| blob | 7059623635b2d32fdaac02856447d6ab851eb04e |
1 Glock internal locking rules
2 ------------------------------
4 This documents the basic principles of the glock state machine
5 internals. Each glock (struct gfs2_glock in fs/gfs2/incore.h)
6 has two main (internal) locks:
8 1. A spinlock (gl_lockref.lock) which protects the internal state such
9 as gl_state, gl_target and the list of holders (gl_holders)
10 2. A non-blocking bit lock, GLF_LOCK, which is used to prevent other
11 threads from making calls to the DLM, etc. at the same time. If a
12 thread takes this lock, it must then call run_queue (usually via the
13 workqueue) when it releases it in order to ensure any pending tasks
14 are completed.
16 The gl_holders list contains all the queued lock requests (not
17 just the holders) associated with the glock. If there are any
18 held locks, then they will be contiguous entries at the head
19 of the list. Locks are granted in strictly the order that they
20 are queued, except for those marked LM_FLAG_PRIORITY which are
21 used only during recovery, and even then only for journal locks.
23 There are three lock states that users of the glock layer can request,
24 namely shared (SH), deferred (DF) and exclusive (EX). Those translate
25 to the following DLM lock modes:
27 Glock mode | DLM lock mode
28 ------------------------------
29 UN | IV/NL Unlocked (no DLM lock associated with glock) or NL
30 SH | PR (Protected read)
31 DF | CW (Concurrent write)
32 EX | EX (Exclusive)
34 Thus DF is basically a shared mode which is incompatible with the "normal"
35 shared lock mode, SH. In GFS2 the DF mode is used exclusively for direct I/O
36 operations. The glocks are basically a lock plus some routines which deal
37 with cache management. The following rules apply for the cache:
39 Glock mode | Cache data | Cache Metadata | Dirty Data | Dirty Metadata
40 --------------------------------------------------------------------------
41 UN | No | No | No | No
42 SH | Yes | Yes | No | No
43 DF | No | Yes | No | No
44 EX | Yes | Yes | Yes | Yes
46 These rules are implemented using the various glock operations which
47 are defined for each type of glock. Not all types of glocks use
48 all the modes. Only inode glocks use the DF mode for example.
50 Table of glock operations and per type constants:
52 Field | Purpose
53 ----------------------------------------------------------------------------
54 go_xmote_th | Called before remote state change (e.g. to sync dirty data)
55 go_xmote_bh | Called after remote state change (e.g. to refill cache)
56 go_inval | Called if remote state change requires invalidating the cache
57 go_demote_ok | Returns boolean value of whether its ok to demote a glock
58 | (e.g. checks timeout, and that there is no cached data)
59 go_lock | Called for the first local holder of a lock
60 go_unlock | Called on the final local unlock of a lock
61 go_dump | Called to print content of object for debugfs file, or on
62 | error to dump glock to the log.
63 go_type | The type of the glock, LM_TYPE_.....
64 go_callback | Called if the DLM sends a callback to drop this lock
65 go_flags | GLOF_ASPACE is set, if the glock has an address space
66 | associated with it
68 The minimum hold time for each lock is the time after a remote lock
69 grant for which we ignore remote demote requests. This is in order to
70 prevent a situation where locks are being bounced around the cluster
71 from node to node with none of the nodes making any progress. This
72 tends to show up most with shared mmaped files which are being written
73 to by multiple nodes. By delaying the demotion in response to a
74 remote callback, that gives the userspace program time to make
75 some progress before the pages are unmapped.
77 There is a plan to try and remove the go_lock and go_unlock callbacks
78 if possible, in order to try and speed up the fast path though the locking.
79 Also, eventually we hope to make the glock "EX" mode locally shared
80 such that any local locking will be done with the i_mutex as required
81 rather than via the glock.
83 Locking rules for glock operations:
85 Operation | GLF_LOCK bit lock held | gl_lockref.lock spinlock held
86 -------------------------------------------------------------------------
87 go_xmote_th | Yes | No
88 go_xmote_bh | Yes | No
89 go_inval | Yes | No
90 go_demote_ok | Sometimes | Yes
91 go_lock | Yes | No
92 go_unlock | Yes | No
93 go_dump | Sometimes | Yes
94 go_callback | Sometimes (N/A) | Yes
96 N.B. Operations must not drop either the bit lock or the spinlock
97 if its held on entry. go_dump and do_demote_ok must never block.
98 Note that go_dump will only be called if the glock's state
99 indicates that it is caching uptodate data.
101 Glock locking order within GFS2:
103 1. i_rwsem (if required)
104 2. Rename glock (for rename only)
105 3. Inode glock(s)
106 (Parents before children, inodes at "same level" with same parent in
107 lock number order)
108 4. Rgrp glock(s) (for (de)allocation operations)
109 5. Transaction glock (via gfs2_trans_begin) for non-read operations
110 6. i_rw_mutex (if required)
111 7. Page lock (always last, very important!)
113 There are two glocks per inode. One deals with access to the inode
114 itself (locking order as above), and the other, known as the iopen
115 glock is used in conjunction with the i_nlink field in the inode to
116 determine the lifetime of the inode in question. Locking of inodes
117 is on a per-inode basis. Locking of rgrps is on a per rgrp basis.
118 In general we prefer to lock local locks prior to cluster locks.
120 Glock Statistics
121 ------------------
123 The stats are divided into two sets: those relating to the
124 super block and those relating to an individual glock. The
125 super block stats are done on a per cpu basis in order to
126 try and reduce the overhead of gathering them. They are also
127 further divided by glock type. All timings are in nanoseconds.
129 In the case of both the super block and glock statistics,
130 the same information is gathered in each case. The super
131 block timing statistics are used to provide default values for
132 the glock timing statistics, so that newly created glocks
133 should have, as far as possible, a sensible starting point.
134 The per-glock counters are initialised to zero when the
135 glock is created. The per-glock statistics are lost when
136 the glock is ejected from memory.
138 The statistics are divided into three pairs of mean and
139 variance, plus two counters. The mean/variance pairs are
140 smoothed exponential estimates and the algorithm used is
141 one which will be very familiar to those used to calculation
142 of round trip times in network code. See "TCP/IP Illustrated,
143 Volume 1", W. Richard Stevens, sect 21.3, "Round-Trip Time Measurement",
144 p. 299 and onwards. Also, Volume 2, Sect. 25.10, p. 838 and onwards.
145 Unlike the TCP/IP Illustrated case, the mean and variance are
146 not scaled, but are in units of integer nanoseconds.
148 The three pairs of mean/variance measure the following
149 things:
151 1. DLM lock time (non-blocking requests)
152 2. DLM lock time (blocking requests)
153 3. Inter-request time (again to the DLM)
155 A non-blocking request is one which will complete right
156 away, whatever the state of the DLM lock in question. That
157 currently means any requests when (a) the current state of
158 the lock is exclusive, i.e. a lock demotion (b) the requested
159 state is either null or unlocked (again, a demotion) or (c) the
160 "try lock" flag is set. A blocking request covers all the other
161 lock requests.
163 There are two counters. The first is there primarily to show
164 how many lock requests have been made, and thus how much data
165 has gone into the mean/variance calculations. The other counter
166 is counting queuing of holders at the top layer of the glock
167 code. Hopefully that number will be a lot larger than the number
168 of dlm lock requests issued.
170 So why gather these statistics? There are several reasons
171 we'd like to get a better idea of these timings:
173 1. To be able to better set the glock "min hold time"
174 2. To spot performance issues more easily
175 3. To improve the algorithm for selecting resource groups for
176 allocation (to base it on lock wait time, rather than blindly
177 using a "try lock")
179 Due to the smoothing action of the updates, a step change in
180 some input quantity being sampled will only fully be taken
181 into account after 8 samples (or 4 for the variance) and this
182 needs to be carefully considered when interpreting the
183 results.
185 Knowing both the time it takes a lock request to complete and
186 the average time between lock requests for a glock means we
187 can compute the total percentage of the time for which the
188 node is able to use a glock vs. time that the rest of the
189 cluster has its share. That will be very useful when setting
190 the lock min hold time.
192 Great care has been taken to ensure that we
193 measure exactly the quantities that we want, as accurately
194 as possible. There are always inaccuracies in any
195 measuring system, but I hope this is as accurate as we
196 can reasonably make it.
198 Per sb stats can be found here:
199 /sys/kernel/debug/gfs2/<fsname>/sbstats
200 Per glock stats can be found here:
201 /sys/kernel/debug/gfs2/<fsname>/glstats
203 Assuming that debugfs is mounted on /sys/kernel/debug and also
204 that <fsname> is replaced with the name of the gfs2 filesystem
205 in question.
207 The abbreviations used in the output as are follows:
209 srtt - Smoothed round trip time for non-blocking dlm requests
210 srttvar - Variance estimate for srtt
211 srttb - Smoothed round trip time for (potentially) blocking dlm requests
212 srttvarb - Variance estimate for srttb
213 sirt - Smoothed inter-request time (for dlm requests)
214 sirtvar - Variance estimate for sirt
215 dlm - Number of dlm requests made (dcnt in glstats file)
216 queue - Number of glock requests queued (qcnt in glstats file)
218 The sbstats file contains a set of these stats for each glock type (so 8 lines
219 for each type) and for each cpu (one column per cpu). The glstats file contains
220 a set of these stats for each glock in a similar format to the glocks file, but
221 using the format mean/variance for each of the timing stats.
223 The gfs2_glock_lock_time tracepoint prints out the current values of the stats
224 for the glock in question, along with some addition information on each dlm
225 reply that is received:
227 status - The status of the dlm request
228 flags - The dlm request flags
229 tdiff - The time taken by this specific request
230 (remaining fields as per above list)