1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
|
Nov. 94 : Work begins.
Dec. 94 v0.01 : First version, directly changing buffers' device
in getblk, bread and co. No swap. Added an entry
in /proc for debugging purpose. (NR)
Jan. 95 v0.10 : Reworked the whole thing, due to some FS problems
with buffers on the same virtual device not belonging
to the same real device. Gave up on buffers device
change, and now using request device, which is
obviously better. (NR)
Jan. 15/95 v0.11: FS is now working fine (using an ext2). Swap added
but not yet tested. (NR)
Jan. 21/95 v0.12: Fixed a king-size bug in swap, now working flawlessly.
Some cosmetic changes to /proc/mdstat. Also cleaned
up md tools and wrote man page. (NR)
Feb. 14/95 v0.13: Fixed a bug preventing from using any md_dev other
than md0 ! Some checking added in the kernel.
Real devices now look like mounted when added to a
md_dev (use an inode as a lock, so it works with
fs_may_mount). Updated md tools (now behaving a bit
like mount). Man page is more confusing. (R)
Mar. 08/95 v0.14: Introduced a per md_dev 'personality' that indicates
the way the device is managed. Two personalities are
currently implemented : linear, which is the old (and
rather reliable) one, and striped, which is a new one
(real devices are no more appended to each other, but
stacked so we have a chunk on first disk, then
another on second...). The striped personality should
show some speed improvement with SCSI disks. Updated
md tools accordingly. (R)
Mar. 20/95 v0.15: Cleaned up ll_rw_block.c a bit. Debug informations in
/proc/mdstat are now optional (default is off). Major
number is now 9 (registered number). (R)
Mar. 27/95 v0.16: First attempt to handle errors in a RAID-(1 and up)
system. Using a hacked version of wait_on_buffer.
It should work fine, but it's not really nice
to the kernel layering. I think I'd better
find another idea for that damned thing. Can
now choose between linear and striped mode at
config time. This should save a few bytes on
low memory systems. (NR)
Apr. 04/95 v0.17: Ok, things are now much more modular (created
linear.c and raid0.c). This should make module
versions very easy to implement. (NR)
Apr. 06/95 v0.18: Since some people around don't like the way I trap
requests in ll_rw_block.c, rewrote the whole
stuff as a REAL device driver. Now working
with minimal support in this file (maybe
somewhat less efficient, since requests are now
stuffed back to the physical device driver). (NR)
Apr. 11/95 v0.19: Arghh ! Fixed a fxcking bug that must have been
here since v0.14 ! Unfortunately, this changes
blocks repartition on physical devices. Had to
re-mkfs my md devices :(. Anyway, it works a lot
better now. Now using read ahead. Really nice
with my old 1542. The best would be to have a
per-md-dev read_ahead value, instead of a
per-major-number value. Adapted to 1.2.5. (R)
Apr. 15/95 v0.20: Fixed a stupid bug in linear mode (introduced
with v0.19, thanks to Jeff Uphoff for reporting
the bug as well as being my first ALPHA tester).
Released as 0.19b not to confuse people that are
downloading the patch. New syntax to mdadd and co,
but this will have to be rewritten soon. (R)
May 01/95 v0.21: Now using early requests removal, as used in scsi
and ide drivers. This saves some requests, and
improves throughput. Fixed a small bug introduced
by a change in a kernel patch somewhere around
1.2.4. Still playing with RAID-1 (giving me
headaches because of non-atomic writes). It is now
usable, but is awfully slow when writing, and
no errors trapping yet. (NR)
May 05/95 v0.22: Added modules support. Only having md.c compiled
into the kernel is mandatory. Personalities are
optional and can be loaded at run time. Didn't
use ksyms.c to register exported symbols, but
instead declared them in md.c . Hope this will
work nicely (it does... ;-) (NR)
May 26/95 v0.23: Added make_double_request to ll_rw_block.c .
This gives really better performances to RAID-1
devices, and solve somewhat the atomicity problem
that bothered the whole thing since weeks... Had to
change some things in the SCSI driver and to
buffers code to make things work. Some cleanups in
error handling, although it's fully untested.
New command 'mdop' to invalid and revalid
physical devices in a raid1 device.
Corrected a small bug in mdrun, and added maximum
fault field in mdtab. Added version number to mdadd
& co. Partially rewrote README, mdadd.8, and created
mdtab.5. (R)
May 28/95 v0.24: Still problems with RAID-1 and SCSI. Trying to delay
sister requests queuing in scsi.c (shouldn't bother
normal activity, but will slow down RAID-1 even more).
The question is : does it all worth it ? It seems to
work, but writing is SLOW, and it eats more than 20k
of kernel space on my machine, because of the extended
buffer_head structure. Won't work with mixed IDE/SCSI
devices until sister_req and shared_count are handled
in non-SCSI disk drivers. (R)
June 05/95 v0.25: Some RAID-5 defines... keep smiling ! Corrected a small
bug that cause a kernel Oops when using a md device
that is not being run... This is a very frequent
mistake that should be included in a FAQ. Chunk size
can now be directly set by indicating the size in
kilobytes. mdrun -p0 -c1 /dev/md0 becomes
mdrun -p0 -c8k /dev/md0. Even works in /etc/mdtab !
Idea from phil@zeus.fasttax.com . (R)
June 10/95 v0.26: Corrected bug that appears with CONFIG_MODVERSIONS
enabled (report from pstrode@exec.com). Switched
linear and raid0 to BETA. (R)
June 13/95 v0.26b: Arghh ! One more silly bug that nobody reported !
Have to learn how to use cpp ! Will bother people
that use a device different from md0. (R)
June 21/95 v0.27: Started RAID-5 integration... Extended md_personality
structure, and first shot of the raid5 kernel daemon.
Some changes in error traping. Adapted to 1.3.4.
Some docs update by cdr@star.net. (R)
July 04/95 v0.28: Tried to reduce memory usage for people that do not
use RAID-1 nor RAID-5. Removed bb from README (sad,
sad, sad...). Some stats are now avaible for chunk
size determination. (R)
July 15/95 v0.29: Patch only. Bug report from Phil Packer (removed
md_cur_req and moved requests handling down to
personalities, now needs some cleanup). Still wonder
why it only happends his machine, and not mine...
Anyway, lotta thanks to Phil for his very nice work
identifying the problem. (R)
July 20/95 v0.30: Some cleanup after 0.29 changes. Still deadlock problem
with RAID-5. Added mdcreate command and some checksum
info to mdadd. (R)
Oct. 06/95 v0.31: First non-mil version ! Lot of work have been done to
speed things up, specially RAID-0 (less original Linux
code used, more custom code, and bigger kernel ;-).
Small bug fix in read_ahead calculation when removing
a device. Adapted to 1.3.31. (R)
Nov. 26/95 v0.32: Linear and RAID-0 only version tuned for kernel
integration. Happy Birthday to MD ! (R)
Jan. 14/96 v0.33: Added some locks here and there. (R)
Feb. 25/96 v0.34: Some bug fixes (1k buffers...). Sent to Linus. (R)
Apr. 19/96 v0.35: Switched back to the ol' good way (just filter,
don't requeue). It works, simply works (1.3.93).
|
