1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/errno.h>
4 #include <linux/fs.h>
5 #include <linux/file.h>
6 #include <linux/proc_fs.h>
7 #include <linux/seq_file.h>
8 #include <linux/io_uring.h>
9
10 #include <uapi/linux/io_uring.h>
11
12 #include "filetable.h"
13 #include "sqpoll.h"
14 #include "fdinfo.h"
15 #include "cancel.h"
16 #include "rsrc.h"
17
18 #ifdef CONFIG_NET_RX_BUSY_POLL
common_tracking_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m,const char * tracking_strategy)19 static __cold void common_tracking_show_fdinfo(struct io_ring_ctx *ctx,
20 struct seq_file *m,
21 const char *tracking_strategy)
22 {
23 seq_puts(m, "NAPI:\tenabled\n");
24 seq_printf(m, "napi tracking:\t%s\n", tracking_strategy);
25 seq_printf(m, "napi_busy_poll_dt:\t%llu\n", ctx->napi_busy_poll_dt);
26 if (ctx->napi_prefer_busy_poll)
27 seq_puts(m, "napi_prefer_busy_poll:\ttrue\n");
28 else
29 seq_puts(m, "napi_prefer_busy_poll:\tfalse\n");
30 }
31
napi_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)32 static __cold void napi_show_fdinfo(struct io_ring_ctx *ctx,
33 struct seq_file *m)
34 {
35 unsigned int mode = READ_ONCE(ctx->napi_track_mode);
36
37 switch (mode) {
38 case IO_URING_NAPI_TRACKING_INACTIVE:
39 seq_puts(m, "NAPI:\tdisabled\n");
40 break;
41 case IO_URING_NAPI_TRACKING_DYNAMIC:
42 common_tracking_show_fdinfo(ctx, m, "dynamic");
43 break;
44 case IO_URING_NAPI_TRACKING_STATIC:
45 common_tracking_show_fdinfo(ctx, m, "static");
46 break;
47 default:
48 seq_printf(m, "NAPI:\tunknown mode (%u)\n", mode);
49 }
50 }
51 #else
napi_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)52 static inline void napi_show_fdinfo(struct io_ring_ctx *ctx,
53 struct seq_file *m)
54 {
55 }
56 #endif
57
__io_uring_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)58 static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
59 {
60 struct io_overflow_cqe *ocqe;
61 struct io_rings *r = ctx->rings;
62 unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
63 unsigned int sq_head = READ_ONCE(r->sq.head);
64 unsigned int sq_tail = READ_ONCE(r->sq.tail);
65 unsigned int cq_head = READ_ONCE(r->cq.head);
66 unsigned int cq_tail = READ_ONCE(r->cq.tail);
67 unsigned int sq_shift = 0;
68 unsigned int sq_entries;
69 int sq_pid = -1, sq_cpu = -1;
70 u64 sq_total_time = 0, sq_work_time = 0;
71 unsigned int i;
72
73 if (ctx->flags & IORING_SETUP_SQE128)
74 sq_shift = 1;
75
76 /*
77 * we may get imprecise sqe and cqe info if uring is actively running
78 * since we get cached_sq_head and cached_cq_tail without uring_lock
79 * and sq_tail and cq_head are changed by userspace. But it's ok since
80 * we usually use these info when it is stuck.
81 */
82 seq_printf(m, "SqMask:\t0x%x\n", sq_mask);
83 seq_printf(m, "SqHead:\t%u\n", sq_head);
84 seq_printf(m, "SqTail:\t%u\n", sq_tail);
85 seq_printf(m, "CachedSqHead:\t%u\n", data_race(ctx->cached_sq_head));
86 seq_printf(m, "CqMask:\t0x%x\n", cq_mask);
87 seq_printf(m, "CqHead:\t%u\n", cq_head);
88 seq_printf(m, "CqTail:\t%u\n", cq_tail);
89 seq_printf(m, "CachedCqTail:\t%u\n", data_race(ctx->cached_cq_tail));
90 seq_printf(m, "SQEs:\t%u\n", sq_tail - sq_head);
91 sq_entries = min(sq_tail - sq_head, ctx->sq_entries);
92 for (i = 0; i < sq_entries; i++) {
93 unsigned int entry = i + sq_head;
94 struct io_uring_sqe *sqe;
95 unsigned int sq_idx;
96
97 if (ctx->flags & IORING_SETUP_NO_SQARRAY)
98 break;
99 sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]);
100 if (sq_idx > sq_mask)
101 continue;
102 sqe = &ctx->sq_sqes[sq_idx << sq_shift];
103 seq_printf(m, "%5u: opcode:%s, fd:%d, flags:%x, off:%llu, "
104 "addr:0x%llx, rw_flags:0x%x, buf_index:%d "
105 "user_data:%llu",
106 sq_idx, io_uring_get_opcode(sqe->opcode), sqe->fd,
107 sqe->flags, (unsigned long long) sqe->off,
108 (unsigned long long) sqe->addr, sqe->rw_flags,
109 sqe->buf_index, sqe->user_data);
110 if (sq_shift) {
111 u64 *sqeb = (void *) (sqe + 1);
112 int size = sizeof(struct io_uring_sqe) / sizeof(u64);
113 int j;
114
115 for (j = 0; j < size; j++) {
116 seq_printf(m, ", e%d:0x%llx", j,
117 (unsigned long long) *sqeb);
118 sqeb++;
119 }
120 }
121 seq_printf(m, "\n");
122 }
123 seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head);
124 while (cq_head < cq_tail) {
125 struct io_uring_cqe *cqe;
126 bool cqe32 = false;
127
128 cqe = &r->cqes[(cq_head & cq_mask)];
129 if (cqe->flags & IORING_CQE_F_32 || ctx->flags & IORING_SETUP_CQE32)
130 cqe32 = true;
131 seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x",
132 cq_head & cq_mask, cqe->user_data, cqe->res,
133 cqe->flags);
134 if (cqe32)
135 seq_printf(m, ", extra1:%llu, extra2:%llu\n",
136 cqe->big_cqe[0], cqe->big_cqe[1]);
137 seq_printf(m, "\n");
138 cq_head++;
139 if (cqe32)
140 cq_head++;
141 }
142
143 if (ctx->flags & IORING_SETUP_SQPOLL) {
144 struct io_sq_data *sq = ctx->sq_data;
145 struct task_struct *tsk;
146
147 rcu_read_lock();
148 tsk = rcu_dereference(sq->thread);
149 /*
150 * sq->thread might be NULL if we raced with the sqpoll
151 * thread termination.
152 */
153 if (tsk) {
154 u64 usec;
155
156 get_task_struct(tsk);
157 rcu_read_unlock();
158 usec = io_sq_cpu_usec(tsk);
159 put_task_struct(tsk);
160 sq_pid = sq->task_pid;
161 sq_cpu = sq->sq_cpu;
162 sq_total_time = usec;
163 sq_work_time = sq->work_time;
164 } else {
165 rcu_read_unlock();
166 }
167 }
168
169 seq_printf(m, "SqThread:\t%d\n", sq_pid);
170 seq_printf(m, "SqThreadCpu:\t%d\n", sq_cpu);
171 seq_printf(m, "SqTotalTime:\t%llu\n", sq_total_time);
172 seq_printf(m, "SqWorkTime:\t%llu\n", sq_work_time);
173 seq_printf(m, "UserFiles:\t%u\n", ctx->file_table.data.nr);
174 for (i = 0; i < ctx->file_table.data.nr; i++) {
175 struct file *f = NULL;
176
177 if (ctx->file_table.data.nodes[i])
178 f = io_slot_file(ctx->file_table.data.nodes[i]);
179 if (f) {
180 seq_printf(m, "%5u: ", i);
181 seq_file_path(m, f, " \t\n\\");
182 seq_puts(m, "\n");
183 }
184 }
185 seq_printf(m, "UserBufs:\t%u\n", ctx->buf_table.nr);
186 for (i = 0; i < ctx->buf_table.nr; i++) {
187 struct io_mapped_ubuf *buf = NULL;
188
189 if (ctx->buf_table.nodes[i])
190 buf = ctx->buf_table.nodes[i]->buf;
191 if (buf)
192 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, buf->len);
193 else
194 seq_printf(m, "%5u: <none>\n", i);
195 }
196
197 seq_puts(m, "PollList:\n");
198 for (i = 0; i < (1U << ctx->cancel_table.hash_bits); i++) {
199 struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
200 struct io_kiocb *req;
201
202 hlist_for_each_entry(req, &hb->list, hash_node)
203 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
204 task_work_pending(req->tctx->task));
205 }
206
207 seq_puts(m, "CqOverflowList:\n");
208 spin_lock(&ctx->completion_lock);
209 list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
210 struct io_uring_cqe *cqe = &ocqe->cqe;
211
212 seq_printf(m, " user_data=%llu, res=%d, flags=%x\n",
213 cqe->user_data, cqe->res, cqe->flags);
214
215 }
216 spin_unlock(&ctx->completion_lock);
217 napi_show_fdinfo(ctx, m);
218 }
219
220 /*
221 * Caller holds a reference to the file already, we don't need to do
222 * anything else to get an extra reference.
223 */
io_uring_show_fdinfo(struct seq_file * m,struct file * file)224 __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *file)
225 {
226 struct io_ring_ctx *ctx = file->private_data;
227
228 /*
229 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
230 * since fdinfo case grabs it in the opposite direction of normal use
231 * cases.
232 */
233 if (mutex_trylock(&ctx->uring_lock)) {
234 __io_uring_show_fdinfo(ctx, m);
235 mutex_unlock(&ctx->uring_lock);
236 }
237 }
238