1.. SPDX-License-Identifier: GPL-2.0 2 3============================ 4Glock internal locking rules 5============================ 6 7This documents the basic principles of the glock state machine 8internals. Each glock (struct gfs2_glock in fs/gfs2/incore.h) 9has two main (internal) locks: 10 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. 18 19The gl_holders list contains all the queued lock requests (not 20just the holders) associated with the glock. If there are any 21held locks, then they will be contiguous entries at the head 22of the list. Locks are granted in strictly the order that they 23are queued. 24 25There are three lock states that users of the glock layer can request, 26namely shared (SH), deferred (DF) and exclusive (EX). Those translate 27to the following DLM lock modes: 28 29========== ====== ===================================================== 30Glock 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========== ====== ===================================================== 37 38Thus DF is basically a shared mode which is incompatible with the "normal" 39shared lock mode, SH. In GFS2 the DF mode is used exclusively for direct I/O 40operations. The glocks are basically a lock plus some routines which deal 41with cache management. The following rules apply for the cache: 42 43========== ========== ============== ========== ============== 44Glock mode Cache data Cache Metadata Dirty Data Dirty Metadata 45========== ========== ============== ========== ============== 46 UN No No No No 47 SH Yes Yes No No 48 DF No Yes No No 49 EX Yes Yes Yes Yes 50========== ========== ============== ========== ============== 51 52These rules are implemented using the various glock operations which 53are defined for each type of glock. Not all types of glocks use 54all the modes. Only inode glocks use the DF mode for example. 55 56Table of glock operations and per type constants: 57 58============= ============================================================= 59Field Purpose 60============= ============================================================= 61go_xmote_th Called before remote state change (e.g. to sync dirty data) 62go_xmote_bh Called after remote state change (e.g. to refill cache) 63go_inval Called if remote state change requires invalidating the cache 64go_demote_ok Returns boolean value of whether its ok to demote a glock 65 (e.g. checks timeout, and that there is no cached data) 66go_lock Called for the first local holder of a lock 67go_unlock Called on the final local unlock of a lock 68go_dump Called to print content of object for debugfs file, or on 69 error to dump glock to the log. 70go_type The type of the glock, ``LM_TYPE_*`` 71go_callback Called if the DLM sends a callback to drop this lock 72go_flags GLOF_ASPACE is set, if the glock has an address space 73 associated with it 74============= ============================================================= 75 76The minimum hold time for each lock is the time after a remote lock 77grant for which we ignore remote demote requests. This is in order to 78prevent a situation where locks are being bounced around the cluster 79from node to node with none of the nodes making any progress. This 80tends to show up most with shared mmapped files which are being written 81to by multiple nodes. By delaying the demotion in response to a 82remote callback, that gives the userspace program time to make 83some progress before the pages are unmapped. 84 85There is a plan to try and remove the go_lock and go_unlock callbacks 86if possible, in order to try and speed up the fast path though the locking. 87Also, eventually we hope to make the glock "EX" mode locally shared 88such that any local locking will be done with the i_mutex as required 89rather than via the glock. 90 91Locking rules for glock operations: 92 93============= ====================== ============================= 94Operation GLF_LOCK bit lock held gl_lockref.lock spinlock held 95============= ====================== ============================= 96go_xmote_th Yes No 97go_xmote_bh Yes No 98go_inval Yes No 99go_demote_ok Sometimes Yes 100go_lock Yes No 101go_unlock Yes No 102go_dump Sometimes Yes 103go_callback Sometimes (N/A) Yes 104============= ====================== ============================= 105 106.. Note:: 107 108 Operations must not drop either the bit lock or the spinlock 109 if its held on entry. go_dump and do_demote_ok must never block. 110 Note that go_dump will only be called if the glock's state 111 indicates that it is caching uptodate data. 112 113Glock locking order within GFS2: 114 115 1. i_rwsem (if required) 116 2. Rename glock (for rename only) 117 3. Inode glock(s) 118 (Parents before children, inodes at "same level" with same parent in 119 lock number order) 120 4. Rgrp glock(s) (for (de)allocation operations) 121 5. Transaction glock (via gfs2_trans_begin) for non-read operations 122 6. i_rw_mutex (if required) 123 7. Page lock (always last, very important!) 124 125There are two glocks per inode. One deals with access to the inode 126itself (locking order as above), and the other, known as the iopen 127glock is used in conjunction with the i_nlink field in the inode to 128determine the lifetime of the inode in question. Locking of inodes 129is on a per-inode basis. Locking of rgrps is on a per rgrp basis. 130In general we prefer to lock local locks prior to cluster locks. 131 132Glock Statistics 133---------------- 134 135The stats are divided into two sets: those relating to the 136super block and those relating to an individual glock. The 137super block stats are done on a per cpu basis in order to 138try and reduce the overhead of gathering them. They are also 139further divided by glock type. All timings are in nanoseconds. 140 141In the case of both the super block and glock statistics, 142the same information is gathered in each case. The super 143block timing statistics are used to provide default values for 144the glock timing statistics, so that newly created glocks 145should have, as far as possible, a sensible starting point. 146The per-glock counters are initialised to zero when the 147glock is created. The per-glock statistics are lost when 148the glock is ejected from memory. 149 150The statistics are divided into three pairs of mean and 151variance, plus two counters. The mean/variance pairs are 152smoothed exponential estimates and the algorithm used is 153one which will be very familiar to those used to calculation 154of round trip times in network code. See "TCP/IP Illustrated, 155Volume 1", W. Richard Stevens, sect 21.3, "Round-Trip Time Measurement", 156p. 299 and onwards. Also, Volume 2, Sect. 25.10, p. 838 and onwards. 157Unlike the TCP/IP Illustrated case, the mean and variance are 158not scaled, but are in units of integer nanoseconds. 159 160The three pairs of mean/variance measure the following 161things: 162 163 1. DLM lock time (non-blocking requests) 164 2. DLM lock time (blocking requests) 165 3. Inter-request time (again to the DLM) 166 167A non-blocking request is one which will complete right 168away, whatever the state of the DLM lock in question. That 169currently means any requests when (a) the current state of 170the lock is exclusive, i.e. a lock demotion (b) the requested 171state is either null or unlocked (again, a demotion) or (c) the 172"try lock" flag is set. A blocking request covers all the other 173lock requests. 174 175There are two counters. The first is there primarily to show 176how many lock requests have been made, and thus how much data 177has gone into the mean/variance calculations. The other counter 178is counting queuing of holders at the top layer of the glock 179code. Hopefully that number will be a lot larger than the number 180of dlm lock requests issued. 181 182So why gather these statistics? There are several reasons 183we'd like to get a better idea of these timings: 184 1851. To be able to better set the glock "min hold time" 1862. To spot performance issues more easily 1873. To improve the algorithm for selecting resource groups for 188 allocation (to base it on lock wait time, rather than blindly 189 using a "try lock") 190 191Due to the smoothing action of the updates, a step change in 192some input quantity being sampled will only fully be taken 193into account after 8 samples (or 4 for the variance) and this 194needs to be carefully considered when interpreting the 195results. 196 197Knowing both the time it takes a lock request to complete and 198the average time between lock requests for a glock means we 199can compute the total percentage of the time for which the 200node is able to use a glock vs. time that the rest of the 201cluster has its share. That will be very useful when setting 202the lock min hold time. 203 204Great care has been taken to ensure that we 205measure exactly the quantities that we want, as accurately 206as possible. There are always inaccuracies in any 207measuring system, but I hope this is as accurate as we 208can reasonably make it. 209 210Per sb stats can be found here:: 211 212 /sys/kernel/debug/gfs2/<fsname>/sbstats 213 214Per glock stats can be found here:: 215 216 /sys/kernel/debug/gfs2/<fsname>/glstats 217 218Assuming that debugfs is mounted on /sys/kernel/debug and also 219that <fsname> is replaced with the name of the gfs2 filesystem 220in question. 221 222The abbreviations used in the output as are follows: 223 224========= ================================================================ 225srtt Smoothed round trip time for non blocking dlm requests 226srttvar Variance estimate for srtt 227srttb Smoothed round trip time for (potentially) blocking dlm requests 228srttvarb Variance estimate for srttb 229sirt Smoothed inter request time (for dlm requests) 230sirtvar Variance estimate for sirt 231dlm Number of dlm requests made (dcnt in glstats file) 232queue Number of glock requests queued (qcnt in glstats file) 233========= ================================================================ 234 235The sbstats file contains a set of these stats for each glock type (so 8 lines 236for each type) and for each cpu (one column per cpu). The glstats file contains 237a set of these stats for each glock in a similar format to the glocks file, but 238using the format mean/variance for each of the timing stats. 239 240The gfs2_glock_lock_time tracepoint prints out the current values of the stats 241for the glock in question, along with some addition information on each dlm 242reply that is received: 243 244====== ======================================= 245status The status of the dlm request 246flags The dlm request flags 247tdiff The time taken by this specific request 248====== ======================================= 249 250(remaining fields as per above list) 251 252 253