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 "io_uring.h"
13 #include "sqpoll.h"
14 #include "fdinfo.h"
15 #include "cancel.h"
16 #include "rsrc.h"
17
18 #ifdef CONFIG_PROC_FS
io_uring_show_cred(struct seq_file * m,unsigned int id,const struct cred * cred)19 static __cold int io_uring_show_cred(struct seq_file *m, unsigned int id,
20 const struct cred *cred)
21 {
22 struct user_namespace *uns = seq_user_ns(m);
23 struct group_info *gi;
24 kernel_cap_t cap;
25 int g;
26
27 seq_printf(m, "%5d\n", id);
28 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
29 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
30 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
31 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
32 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
33 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
34 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
35 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
36 seq_puts(m, "\n\tGroups:\t");
37 gi = cred->group_info;
38 for (g = 0; g < gi->ngroups; g++) {
39 seq_put_decimal_ull(m, g ? " " : "",
40 from_kgid_munged(uns, gi->gid[g]));
41 }
42 seq_puts(m, "\n\tCapEff:\t");
43 cap = cred->cap_effective;
44 seq_put_hex_ll(m, NULL, cap.val, 16);
45 seq_putc(m, '\n');
46 return 0;
47 }
48
49 #ifdef CONFIG_NET_RX_BUSY_POLL
common_tracking_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m,const char * tracking_strategy)50 static __cold void common_tracking_show_fdinfo(struct io_ring_ctx *ctx,
51 struct seq_file *m,
52 const char *tracking_strategy)
53 {
54 seq_puts(m, "NAPI:\tenabled\n");
55 seq_printf(m, "napi tracking:\t%s\n", tracking_strategy);
56 seq_printf(m, "napi_busy_poll_dt:\t%llu\n", ctx->napi_busy_poll_dt);
57 if (ctx->napi_prefer_busy_poll)
58 seq_puts(m, "napi_prefer_busy_poll:\ttrue\n");
59 else
60 seq_puts(m, "napi_prefer_busy_poll:\tfalse\n");
61 }
62
napi_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)63 static __cold void napi_show_fdinfo(struct io_ring_ctx *ctx,
64 struct seq_file *m)
65 {
66 unsigned int mode = READ_ONCE(ctx->napi_track_mode);
67
68 switch (mode) {
69 case IO_URING_NAPI_TRACKING_INACTIVE:
70 seq_puts(m, "NAPI:\tdisabled\n");
71 break;
72 case IO_URING_NAPI_TRACKING_DYNAMIC:
73 common_tracking_show_fdinfo(ctx, m, "dynamic");
74 break;
75 case IO_URING_NAPI_TRACKING_STATIC:
76 common_tracking_show_fdinfo(ctx, m, "static");
77 break;
78 default:
79 seq_printf(m, "NAPI:\tunknown mode (%u)\n", mode);
80 }
81 }
82 #else
napi_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)83 static inline void napi_show_fdinfo(struct io_ring_ctx *ctx,
84 struct seq_file *m)
85 {
86 }
87 #endif
88
__io_uring_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)89 static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
90 {
91 struct io_overflow_cqe *ocqe;
92 struct io_rings *r = ctx->rings;
93 struct rusage sq_usage;
94 unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
95 unsigned int sq_head = READ_ONCE(r->sq.head);
96 unsigned int sq_tail = READ_ONCE(r->sq.tail);
97 unsigned int cq_head = READ_ONCE(r->cq.head);
98 unsigned int cq_tail = READ_ONCE(r->cq.tail);
99 unsigned int cq_shift = 0;
100 unsigned int sq_shift = 0;
101 unsigned int sq_entries, cq_entries;
102 int sq_pid = -1, sq_cpu = -1;
103 u64 sq_total_time = 0, sq_work_time = 0;
104 unsigned int i;
105
106 if (ctx->flags & IORING_SETUP_CQE32)
107 cq_shift = 1;
108 if (ctx->flags & IORING_SETUP_SQE128)
109 sq_shift = 1;
110
111 /*
112 * we may get imprecise sqe and cqe info if uring is actively running
113 * since we get cached_sq_head and cached_cq_tail without uring_lock
114 * and sq_tail and cq_head are changed by userspace. But it's ok since
115 * we usually use these info when it is stuck.
116 */
117 seq_printf(m, "SqMask:\t0x%x\n", sq_mask);
118 seq_printf(m, "SqHead:\t%u\n", sq_head);
119 seq_printf(m, "SqTail:\t%u\n", sq_tail);
120 seq_printf(m, "CachedSqHead:\t%u\n", data_race(ctx->cached_sq_head));
121 seq_printf(m, "CqMask:\t0x%x\n", cq_mask);
122 seq_printf(m, "CqHead:\t%u\n", cq_head);
123 seq_printf(m, "CqTail:\t%u\n", cq_tail);
124 seq_printf(m, "CachedCqTail:\t%u\n", data_race(ctx->cached_cq_tail));
125 seq_printf(m, "SQEs:\t%u\n", sq_tail - sq_head);
126 sq_entries = min(sq_tail - sq_head, ctx->sq_entries);
127 for (i = 0; i < sq_entries; i++) {
128 unsigned int entry = i + sq_head;
129 struct io_uring_sqe *sqe;
130 unsigned int sq_idx;
131
132 if (ctx->flags & IORING_SETUP_NO_SQARRAY)
133 break;
134 sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]);
135 if (sq_idx > sq_mask)
136 continue;
137 sqe = &ctx->sq_sqes[sq_idx << sq_shift];
138 seq_printf(m, "%5u: opcode:%s, fd:%d, flags:%x, off:%llu, "
139 "addr:0x%llx, rw_flags:0x%x, buf_index:%d "
140 "user_data:%llu",
141 sq_idx, io_uring_get_opcode(sqe->opcode), sqe->fd,
142 sqe->flags, (unsigned long long) sqe->off,
143 (unsigned long long) sqe->addr, sqe->rw_flags,
144 sqe->buf_index, sqe->user_data);
145 if (sq_shift) {
146 u64 *sqeb = (void *) (sqe + 1);
147 int size = sizeof(struct io_uring_sqe) / sizeof(u64);
148 int j;
149
150 for (j = 0; j < size; j++) {
151 seq_printf(m, ", e%d:0x%llx", j,
152 (unsigned long long) *sqeb);
153 sqeb++;
154 }
155 }
156 seq_printf(m, "\n");
157 }
158 seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head);
159 cq_entries = min(cq_tail - cq_head, ctx->cq_entries);
160 for (i = 0; i < cq_entries; i++) {
161 unsigned int entry = i + cq_head;
162 struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift];
163
164 seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x",
165 entry & cq_mask, cqe->user_data, cqe->res,
166 cqe->flags);
167 if (cq_shift)
168 seq_printf(m, ", extra1:%llu, extra2:%llu\n",
169 cqe->big_cqe[0], cqe->big_cqe[1]);
170 seq_printf(m, "\n");
171 }
172
173 if (ctx->flags & IORING_SETUP_SQPOLL) {
174 struct io_sq_data *sq = ctx->sq_data;
175
176 /*
177 * sq->thread might be NULL if we raced with the sqpoll
178 * thread termination.
179 */
180 if (sq->thread) {
181 sq_pid = sq->task_pid;
182 sq_cpu = sq->sq_cpu;
183 getrusage(sq->thread, RUSAGE_SELF, &sq_usage);
184 sq_total_time = (sq_usage.ru_stime.tv_sec * 1000000
185 + sq_usage.ru_stime.tv_usec);
186 sq_work_time = sq->work_time;
187 }
188 }
189
190 seq_printf(m, "SqThread:\t%d\n", sq_pid);
191 seq_printf(m, "SqThreadCpu:\t%d\n", sq_cpu);
192 seq_printf(m, "SqTotalTime:\t%llu\n", sq_total_time);
193 seq_printf(m, "SqWorkTime:\t%llu\n", sq_work_time);
194 seq_printf(m, "UserFiles:\t%u\n", ctx->file_table.data.nr);
195 for (i = 0; i < ctx->file_table.data.nr; i++) {
196 struct file *f = NULL;
197
198 if (ctx->file_table.data.nodes[i])
199 f = io_slot_file(ctx->file_table.data.nodes[i]);
200 if (f) {
201 seq_printf(m, "%5u: ", i);
202 seq_file_path(m, f, " \t\n\\");
203 seq_puts(m, "\n");
204 }
205 }
206 seq_printf(m, "UserBufs:\t%u\n", ctx->buf_table.nr);
207 for (i = 0; i < ctx->buf_table.nr; i++) {
208 struct io_mapped_ubuf *buf = NULL;
209
210 if (ctx->buf_table.nodes[i])
211 buf = ctx->buf_table.nodes[i]->buf;
212 if (buf)
213 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, buf->len);
214 else
215 seq_printf(m, "%5u: <none>\n", i);
216 }
217 if (!xa_empty(&ctx->personalities)) {
218 unsigned long index;
219 const struct cred *cred;
220
221 seq_printf(m, "Personalities:\n");
222 xa_for_each(&ctx->personalities, index, cred)
223 io_uring_show_cred(m, index, cred);
224 }
225
226 seq_puts(m, "PollList:\n");
227 for (i = 0; i < (1U << ctx->cancel_table.hash_bits); i++) {
228 struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
229 struct io_kiocb *req;
230
231 hlist_for_each_entry(req, &hb->list, hash_node)
232 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
233 task_work_pending(req->tctx->task));
234 }
235
236 seq_puts(m, "CqOverflowList:\n");
237 spin_lock(&ctx->completion_lock);
238 list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
239 struct io_uring_cqe *cqe = &ocqe->cqe;
240
241 seq_printf(m, " user_data=%llu, res=%d, flags=%x\n",
242 cqe->user_data, cqe->res, cqe->flags);
243
244 }
245 spin_unlock(&ctx->completion_lock);
246 napi_show_fdinfo(ctx, m);
247 }
248
249 /*
250 * Caller holds a reference to the file already, we don't need to do
251 * anything else to get an extra reference.
252 */
io_uring_show_fdinfo(struct seq_file * m,struct file * file)253 __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *file)
254 {
255 struct io_ring_ctx *ctx = file->private_data;
256
257 /*
258 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
259 * since fdinfo case grabs it in the opposite direction of normal use
260 * cases.
261 */
262 if (mutex_trylock(&ctx->uring_lock)) {
263 __io_uring_show_fdinfo(ctx, m);
264 mutex_unlock(&ctx->uring_lock);
265 }
266 }
267 #endif
268