1=====================
2The errseq_t datatype
3=====================
4
5An errseq_t is a way of recording errors in one place, and allowing any
6number of "subscribers" to tell whether it has changed since a previous
7point where it was sampled.
8
9The initial use case for this is tracking errors for file
10synchronization syscalls (fsync, fdatasync, msync and sync_file_range),
11but it may be usable in other situations.
12
13It's implemented as an unsigned 32-bit value.  The low order bits are
14designated to hold an error code (between 1 and MAX_ERRNO).  The upper bits
15are used as a counter.  This is done with atomics instead of locking so that
16these functions can be called from any context.
17
18Note that there is a risk of collisions if new errors are being recorded
19frequently, since we have so few bits to use as a counter.
20
21To mitigate this, the bit between the error value and counter is used as
22a flag to tell whether the value has been sampled since a new value was
23recorded.  That allows us to avoid bumping the counter if no one has
24sampled it since the last time an error was recorded.
25
26Thus we end up with a value that looks something like this:
27
28+--------------------------------------+----+------------------------+
29| 31..13                               | 12 | 11..0                  |
30+--------------------------------------+----+------------------------+
31| counter                              | SF | errno                  |
32+--------------------------------------+----+------------------------+
33
34The general idea is for "watchers" to sample an errseq_t value and keep
35it as a running cursor.  That value can later be used to tell whether
36any new errors have occurred since that sampling was done, and atomically
37record the state at the time that it was checked.  This allows us to
38record errors in one place, and then have a number of "watchers" that
39can tell whether the value has changed since they last checked it.
40
41A new errseq_t should always be zeroed out.  An errseq_t value of all zeroes
42is the special (but common) case where there has never been an error. An all
43zero value thus serves as the "epoch" if one wishes to know whether there
44has ever been an error set since it was first initialized.
45
46API usage
47=========
48
49Let me tell you a story about a worker drone.  Now, he's a good worker
50overall, but the company is a little...management heavy.  He has to
51report to 77 supervisors today, and tomorrow the "big boss" is coming in
52from out of town and he's sure to test the poor fellow too.
53
54They're all handing him work to do -- so much he can't keep track of who
55handed him what, but that's not really a big problem.  The supervisors
56just want to know when he's finished all of the work they've handed him so
57far and whether he made any mistakes since they last asked.
58
59He might have made the mistake on work they didn't actually hand him,
60but he can't keep track of things at that level of detail, all he can
61remember is the most recent mistake that he made.
62
63Here's our worker_drone representation::
64
65        struct worker_drone {
66                errseq_t        wd_err; /* for recording errors */
67        };
68
69Every day, the worker_drone starts out with a blank slate::
70
71        struct worker_drone wd;
72
73        wd.wd_err = (errseq_t)0;
74
75The supervisors come in and get an initial read for the day.  They
76don't care about anything that happened before their watch begins::
77
78        struct supervisor {
79                errseq_t        s_wd_err; /* private "cursor" for wd_err */
80                spinlock_t      s_wd_err_lock; /* protects s_wd_err */
81        }
82
83        struct supervisor       su;
84
85        su.s_wd_err = errseq_sample(&wd.wd_err);
86        spin_lock_init(&su.s_wd_err_lock);
87
88Now they start handing him tasks to do.  Every few minutes they ask him to
89finish up all of the work they've handed him so far.  Then they ask him
90whether he made any mistakes on any of it::
91
92        spin_lock(&su.su_wd_err_lock);
93        err = errseq_check_and_advance(&wd.wd_err, &su.s_wd_err);
94        spin_unlock(&su.su_wd_err_lock);
95
96Up to this point, that just keeps returning 0.
97
98Now, the owners of this company are quite miserly and have given him
99substandard equipment with which to do his job. Occasionally it
100glitches and he makes a mistake.  He sighs a heavy sigh, and marks it
101down::
102
103        errseq_set(&wd.wd_err, -EIO);
104
105...and then gets back to work.  The supervisors eventually poll again
106and they each get the error when they next check.  Subsequent calls will
107return 0, until another error is recorded, at which point it's reported
108to each of them once.
109
110Note that the supervisors can't tell how many mistakes he made, only
111whether one was made since they last checked, and the latest value
112recorded.
113
114Occasionally the big boss comes in for a spot check and asks the worker
115to do a one-off job for him. He's not really watching the worker
116full-time like the supervisors, but he does need to know whether a
117mistake occurred while his job was processing.
118
119He can just sample the current errseq_t in the worker, and then use that
120to tell whether an error has occurred later::
121
122        errseq_t since = errseq_sample(&wd.wd_err);
123        /* submit some work and wait for it to complete */
124        err = errseq_check(&wd.wd_err, since);
125
126Since he's just going to discard "since" after that point, he doesn't
127need to advance it here. He also doesn't need any locking since it's
128not usable by anyone else.
129
130Serializing errseq_t cursor updates
131===================================
132
133Note that the errseq_t API does not protect the errseq_t cursor during a
134check_and_advance_operation. Only the canonical error code is handled
135atomically.  In a situation where more than one task might be using the
136same errseq_t cursor at the same time, it's important to serialize
137updates to that cursor.
138
139If that's not done, then it's possible for the cursor to go backward
140in which case the same error could be reported more than once.
141
142Because of this, it's often advantageous to first do an errseq_check to
143see if anything has changed, and only later do an
144errseq_check_and_advance after taking the lock. e.g.::
145
146        if (errseq_check(&wd.wd_err, READ_ONCE(su.s_wd_err)) {
147                /* su.s_wd_err is protected by s_wd_err_lock */
148                spin_lock(&su.s_wd_err_lock);
149                err = errseq_check_and_advance(&wd.wd_err, &su.s_wd_err);
150                spin_unlock(&su.s_wd_err_lock);
151        }
152
153That avoids the spinlock in the common case where nothing has changed
154since the last time it was checked.
155
156Functions
157=========
158
159.. kernel-doc:: lib/errseq.c
160