1// SPDX-License-Identifier: GPL-2.0
2#include <linux/err.h>
3#include <linux/bug.h>
4#include <linux/atomic.h>
5#include <linux/errseq.h>
6#include <linux/log2.h>
7
8/*
9 * An errseq_t is a way of recording errors in one place, and allowing any
10 * number of "subscribers" to tell whether it has changed since a previous
11 * point where it was sampled.
12 *
13 * It's implemented as an unsigned 32-bit value. The low order bits are
14 * designated to hold an error code (between 0 and -MAX_ERRNO). The upper bits
15 * are used as a counter. This is done with atomics instead of locking so that
16 * these functions can be called from any context.
17 *
18 * The general idea is for consumers to sample an errseq_t value. That value
19 * can later be used to tell whether any new errors have occurred since that
20 * sampling was done.
21 *
22 * Note that there is a risk of collisions if new errors are being recorded
23 * frequently, since we have so few bits to use as a counter.
24 *
25 * To mitigate this, one bit is used as a flag to tell whether the value has
26 * been sampled since a new value was recorded. That allows us to avoid bumping
27 * the counter if no one has sampled it since the last time an error was
28 * recorded.
29 *
30 * A new errseq_t should always be zeroed out. A errseq_t value of all zeroes
31 * is the special (but common) case where there has never been an error. An all
32 * zero value thus serves as the "epoch" if one wishes to know whether there
33 * has ever been an error set since it was first initialized.
34 */
35
36/* The low bits are designated for error code (max of MAX_ERRNO) */
37#define ERRSEQ_SHIFT (ilog2(MAX_ERRNO) + 1)
38
39/* This bit is used as a flag to indicate whether the value has been seen */
40#define ERRSEQ_SEEN (1 << ERRSEQ_SHIFT)
41
42/* Leverage macro ERRSEQ_SEEN to define errno mask macro here */
43#define ERRNO_MASK (ERRSEQ_SEEN - 1)
44
45/* The lowest bit of the counter */
46#define ERRSEQ_CTR_INC (1 << (ERRSEQ_SHIFT + 1))
47
48/**
49 * errseq_set - set a errseq_t for later reporting
50 * @eseq: errseq_t field that should be set
51 * @err: error to set (must be between -1 and -MAX_ERRNO)
52 *
53 * This function sets the error in @eseq, and increments the sequence counter
54 * if the last sequence was sampled at some point in the past.
55 *
56 * Any error set will always overwrite an existing error.
57 *
58 * Return: The previous value, primarily for debugging purposes. The
59 * return value should not be used as a previously sampled value in later
60 * calls as it will not have the SEEN flag set.
61 */
62errseq_t errseq_set(errseq_t *eseq, int err)
63{
64 errseq_t cur, old;
65
66
67 /*
68 * Ensure the error code actually fits where we want it to go. If it
69 * doesn't then just throw a warning and don't record anything. We
70 * also don't accept zero here as that would effectively clear a
71 * previous error.
72 */
73 old = READ_ONCE(*eseq);
74
75 if (WARN(unlikely(err == 0 || (unsigned int)-err > MAX_ERRNO),
76 "err = %d\n", err))
77 return old;
78
79 for (;;) {
80 errseq_t new;
81
82 /* Clear out error bits and set new error */
83 new = (old & ~(ERRNO_MASK | ERRSEQ_SEEN)) | -err;
84
85 /* Only increment if someone has looked at it */
86 if (old & ERRSEQ_SEEN)
87 new += ERRSEQ_CTR_INC;
88
89 /* If there would be no change, then call it done */
90 if (new == old) {
91 cur = new;
92 break;
93 }
94
95 /* Try to swap the new value into place */
96 cur = cmpxchg(eseq, old, new);
97
98 /*
99 * Call it success if we did the swap or someone else beat us
100 * to it for the same value.
101 */
102 if (likely(cur == old || cur == new))
103 break;
104
105 /* Raced with an update, try again */
106 old = cur;
107 }
108 return cur;
109}
110EXPORT_SYMBOL(errseq_set);
111
112/**
113 * errseq_sample() - Grab current errseq_t value.
114 * @eseq: Pointer to errseq_t to be sampled.
115 *
116 * This function allows callers to initialise their errseq_t variable.
117 * If the error has been "seen", new callers will not see an old error.
118 * If there is an unseen error in @eseq, the caller of this function will
119 * see it the next time it checks for an error.
120 *
121 * Context: Any context.
122 * Return: The current errseq value.
123 */
124errseq_t errseq_sample(errseq_t *eseq)
125{
126 errseq_t old = READ_ONCE(*eseq);
127
128 /* If nobody has seen this error yet, then we can be the first. */
129 if (!(old & ERRSEQ_SEEN))
130 old = 0;
131 return old;
132}
133EXPORT_SYMBOL(errseq_sample);
134
135/**
136 * errseq_check() - Has an error occurred since a particular sample point?
137 * @eseq: Pointer to errseq_t value to be checked.
138 * @since: Previously-sampled errseq_t from which to check.
139 *
140 * Grab the value that eseq points to, and see if it has changed @since
141 * the given value was sampled. The @since value is not advanced, so there
142 * is no need to mark the value as seen.
143 *
144 * Return: The latest error set in the errseq_t or 0 if it hasn't changed.
145 */
146int errseq_check(errseq_t *eseq, errseq_t since)
147{
148 errseq_t cur = READ_ONCE(*eseq);
149
150 if (likely(cur == since))
151 return 0;
152 return -(cur & ERRNO_MASK);
153}
154EXPORT_SYMBOL(errseq_check);
155
156/**
157 * errseq_check_and_advance() - Check an errseq_t and advance to current value.
158 * @eseq: Pointer to value being checked and reported.
159 * @since: Pointer to previously-sampled errseq_t to check against and advance.
160 *
161 * Grab the eseq value, and see whether it matches the value that @since
162 * points to. If it does, then just return 0.
163 *
164 * If it doesn't, then the value has changed. Set the "seen" flag, and try to
165 * swap it into place as the new eseq value. Then, set that value as the new
166 * "since" value, and return whatever the error portion is set to.
167 *
168 * Note that no locking is provided here for concurrent updates to the "since"
169 * value. The caller must provide that if necessary. Because of this, callers
170 * may want to do a lockless errseq_check before taking the lock and calling
171 * this.
172 *
173 * Return: Negative errno if one has been stored, or 0 if no new error has
174 * occurred.
175 */
176int errseq_check_and_advance(errseq_t *eseq, errseq_t *since)
177{
178 int err = 0;
179 errseq_t old, new;
180
181 /*
182 * Most callers will want to use the inline wrapper to check this,
183 * so that the common case of no error is handled without needing
184 * to take the lock that protects the "since" value.
185 */
186 old = READ_ONCE(*eseq);
187 if (old != *since) {
188 /*
189 * Set the flag and try to swap it into place if it has
190 * changed.
191 *
192 * We don't care about the outcome of the swap here. If the
193 * swap doesn't occur, then it has either been updated by a
194 * writer who is altering the value in some way (updating
195 * counter or resetting the error), or another reader who is
196 * just setting the "seen" flag. Either outcome is OK, and we
197 * can advance "since" and return an error based on what we
198 * have.
199 */
200 new = old | ERRSEQ_SEEN;
201 if (new != old)
202 cmpxchg(eseq, old, new);
203 *since = new;
204 err = -(new & ERRNO_MASK);
205 }
206 return err;
207}
208EXPORT_SYMBOL(errseq_check_and_advance);
209