Gentoo Archives: gentoo-user

From: tuxic@××××××.de
To: gentoo-user@l.g.o
Subject: Re: [gentoo-user] Misunderstanding fstrim...
Date: Sun, 26 Apr 2020 16:07:42
Message-Id: 20200426160731.sr3oly6bs4crjfy7@solfire
In Reply to: [gentoo-user] Misunderstanding fstrim... by Alan Mackenzie
1 On 04/26 03:26, Alan Mackenzie wrote:
2 > Hello, Gentoo.
3 >
4 > Seeing that I've had my "new" box for three years, now, I'm thinking
5 > it's about time to run fstrim. But how?
6 >
7 > I've got a pair of NVMe drives in a RAID 1 configuration. On the main
8 > partition, /dev/md126, I've got several LVM2 things. In the volume
9 > group(s?) I've got things like /dev/mapper/vg-home mounted on /home.
10 > Most of my partitions are ext4.
11 >
12 > When I try things like
13 >
14 > # fstrim --dry-run /home
15 >
16 > , I get the response:
17 >
18 > /home: 0 B (dry run) trimmed
19 >
20 > . What does this mean, in particular the "0 B" bit? Does it mean that
21 > if I let it rip, it would trim 0 Bytes?
22 >
23 > Will I do any damage if I let it go ahead (without the --dry-run)? Will
24 > this actually do any good? Will fstrim work together with LVM volumes?
25 >
26 > Thanks for any help!
27 >
28 > --
29 > Alan Mackenzie (Nuremberg, Germany).
30 >
31
32 Hi Alan,
33
34 NOW I am no longer feel alone with all my fstrim/SSD related
35 questions! :)
36
37 I exactly fall over this before.
38
39 It looks like fstrim would take its "dry run" really seriously
40 and does NOTHING - it even does not communicate with the SSD.
41 Therefore there is no responds from the SSD how much has been
42 trimmed...or "0 bytes" in other words.
43
44 If you start it with no "dry run", it will send the the SSD
45 informations, what parts, which still keep data, can be freed,
46 because this data is no longer linked to anything.
47
48 Fstrimming "does harm" ;) in the first place. It will cause
49 some kind of writing, which is -- in the purest sense of
50 the word -- "bad" to the SSD.
51
52 But fstrimming is good for the SSD in the more global context, since
53 all freed blocks now are available for wear leveling to the SSD
54 controller again.
55
56 I cannot say anything in the context of LVM, since I
57 don't use it.
58
59 But I don't *think* (read: don't know for sure), that
60 it would harm, since you are dealing with data, which
61 already has been flagged for "can be overwritte" by
62 the filesystem.
63
64 Since the SSD controller handles chunks of data one
65 level lower (the controller does not know anything
66 about filesystems), fstrim tells it, which data
67 can be overwritten from the filesystems point
68 of view.
69
70 BUT: Any undelete operation is made impossible
71 after a fstrim.
72
73 Some inside view to the status of the ssd can be taken
74 by this command:
75
76 sudo smartctl -a <your ssd device here>
77
78 This command only reads from the SSD.
79
80 Cheers!
81 Meino