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, except for those marked LM_FLAG_PRIORITY which are 24used only during recovery, and even then only for journal locks. 25 26There are three lock states that users of the glock layer can request, 27namely shared (SH), deferred (DF) and exclusive (EX). Those translate 28to the following DLM lock modes: 29 30========== ====== ===================================================== 31Glock mode DLM lock mode 32========== ====== ===================================================== 33 UN IV/NL Unlocked (no DLM lock associated with glock) or NL 34 SH PR (Protected read) 35 DF CW (Concurrent write) 36 EX EX (Exclusive) 37========== ====== ===================================================== 38 39Thus DF is basically a shared mode which is incompatible with the "normal" 40shared lock mode, SH. In GFS2 the DF mode is used exclusively for direct I/O 41operations. The glocks are basically a lock plus some routines which deal 42with cache management. The following rules apply for the cache: 43 44========== ========== ============== ========== ============== 45Glock mode Cache data Cache Metadata Dirty Data Dirty Metadata 46========== ========== ============== ========== ============== 47 UN No No No No 48 SH Yes Yes No No 49 DF No Yes No No 50 EX Yes Yes Yes Yes 51========== ========== ============== ========== ============== 52 53These rules are implemented using the various glock operations which 54are defined for each type of glock. Not all types of glocks use 55all the modes. Only inode glocks use the DF mode for example. 56 57Table of glock operations and per type constants: 58 59============= ============================================================= 60Field Purpose 61============= ============================================================= 62go_xmote_th Called before remote state change (e.g. to sync dirty data) 63go_xmote_bh Called after remote state change (e.g. to refill cache) 64go_inval Called if remote state change requires invalidating the cache 65go_demote_ok Returns boolean value of whether its ok to demote a glock 66 (e.g. checks timeout, and that there is no cached data) 67go_lock Called for the first local holder of a lock 68go_unlock Called on the final local unlock of a lock 69go_dump Called to print content of object for debugfs file, or on 70 error to dump glock to the log. 71go_type The type of the glock, ``LM_TYPE_*`` 72go_callback Called if the DLM sends a callback to drop this lock 73go_flags GLOF_ASPACE is set, if the glock has an address space 74 associated with it 75============= ============================================================= 76 77The minimum hold time for each lock is the time after a remote lock 78grant for which we ignore remote demote requests. This is in order to 79prevent a situation where locks are being bounced around the cluster 80from node to node with none of the nodes making any progress. This 81tends to show up most with shared mmapped files which are being written 82to by multiple nodes. By delaying the demotion in response to a 83remote callback, that gives the userspace program time to make 84some progress before the pages are unmapped. 85 86There is a plan to try and remove the go_lock and go_unlock callbacks 87if possible, in order to try and speed up the fast path though the locking. 88Also, eventually we hope to make the glock "EX" mode locally shared 89such that any local locking will be done with the i_mutex as required 90rather than via the glock. 91 92Locking rules for glock operations: 93 94============= ====================== ============================= 95Operation GLF_LOCK bit lock held gl_lockref.lock spinlock held 96============= ====================== ============================= 97go_xmote_th Yes No 98go_xmote_bh Yes No 99go_inval Yes No 100go_demote_ok Sometimes Yes 101go_lock Yes No 102go_unlock Yes No 103go_dump Sometimes Yes 104go_callback Sometimes (N/A) Yes 105============= ====================== ============================= 106 107.. Note:: 108 109 Operations must not drop either the bit lock or the spinlock 110 if its held on entry. go_dump and do_demote_ok must never block. 111 Note that go_dump will only be called if the glock's state 112 indicates that it is caching uptodate data. 113 114Glock locking order within GFS2: 115 116 1. i_rwsem (if required) 117 2. Rename glock (for rename only) 118 3. Inode glock(s) 119 (Parents before children, inodes at "same level" with same parent in 120 lock number order) 121 4. Rgrp glock(s) (for (de)allocation operations) 122 5. Transaction glock (via gfs2_trans_begin) for non-read operations 123 6. i_rw_mutex (if required) 124 7. Page lock (always last, very important!) 125 126There are two glocks per inode. One deals with access to the inode 127itself (locking order as above), and the other, known as the iopen 128glock is used in conjunction with the i_nlink field in the inode to 129determine the lifetime of the inode in question. Locking of inodes 130is on a per-inode basis. Locking of rgrps is on a per rgrp basis. 131In general we prefer to lock local locks prior to cluster locks. 132 133Glock Statistics 134---------------- 135 136The stats are divided into two sets: those relating to the 137super block and those relating to an individual glock. The 138super block stats are done on a per cpu basis in order to 139try and reduce the overhead of gathering them. They are also 140further divided by glock type. All timings are in nanoseconds. 141 142In the case of both the super block and glock statistics, 143the same information is gathered in each case. The super 144block timing statistics are used to provide default values for 145the glock timing statistics, so that newly created glocks 146should have, as far as possible, a sensible starting point. 147The per-glock counters are initialised to zero when the 148glock is created. The per-glock statistics are lost when 149the glock is ejected from memory. 150 151The statistics are divided into three pairs of mean and 152variance, plus two counters. The mean/variance pairs are 153smoothed exponential estimates and the algorithm used is 154one which will be very familiar to those used to calculation 155of round trip times in network code. See "TCP/IP Illustrated, 156Volume 1", W. Richard Stevens, sect 21.3, "Round-Trip Time Measurement", 157p. 299 and onwards. Also, Volume 2, Sect. 25.10, p. 838 and onwards. 158Unlike the TCP/IP Illustrated case, the mean and variance are 159not scaled, but are in units of integer nanoseconds. 160 161The three pairs of mean/variance measure the following 162things: 163 164 1. DLM lock time (non-blocking requests) 165 2. DLM lock time (blocking requests) 166 3. Inter-request time (again to the DLM) 167 168A non-blocking request is one which will complete right 169away, whatever the state of the DLM lock in question. That 170currently means any requests when (a) the current state of 171the lock is exclusive, i.e. a lock demotion (b) the requested 172state is either null or unlocked (again, a demotion) or (c) the 173"try lock" flag is set. A blocking request covers all the other 174lock requests. 175 176There are two counters. The first is there primarily to show 177how many lock requests have been made, and thus how much data 178has gone into the mean/variance calculations. The other counter 179is counting queuing of holders at the top layer of the glock 180code. Hopefully that number will be a lot larger than the number 181of dlm lock requests issued. 182 183So why gather these statistics? There are several reasons 184we'd like to get a better idea of these timings: 185 1861. To be able to better set the glock "min hold time" 1872. To spot performance issues more easily 1883. To improve the algorithm for selecting resource groups for 189 allocation (to base it on lock wait time, rather than blindly 190 using a "try lock") 191 192Due to the smoothing action of the updates, a step change in 193some input quantity being sampled will only fully be taken 194into account after 8 samples (or 4 for the variance) and this 195needs to be carefully considered when interpreting the 196results. 197 198Knowing both the time it takes a lock request to complete and 199the average time between lock requests for a glock means we 200can compute the total percentage of the time for which the 201node is able to use a glock vs. time that the rest of the 202cluster has its share. That will be very useful when setting 203the lock min hold time. 204 205Great care has been taken to ensure that we 206measure exactly the quantities that we want, as accurately 207as possible. There are always inaccuracies in any 208measuring system, but I hope this is as accurate as we 209can reasonably make it. 210 211Per sb stats can be found here:: 212 213 /sys/kernel/debug/gfs2/<fsname>/sbstats 214 215Per glock stats can be found here:: 216 217 /sys/kernel/debug/gfs2/<fsname>/glstats 218 219Assuming that debugfs is mounted on /sys/kernel/debug and also 220that <fsname> is replaced with the name of the gfs2 filesystem 221in question. 222 223The abbreviations used in the output as are follows: 224 225========= ================================================================ 226srtt Smoothed round trip time for non blocking dlm requests 227srttvar Variance estimate for srtt 228srttb Smoothed round trip time for (potentially) blocking dlm requests 229srttvarb Variance estimate for srttb 230sirt Smoothed inter request time (for dlm requests) 231sirtvar Variance estimate for sirt 232dlm Number of dlm requests made (dcnt in glstats file) 233queue Number of glock requests queued (qcnt in glstats file) 234========= ================================================================ 235 236The sbstats file contains a set of these stats for each glock type (so 8 lines 237for each type) and for each cpu (one column per cpu). The glstats file contains 238a set of these stats for each glock in a similar format to the glocks file, but 239using the format mean/variance for each of the timing stats. 240 241The gfs2_glock_lock_time tracepoint prints out the current values of the stats 242for the glock in question, along with some addition information on each dlm 243reply that is received: 244 245====== ======================================= 246status The status of the dlm request 247flags The dlm request flags 248tdiff The time taken by this specific request 249====== ======================================= 250 251(remaining fields as per above list) 252 253 254