(Illustration by Gaich Muramatsu)
(Illustration by Gaich Muramatsu)
rvmutl is the maintenance utility for RVM. Its principle features are log creation and maintenance, but there are also status and statistic display commands as well as printing commands for log records. A search facility for modifications to segments gains additional functionality from the log by allowing its use as a debugging history.
rvmutl is included in the
/bin
directory of the RVM
distribution directory. It can be invoked with a single, optional
parameter, which is interpreted as the name of a log file to be
opened. If the file doesnt exist, rvmutl will ask if you wish to
create and initialize a log with the specified name.
rvmutl prompts for commands with a * prompt, after which commands can be typed. Most commands have one or more parameters, which are separated by one or more spaces. In general, command names can be abbreviated, provided that sufficient characters are typed to uniquely specify a command. To allow typing short-cuts, some non-unique abbreviations have defined interpretations. These are specified with the command documentation
As an example, the following sequence will initialize the file log_file to have one megabyte of log record space:
% rvmutl * init_log log_file 1M * quit
After this simple sequence, the log file is ready for use.
The following pages describe the commands available in rvmutl.
rvmutl -- maintenance utility for RVM log
SYNOPSIS
rvmutl log
rvmutl
DESCRIPTION
The program
rvmutl
is the maintenance utility for
RVM log. Its priniciple features are log creation and maintenance,
but there are also status and statistic display commands, as well
as printing commands for log records. There is also a search
facility for modifications to segments. You may find rvmutl a great
debugging tools.
The are two ways to use
rvmutl
: You may supply the
name of the RVM log on the command line, or you do not supply the
name on command line but specify the log within
rvmutl
(see below).
rvmutl
prompts for commands with a * (asterisk)
prompt. Most commands have one or more parameters, which are
separated by one or more spaces. Some commands have abbreviations,
these will be specified in the command section.
For example, to examine the status of a RVM log (called logfile in the example), you can do
> rvmutl * open_log logfile * status Status of log: /home/clement/linuxobj8/rvm-src/tests/logfile log created on: Tue Nov 4 1997 17:31:10.244793 log created with: RVM Interface Version 1.3 7 Mar 1994 (more display omitted ...) * quit
As another example, suppose you want to create a RVM log device on the partition /dev/hdc7, of size 1 Megabyte, you can do
> rvmutl * i /dev/hdc7 1M * quit
Note that in the preceding example, we use the abbreviation "i" for the command "init_log", and we can use the unit "M" to stand for Megabyte.
I/O REDIRECTION
You can redirect input and output as in typical Unix shells. For example, the following command will have all display redirected to the file allrec:
* show all > allrec *
COMMANDS
In the following, I will detail all the available commands:
Show all special log records. Special log records are those non-transaction records, such as wrap-around marker and segment dictionary entry.
Build the segment dictionary by scanning through the log. Segment dictionary is a in-memory structure that maps segment code to segment file. Segment code is used in each transaction range as a short hand of the segment file involved. If the all option is specified, the whole log will be scanned, otherwise the scanning will stop when one segment dictionary entry is found. If the clear option is specified, the segment dictionary will be cleared before starting the scanning process.
Close the log file.
This provide a handy way to copy an existing log to another file or partition. Since rvm logs start on different offset, depending on whether they are files or partitions, using stock Unix commands such as cp or dd sometimes is not as nice as using this command. Some meta data (such as name of log) in the log status area in the destination log will also be updated to reflect new meta data after log copying.
Show the earliest transaction record in the log.
Initialize a log if it does not exist, or re-initialize it if it does exist. The log will have the name log and the length of data area will be length . The actual size of the log file or partition will be slightly bigger because: 1. there is a log status area (of size 3 sectors); and 2. for log used on raw partition, there will be an offset (of 16 sectors) to jump off possible disklabel of the partition.
The length parameter is specified as an integer, optionally followed by a single character size code. The following size codes are defined: K or k for kilobytes, M or m for megabytes, S or s for sectors (512 bytes), P or p for pages (which have system dependent sizes) and B or b for bytes (default).
If either parameter is not specified, init_log will prompt for the missing information.
RVM maintains the log continuously after creation. If you are going to re-initialize an existing log, make sure the log is empty before the re-initialization, otherwise, the transaction records in the log would be lost (i.e., they would not be applied to the data segment). You can apply all transaction records to the data segments and empty the log (also known as truncate in this manual) by the recover command. You can use the command status to check whether a log is empty.
Print the offset of the earliest transaction record in the log.
Print the offset of the tail of the log.
Print the transaction record at the head of log. (Note that there could be even earlier than the head record -- smaller record number and smaller time stamp. They are considered truncated from the log and can be overwritten upon them freely. They are still on the log just because it happens that rvm has not yet overwritten on them, and they may be useful for debugging. The command earliest prints the earliest record.)
Open the rvm log named
log
. You must open a log before
you can use most of the rest commands. The optional
no_tail
switch turns of the search of tail of the log (for example, when
the log tail is corrupted). The optional
no_update
and
update
switch controls whether update will be made on the
log and segment (they turn on the rvm internal global variable
rvm_no_update
).
Print the next num records. If the optional argument num is not specified, then the next one record will be printed.
Print the next num sub-records. Sub-records are the individual ranges of a transaction. If the optional argument num is not specified, then the next one sub-record will be printed.
In the following two examples, the first command dumps, from offset 0, eight bytes in hexadecimal format; the second command dumps two long integers in decimal format. Note that how the same file content can be dumped into different format.
* peek 0 / 8xb 00000000000: 0x18 0000 0000 0000 0x23 0x17 0x14 0x63 * peek 0 / 2dl 00000000000: 24 1662261027
There are seven different possible sizes: b for byte, l for long, s for short, O for offset (two long integers representing the high and low order bits respectively), f for float, D for double and t for time value.
There are nine different possible formats: c for character, d for decimal, o for octal, x for hexidecimal, u for unsigned, f for float, D for double, O for offset (two long integers representing the high and low order bits respectively) and t for time value.
You may omit the arguments c , s and f , and rvmutl will use the defaults, which are 1, d (decimal) and l (long) respectively, for the three arguments respectively.
Poke on (modify) the content of file . You must specify the argument file when you first peek or poke on a file, afterward you can omit the argument and rvmutl will continue using the current file. The argument offset determines from where the modification should start. The argument cs determines how the modifications should be done, it comprises of two components: c and s ; each of them represents the count and size respectively. c is an integer and determines how many units, as specified by s , should be modified.
Examples:
* poke 512 / 4db = 25
* peek 512 / 8db
512: 25 25 25 25 0 0 0 0
* poke 512 / 2xl = 25
* peek 512 / 8db
512: 25 0 0 0 25 0 0 0
* poke 512 / 2b = 0x26 0x27
* peek 512 / 8db
512: 38 39 38 39 25 0 0 0
* poke 512 / b = "abcde"
* peek 512 / 8db
512: 97 98 99 100 101 0 0 0
Note that in the second example, the character
x
, which
indicates hexidecimal format in peek, is sliently ignored. To enter
a hexidecimal number, you should us prefix '0x' as in the third
example. Note also that how a pattern is repeated by using the
argument
c
in first and third example.
There are seven different possible sizes: b for byte, l for long, s for short, O for offset (two long integers representing the high and low order bits respectively), f for float, D for double and t for time value.
You can also enter string by using the double-quote mark ("), as in the last example.
Print the previous num records. If the optional argument num is not specified, then the previous one record will be printed.
Print the previous num sub-records. Sub-records are the individual ranges of a transaction. If the optional argument num is not specified, then the previous one sub-record will be printed.
Quit rvmutl.
Re-read the log status from on-disk log device (file or
partition) to in-memory structure (
status
). Note that
the command
status
show the content of the in-memory
structure, but not the on-disk device. If you have made changes to
the on-disk device, you may want to re-read them into the
memory.
You can specify memory ranges to be monitored by this command. When a memory range is monitored, there will be message printed out whenever there are modification within the range (E.g. in recover ). The switch clear will clear all the monitor previous set. The switch no_update and update carry the usual meaning (c.f. open_log ). You will be asked, interactively, for the addresses and formats for the ranges you want to monitor, you should use the syntax similar to poke . The following example shows how to use monitor , show monitor , as well as how recover prints out the modification.
* open_log log3 * monitor update More ranges (y or n [n])? y Enter list of addresses/format, terminate with null line : 0x200d7550/1l : * show monitor Range Address/format Length 1 0x200d7550/1dl 4 * recover do_nv: data from log record 2101, range 52 monitored range 1, 0x200d7550, length 4 matched by modified range 0x200d7550, length 4 0x200d7550: 537286660 change_tree_insert: inserting entire range monitored range 1, 0x200d7550, length 4 matched by modified range 0x200d7550, length 4 0x200d7550: 537286660 do_nv: data from log record 2101, range 51 monitored range 1, 0x200d7550, length 4 matched by modified range 0x200ced40, length 8367 0x200d7550: 0 change_tree_insert: inserting entire range monitored range 1, 0x200d7550, length 4 matched by modified range 0x200ced40, length 8367 0x200d7550: 0 do_nv: data from log record 2101, range 50 monitored range 1, 0x200d7550, length 4 matched by modified range 0x200cdcc4, length 16 0x200d7550: 0 change_tree_insert: inserting entire range monitored range 1, 0x200d7550, length 4 matched by modified range 0x200cdcc4, length 16
Truncate the log. Truncation means applying the transaction records, that were logged on the log, to the data segment, and then reclaiming the space on the log used by those records.
Set field (or memeory locateion at addr ) to val . There are five fields that you can set: head , tail , prev_head , prev_tail and log_start . (You can use prev head and prev tail to represent prev_head and prev_tail respectively.) Only the in-memory structure of the field is changed by this command, you can use write_status to write the in-memory structure (status area) to disk.
E.g.
* show mods 0x2000005c / 8b
Record number: 40756 modifies specified range:
Modification range: 1 Log offset: 224664
VM address: 0x2000005c Length: 52
Segment code: 1 Offset: 4188
Record length: 108 Back link: 48
0x2000005c: 74 -107 0 0 0 0 0 0
* show mods 0x2000005c / 1b = 74
Record number: 40756 assigns specified values
Modification range: 1 Log offset: 224664
VM address: 0x2000005c Length: 52
Segment code: 1 Offset: 4188
Record length: 108 Back link: 48
0x2000005c: 74
* show mods 0x2000005c / 1b = 75
-- no more records
Show the size of a the give structure struct , or size of all structure. The unit is byte. The following structure can be shown: condition , device_t ( dev ), dev_region_t , FLUSH_BUF_LEN , free_page_t , MAXPATHLEN , int , list_entry_t , log_t ( log ), log_buf_t , LOG_DEV_STATUS_SIZE , log_dev_status_t , log_seg_t , LOG_SPECIAL_IOV_MAX , LOG_SPECIAL_SIZE , log_special_t , log_status_t ( status ), log_wrap_t , long , MAX_READ_LEN , mem_region_t , MIN_NV_RANGE_SIZE , MIN_FLUSH_BUF_LEN , MIN_RECOVERY_BUF_LEN , MIN_TRANS_SIZE , mutex , NUM_CACHE_TYPES , NV_LOCAL_MAX , nv_range_t ( nv ), NV_RANGE_OVERHEAD , page_size , range_t , rec_end_t , rec_hdr_t , RECOVERY_BUF_LEN , region_t , rvm_length_t , rvm_offset_t , rvm_options_t , rvm_region_t , rvm_tid_t , rw_lock_t ( rw_lock ), rw_lock_mode_t , seg_t , struct_id_t , int_tid_t , timeval , trans_hdr_t , TRANS_SIZE , tree_links_t , tree_node_t , TRUNCATE , ulong , unsigned .
Show the log status area, which is the meta data about the log (include head offset, tail offset, space used by records, total log size, first record number, last record number, log creation time, last truncation time etc).
Show the current subrecord.
Show the statistics of rvm activities.
Show the transaction record at the tail of log.
Turn on update (i.e. set to false the rvm-internal global
variable
rvm_no_update
.) Update will not be made on
the log and segment.
Turn off update (i.e. set to true the rvm-internal global
variable
rvm_no_update
.) Update will be made on the
log and segment.
Write out the in-memory log status structure to the log status block on disk.
BUGS
The command find_hole is not yet documented.
Peeking using the format or size of time value ( t ) does not work.
The way to poke an offset value is not yet documented.
sizeof of some structs (e.g. condition , int ) wrongly displays the page size rather than the size of the structure.
The use of no_update does not quite work. Rvmutl easily crashes when the switch is used.
The exact semantic of replay command is not documented, and the command itself crashes quite easily.
The use of seg_dict in set is not yet documented.
The command show timestamp has no effect.
AUTHOR
Yui Wah LEE completely rewrote this man page (Nov. 1997)