong - just return an empty map. *)
309 Not_found -> [], 0L in
310 (vgname, (pvuuids, vgmeta, pvdevs, extent_size))
313 (* Scan for logical volumes. Each VG contains several LVs.
314 * This gives us a list of LVs within each VG (hence extends
318 fun (vgname, (pvuuids, vgmeta, pvdevs, extent_size)) ->
321 let lvs = get_meta "logical_volumes" vgmeta in
322 let lvs = List.filter_map (
324 | lvname, Metadata lvmeta ->
326 let segment_count = get_int "segment_count" lvmeta 0 1024 in
328 (* Get the segments for this LV. *)
329 let segments = range 1 (segment_count+1) in
332 (fun i -> get_meta ("segment" ^ string_of_int i) lvmeta)
339 get_int64 "start_extent" segmeta in
341 get_int64 "extent_count" segmeta in
342 let segtype = get_string "type" segmeta in
344 (* Can only handle striped segments at the
347 if segtype <> "striped" then raise Not_found;
350 get_int "stripe_count" segmeta 0 1024 in
351 let stripes = get_stripes "stripes" segmeta in
353 if List.length stripes <> stripe_count then
356 (* Can only handle linear striped segments at
359 if stripe_count <> 1 then raise Not_found;
360 let pvname, pvoffset = List.hd stripes in
362 (start_extent, extent_count, pvname, pvoffset)
365 Some (lvname, segments)
367 (* Something went wrong with segments - omit this LV. *)
375 (* Something went wrong - assume no LVs found. *)
377 (vgname, (pvuuids, vgmeta, pvdevs, extent_size, lvs))
383 fun (vgname, (pvuuids, vgmeta, pvdevs, extent_size, lvs)) ->
384 eprintf "VG %s: (extent_size = %Ld bytes)\n" vgname extent_size;
386 fun (lvname, segments) ->
387 eprintf " %s/%s:\n" vgname lvname;
389 fun (start_extent, extent_count, pvname, pvoffset) ->
390 eprintf " start %Ld count %Ld at %s:%Ld\n"
391 start_extent extent_count pvname pvoffset
398 (* Finally we can set up devices for the LVs. *)
401 fun (vgname, (pvuuid, vgmeta, pvdevs, extent_size, lvs)) ->
404 fun (lvname, segments) ->
405 let name = vgname ^ "/" ^ lvname in
406 let segments = List.map (
407 fun (start_extent, extent_count, pvname, pvoffset) ->
408 (* Get the PV device. *)
409 let pvdev = List.assoc pvname pvdevs in
411 (* Extents mapped to: *)
412 (start_extent, extent_count, pvdev, pvoffset)
415 (* Create a linear mapping device. *)
416 let lv_dev = new linear_map_device name extent_size segments in
423 let lvs = List.concat lvs in
425 (* Return the list of LV devices. *)
428 (*----------------------------------------------------------------------*)
429 (* Register with main code. *)
431 lvm_type_register plugin_name probe_pv list_lvs