| blob | adc0d4c4d9798fa0725757e5dff752c1a98f09d8 |
1 .. SPDX-License-Identifier: GPL-2.0
3 ============================
4 Glock internal locking rules
5 ============================
7 This documents the basic principles of the glock state machine
8 internals. Each glock (struct gfs2_glock in fs/gfs2/incore.h)
9 has two main (internal) locks:
11 1. A spinlock (gl_lockref.lock) which protects the internal state such
12 as gl_state, gl_target and the list of holders (gl_holders)
13 2. A non-blocking bit lock, GLF_LOCK, which is used to prevent other
14 threads from making calls to the DLM, etc. at the same time. If a
15 thread takes this lock, it must then call run_queue (usually via the
16 workqueue) when it releases it in order to ensure any pending tasks
17 are completed.
19 The gl_holders list contains all the queued lock requests (not
20 just the holders) associated with the glock. If there are any
21 held locks, then they will be contiguous entries at the head
22 of the list. Locks are granted in strictly the order that they
23 are queued.
25 There are three lock states that users of the glock layer can request,
26 namely shared (SH), deferred (DF) and exclusive (EX). Those translate
27 to the following DLM lock modes:
29 ========== ====== =====================================================
30 Glock mode DLM lock mode
31 ========== ====== =====================================================
32 UN IV/NL Unlocked (no DLM lock associated with glock) or NL
33 SH PR (Protected read)
34 DF CW (Concurrent write)
35 EX EX (Exclusive)
36 ========== ====== =====================================================
38 Thus DF is basically a shared mode which is incompatible with the "normal"
39 shared lock mode, SH. In GFS2 the DF mode is used exclusively for direct I/O
40 operations. The glocks are basically a lock plus some routines which deal
41 with cache management. The following rules apply for the cache:
43 ========== ============== ========== ========== ==============
44 Glock mode Cache Metadata Cache data Dirty Data Dirty Metadata
45 ========== ============== ========== ========== ==============
46 UN No No No No
47 DF Yes No No No
48 SH Yes Yes No No
49 EX Yes Yes Yes Yes
50 ========== ============== ========== ========== ==============
52 These rules are implemented using the various glock operations which
53 are defined for each type of glock. Not all types of glocks use
54 all the modes. Only inode glocks use the DF mode for example.
56 Table of glock operations and per type constants:
58 ============== =============================================================
59 Field Purpose
60 ============== =============================================================
61 go_sync Called before remote state change (e.g. to sync dirty data)
62 go_xmote_bh Called after remote state change (e.g. to refill cache)
63 go_inval Called if remote state change requires invalidating the cache
64 go_instantiate Called when a glock has been acquired
65 go_held Called every time a glock holder is acquired
66 go_dump Called to print content of object for debugfs file, or on
67 error to dump glock to the log.
68 go_callback Called if the DLM sends a callback to drop this lock
69 go_unlocked Called when a glock is unlocked (dlm_unlock())
70 go_type The type of the glock, ``LM_TYPE_*``
71 go_flags GLOF_ASPACE is set, if the glock has an address space
72 associated with it
73 ============== =============================================================
75 The minimum hold time for each lock is the time after a remote lock
76 grant for which we ignore remote demote requests. This is in order to
77 prevent a situation where locks are being bounced around the cluster
78 from node to node with none of the nodes making any progress. This
79 tends to show up most with shared mmapped files which are being written
80 to by multiple nodes. By delaying the demotion in response to a
81 remote callback, that gives the userspace program time to make
82 some progress before the pages are unmapped.
84 Eventually, we hope to make the glock "EX" mode locally shared such that any
85 local locking will be done with the i_mutex as required rather than via the
86 glock.
88 Locking rules for glock operations:
90 ============== ====================== =============================
91 Operation GLF_LOCK bit lock held gl_lockref.lock spinlock held
92 ============== ====================== =============================
93 go_sync Yes No
94 go_xmote_bh Yes No
95 go_inval Yes No
96 go_instantiate No No
97 go_held No No
98 go_dump Sometimes Yes
99 go_callback Sometimes (N/A) Yes
100 go_unlocked Yes No
101 ============== ====================== =============================
103 .. Note::
105 Operations must not drop either the bit lock or the spinlock
106 if its held on entry. go_dump and do_demote_ok must never block.
107 Note that go_dump will only be called if the glock's state
108 indicates that it is caching uptodate data.
110 Glock locking order within GFS2:
112 1. i_rwsem (if required)
113 2. Rename glock (for rename only)
114 3. Inode glock(s)
115 (Parents before children, inodes at "same level" with same parent in
116 lock number order)
117 4. Rgrp glock(s) (for (de)allocation operations)
118 5. Transaction glock (via gfs2_trans_begin) for non-read operations
119 6. i_rw_mutex (if required)
120 7. Page lock (always last, very important!)
122 There are two glocks per inode. One deals with access to the inode
123 itself (locking order as above), and the other, known as the iopen
124 glock is used in conjunction with the i_nlink field in the inode to
125 determine the lifetime of the inode in question. Locking of inodes
126 is on a per-inode basis. Locking of rgrps is on a per rgrp basis.
127 In general we prefer to lock local locks prior to cluster locks.
129 Glock Statistics
130 ----------------
132 The stats are divided into two sets: those relating to the
133 super block and those relating to an individual glock. The
134 super block stats are done on a per cpu basis in order to
135 try and reduce the overhead of gathering them. They are also
136 further divided by glock type. All timings are in nanoseconds.
138 In the case of both the super block and glock statistics,
139 the same information is gathered in each case. The super
140 block timing statistics are used to provide default values for
141 the glock timing statistics, so that newly created glocks
142 should have, as far as possible, a sensible starting point.
143 The per-glock counters are initialised to zero when the
144 glock is created. The per-glock statistics are lost when
145 the glock is ejected from memory.
147 The statistics are divided into three pairs of mean and
148 variance, plus two counters. The mean/variance pairs are
149 smoothed exponential estimates and the algorithm used is
150 one which will be very familiar to those used to calculation
151 of round trip times in network code. See "TCP/IP Illustrated,
152 Volume 1", W. Richard Stevens, sect 21.3, "Round-Trip Time Measurement",
153 p. 299 and onwards. Also, Volume 2, Sect. 25.10, p. 838 and onwards.
154 Unlike the TCP/IP Illustrated case, the mean and variance are
155 not scaled, but are in units of integer nanoseconds.
157 The three pairs of mean/variance measure the following
158 things:
160 1. DLM lock time (non-blocking requests)
161 2. DLM lock time (blocking requests)
162 3. Inter-request time (again to the DLM)
164 A non-blocking request is one which will complete right
165 away, whatever the state of the DLM lock in question. That
166 currently means any requests when (a) the current state of
167 the lock is exclusive, i.e. a lock demotion (b) the requested
168 state is either null or unlocked (again, a demotion) or (c) the
169 "try lock" flag is set. A blocking request covers all the other
170 lock requests.
172 There are two counters. The first is there primarily to show
173 how many lock requests have been made, and thus how much data
174 has gone into the mean/variance calculations. The other counter
175 is counting queuing of holders at the top layer of the glock
176 code. Hopefully that number will be a lot larger than the number
177 of dlm lock requests issued.
179 So why gather these statistics? There are several reasons
180 we'd like to get a better idea of these timings:
182 1. To be able to better set the glock "min hold time"
183 2. To spot performance issues more easily
184 3. To improve the algorithm for selecting resource groups for
185 allocation (to base it on lock wait time, rather than blindly
186 using a "try lock")
188 Due to the smoothing action of the updates, a step change in
189 some input quantity being sampled will only fully be taken
190 into account after 8 samples (or 4 for the variance) and this
191 needs to be carefully considered when interpreting the
192 results.
194 Knowing both the time it takes a lock request to complete and
195 the average time between lock requests for a glock means we
196 can compute the total percentage of the time for which the
197 node is able to use a glock vs. time that the rest of the
198 cluster has its share. That will be very useful when setting
199 the lock min hold time.
201 Great care has been taken to ensure that we
202 measure exactly the quantities that we want, as accurately
203 as possible. There are always inaccuracies in any
204 measuring system, but I hope this is as accurate as we
205 can reasonably make it.
207 Per sb stats can be found here::
209 /sys/kernel/debug/gfs2/<fsname>/sbstats
211 Per glock stats can be found here::
213 /sys/kernel/debug/gfs2/<fsname>/glstats
215 Assuming that debugfs is mounted on /sys/kernel/debug and also
216 that <fsname> is replaced with the name of the gfs2 filesystem
217 in question.
219 The abbreviations used in the output as are follows:
221 ========= ================================================================
222 srtt Smoothed round trip time for non blocking dlm requests
223 srttvar Variance estimate for srtt
224 srttb Smoothed round trip time for (potentially) blocking dlm requests
225 srttvarb Variance estimate for srttb
226 sirt Smoothed inter request time (for dlm requests)
227 sirtvar Variance estimate for sirt
228 dlm Number of dlm requests made (dcnt in glstats file)
229 queue Number of glock requests queued (qcnt in glstats file)
230 ========= ================================================================
232 The sbstats file contains a set of these stats for each glock type (so 8 lines
233 for each type) and for each cpu (one column per cpu). The glstats file contains
234 a set of these stats for each glock in a similar format to the glocks file, but
235 using the format mean/variance for each of the timing stats.
237 The gfs2_glock_lock_time tracepoint prints out the current values of the stats
238 for the glock in question, along with some addition information on each dlm
239 reply that is received:
241 ====== =======================================
242 status The status of the dlm request
243 flags The dlm request flags
244 tdiff The time taken by this specific request
245 ====== =======================================
247 (remaining fields as per above list)