xref: /linux/io_uring/fdinfo.c (revision 3fd6c59042dbba50391e30862beac979491145fe)
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 
89 /*
90  * Caller holds a reference to the file already, we don't need to do
91  * anything else to get an extra reference.
92  */
io_uring_show_fdinfo(struct seq_file * m,struct file * file)93 __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *file)
94 {
95 	struct io_ring_ctx *ctx = file->private_data;
96 	struct io_overflow_cqe *ocqe;
97 	struct io_rings *r = ctx->rings;
98 	struct rusage sq_usage;
99 	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
100 	unsigned int sq_head = READ_ONCE(r->sq.head);
101 	unsigned int sq_tail = READ_ONCE(r->sq.tail);
102 	unsigned int cq_head = READ_ONCE(r->cq.head);
103 	unsigned int cq_tail = READ_ONCE(r->cq.tail);
104 	unsigned int cq_shift = 0;
105 	unsigned int sq_shift = 0;
106 	unsigned int sq_entries, cq_entries;
107 	int sq_pid = -1, sq_cpu = -1;
108 	u64 sq_total_time = 0, sq_work_time = 0;
109 	bool has_lock;
110 	unsigned int i;
111 
112 	if (ctx->flags & IORING_SETUP_CQE32)
113 		cq_shift = 1;
114 	if (ctx->flags & IORING_SETUP_SQE128)
115 		sq_shift = 1;
116 
117 	/*
118 	 * we may get imprecise sqe and cqe info if uring is actively running
119 	 * since we get cached_sq_head and cached_cq_tail without uring_lock
120 	 * and sq_tail and cq_head are changed by userspace. But it's ok since
121 	 * we usually use these info when it is stuck.
122 	 */
123 	seq_printf(m, "SqMask:\t0x%x\n", sq_mask);
124 	seq_printf(m, "SqHead:\t%u\n", sq_head);
125 	seq_printf(m, "SqTail:\t%u\n", sq_tail);
126 	seq_printf(m, "CachedSqHead:\t%u\n", ctx->cached_sq_head);
127 	seq_printf(m, "CqMask:\t0x%x\n", cq_mask);
128 	seq_printf(m, "CqHead:\t%u\n", cq_head);
129 	seq_printf(m, "CqTail:\t%u\n", cq_tail);
130 	seq_printf(m, "CachedCqTail:\t%u\n", ctx->cached_cq_tail);
131 	seq_printf(m, "SQEs:\t%u\n", sq_tail - sq_head);
132 	sq_entries = min(sq_tail - sq_head, ctx->sq_entries);
133 	for (i = 0; i < sq_entries; i++) {
134 		unsigned int entry = i + sq_head;
135 		struct io_uring_sqe *sqe;
136 		unsigned int sq_idx;
137 
138 		if (ctx->flags & IORING_SETUP_NO_SQARRAY)
139 			break;
140 		sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]);
141 		if (sq_idx > sq_mask)
142 			continue;
143 		sqe = &ctx->sq_sqes[sq_idx << sq_shift];
144 		seq_printf(m, "%5u: opcode:%s, fd:%d, flags:%x, off:%llu, "
145 			      "addr:0x%llx, rw_flags:0x%x, buf_index:%d "
146 			      "user_data:%llu",
147 			   sq_idx, io_uring_get_opcode(sqe->opcode), sqe->fd,
148 			   sqe->flags, (unsigned long long) sqe->off,
149 			   (unsigned long long) sqe->addr, sqe->rw_flags,
150 			   sqe->buf_index, sqe->user_data);
151 		if (sq_shift) {
152 			u64 *sqeb = (void *) (sqe + 1);
153 			int size = sizeof(struct io_uring_sqe) / sizeof(u64);
154 			int j;
155 
156 			for (j = 0; j < size; j++) {
157 				seq_printf(m, ", e%d:0x%llx", j,
158 						(unsigned long long) *sqeb);
159 				sqeb++;
160 			}
161 		}
162 		seq_printf(m, "\n");
163 	}
164 	seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head);
165 	cq_entries = min(cq_tail - cq_head, ctx->cq_entries);
166 	for (i = 0; i < cq_entries; i++) {
167 		unsigned int entry = i + cq_head;
168 		struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift];
169 
170 		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x",
171 			   entry & cq_mask, cqe->user_data, cqe->res,
172 			   cqe->flags);
173 		if (cq_shift)
174 			seq_printf(m, ", extra1:%llu, extra2:%llu\n",
175 					cqe->big_cqe[0], cqe->big_cqe[1]);
176 		seq_printf(m, "\n");
177 	}
178 
179 	/*
180 	 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
181 	 * since fdinfo case grabs it in the opposite direction of normal use
182 	 * cases. If we fail to get the lock, we just don't iterate any
183 	 * structures that could be going away outside the io_uring mutex.
184 	 */
185 	has_lock = mutex_trylock(&ctx->uring_lock);
186 
187 	if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
188 		struct io_sq_data *sq = ctx->sq_data;
189 
190 		/*
191 		 * sq->thread might be NULL if we raced with the sqpoll
192 		 * thread termination.
193 		 */
194 		if (sq->thread) {
195 			sq_pid = sq->task_pid;
196 			sq_cpu = sq->sq_cpu;
197 			getrusage(sq->thread, RUSAGE_SELF, &sq_usage);
198 			sq_total_time = (sq_usage.ru_stime.tv_sec * 1000000
199 					 + sq_usage.ru_stime.tv_usec);
200 			sq_work_time = sq->work_time;
201 		}
202 	}
203 
204 	seq_printf(m, "SqThread:\t%d\n", sq_pid);
205 	seq_printf(m, "SqThreadCpu:\t%d\n", sq_cpu);
206 	seq_printf(m, "SqTotalTime:\t%llu\n", sq_total_time);
207 	seq_printf(m, "SqWorkTime:\t%llu\n", sq_work_time);
208 	seq_printf(m, "UserFiles:\t%u\n", ctx->file_table.data.nr);
209 	for (i = 0; has_lock && i < ctx->file_table.data.nr; i++) {
210 		struct file *f = NULL;
211 
212 		if (ctx->file_table.data.nodes[i])
213 			f = io_slot_file(ctx->file_table.data.nodes[i]);
214 		if (f)
215 			seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
216 		else
217 			seq_printf(m, "%5u: <none>\n", i);
218 	}
219 	seq_printf(m, "UserBufs:\t%u\n", ctx->buf_table.nr);
220 	for (i = 0; has_lock && i < ctx->buf_table.nr; i++) {
221 		struct io_mapped_ubuf *buf = NULL;
222 
223 		if (ctx->buf_table.nodes[i])
224 			buf = ctx->buf_table.nodes[i]->buf;
225 		if (buf)
226 			seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, buf->len);
227 		else
228 			seq_printf(m, "%5u: <none>\n", i);
229 	}
230 	if (has_lock && !xa_empty(&ctx->personalities)) {
231 		unsigned long index;
232 		const struct cred *cred;
233 
234 		seq_printf(m, "Personalities:\n");
235 		xa_for_each(&ctx->personalities, index, cred)
236 			io_uring_show_cred(m, index, cred);
237 	}
238 
239 	seq_puts(m, "PollList:\n");
240 	for (i = 0; has_lock && i < (1U << ctx->cancel_table.hash_bits); i++) {
241 		struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
242 		struct io_kiocb *req;
243 
244 		hlist_for_each_entry(req, &hb->list, hash_node)
245 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
246 					task_work_pending(req->tctx->task));
247 	}
248 
249 	if (has_lock)
250 		mutex_unlock(&ctx->uring_lock);
251 
252 	seq_puts(m, "CqOverflowList:\n");
253 	spin_lock(&ctx->completion_lock);
254 	list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
255 		struct io_uring_cqe *cqe = &ocqe->cqe;
256 
257 		seq_printf(m, "  user_data=%llu, res=%d, flags=%x\n",
258 			   cqe->user_data, cqe->res, cqe->flags);
259 
260 	}
261 	spin_unlock(&ctx->completion_lock);
262 	napi_show_fdinfo(ctx, m);
263 }
264 #endif
265