xref: /linux/io_uring/fdinfo.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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
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 /*
50  * Caller holds a reference to the file already, we don't need to do
51  * anything else to get an extra reference.
52  */
53 __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *file)
54 {
55 	struct io_ring_ctx *ctx = file->private_data;
56 	struct io_overflow_cqe *ocqe;
57 	struct io_rings *r = ctx->rings;
58 	struct rusage sq_usage;
59 	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
60 	unsigned int sq_head = READ_ONCE(r->sq.head);
61 	unsigned int sq_tail = READ_ONCE(r->sq.tail);
62 	unsigned int cq_head = READ_ONCE(r->cq.head);
63 	unsigned int cq_tail = READ_ONCE(r->cq.tail);
64 	unsigned int cq_shift = 0;
65 	unsigned int sq_shift = 0;
66 	unsigned int sq_entries, cq_entries;
67 	int sq_pid = -1, sq_cpu = -1;
68 	u64 sq_total_time = 0, sq_work_time = 0;
69 	bool has_lock;
70 	unsigned int i;
71 
72 	if (ctx->flags & IORING_SETUP_CQE32)
73 		cq_shift = 1;
74 	if (ctx->flags & IORING_SETUP_SQE128)
75 		sq_shift = 1;
76 
77 	/*
78 	 * we may get imprecise sqe and cqe info if uring is actively running
79 	 * since we get cached_sq_head and cached_cq_tail without uring_lock
80 	 * and sq_tail and cq_head are changed by userspace. But it's ok since
81 	 * we usually use these info when it is stuck.
82 	 */
83 	seq_printf(m, "SqMask:\t0x%x\n", sq_mask);
84 	seq_printf(m, "SqHead:\t%u\n", sq_head);
85 	seq_printf(m, "SqTail:\t%u\n", sq_tail);
86 	seq_printf(m, "CachedSqHead:\t%u\n", ctx->cached_sq_head);
87 	seq_printf(m, "CqMask:\t0x%x\n", cq_mask);
88 	seq_printf(m, "CqHead:\t%u\n", cq_head);
89 	seq_printf(m, "CqTail:\t%u\n", cq_tail);
90 	seq_printf(m, "CachedCqTail:\t%u\n", ctx->cached_cq_tail);
91 	seq_printf(m, "SQEs:\t%u\n", sq_tail - sq_head);
92 	sq_entries = min(sq_tail - sq_head, ctx->sq_entries);
93 	for (i = 0; i < sq_entries; i++) {
94 		unsigned int entry = i + sq_head;
95 		struct io_uring_sqe *sqe;
96 		unsigned int sq_idx;
97 
98 		if (ctx->flags & IORING_SETUP_NO_SQARRAY)
99 			break;
100 		sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]);
101 		if (sq_idx > sq_mask)
102 			continue;
103 		sqe = &ctx->sq_sqes[sq_idx << sq_shift];
104 		seq_printf(m, "%5u: opcode:%s, fd:%d, flags:%x, off:%llu, "
105 			      "addr:0x%llx, rw_flags:0x%x, buf_index:%d "
106 			      "user_data:%llu",
107 			   sq_idx, io_uring_get_opcode(sqe->opcode), sqe->fd,
108 			   sqe->flags, (unsigned long long) sqe->off,
109 			   (unsigned long long) sqe->addr, sqe->rw_flags,
110 			   sqe->buf_index, sqe->user_data);
111 		if (sq_shift) {
112 			u64 *sqeb = (void *) (sqe + 1);
113 			int size = sizeof(struct io_uring_sqe) / sizeof(u64);
114 			int j;
115 
116 			for (j = 0; j < size; j++) {
117 				seq_printf(m, ", e%d:0x%llx", j,
118 						(unsigned long long) *sqeb);
119 				sqeb++;
120 			}
121 		}
122 		seq_printf(m, "\n");
123 	}
124 	seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head);
125 	cq_entries = min(cq_tail - cq_head, ctx->cq_entries);
126 	for (i = 0; i < cq_entries; i++) {
127 		unsigned int entry = i + cq_head;
128 		struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift];
129 
130 		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x",
131 			   entry & cq_mask, cqe->user_data, cqe->res,
132 			   cqe->flags);
133 		if (cq_shift)
134 			seq_printf(m, ", extra1:%llu, extra2:%llu\n",
135 					cqe->big_cqe[0], cqe->big_cqe[1]);
136 		seq_printf(m, "\n");
137 	}
138 
139 	/*
140 	 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
141 	 * since fdinfo case grabs it in the opposite direction of normal use
142 	 * cases. If we fail to get the lock, we just don't iterate any
143 	 * structures that could be going away outside the io_uring mutex.
144 	 */
145 	has_lock = mutex_trylock(&ctx->uring_lock);
146 
147 	if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
148 		struct io_sq_data *sq = ctx->sq_data;
149 
150 		/*
151 		 * sq->thread might be NULL if we raced with the sqpoll
152 		 * thread termination.
153 		 */
154 		if (sq->thread) {
155 			sq_pid = sq->task_pid;
156 			sq_cpu = sq->sq_cpu;
157 			getrusage(sq->thread, RUSAGE_SELF, &sq_usage);
158 			sq_total_time = (sq_usage.ru_stime.tv_sec * 1000000
159 					 + sq_usage.ru_stime.tv_usec);
160 			sq_work_time = sq->work_time;
161 		}
162 	}
163 
164 	seq_printf(m, "SqThread:\t%d\n", sq_pid);
165 	seq_printf(m, "SqThreadCpu:\t%d\n", sq_cpu);
166 	seq_printf(m, "SqTotalTime:\t%llu\n", sq_total_time);
167 	seq_printf(m, "SqWorkTime:\t%llu\n", sq_work_time);
168 	seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
169 	for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
170 		struct file *f = io_file_from_index(&ctx->file_table, i);
171 
172 		if (f)
173 			seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
174 		else
175 			seq_printf(m, "%5u: <none>\n", i);
176 	}
177 	seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
178 	for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
179 		struct io_mapped_ubuf *buf = ctx->user_bufs[i];
180 
181 		seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, buf->len);
182 	}
183 	if (has_lock && !xa_empty(&ctx->personalities)) {
184 		unsigned long index;
185 		const struct cred *cred;
186 
187 		seq_printf(m, "Personalities:\n");
188 		xa_for_each(&ctx->personalities, index, cred)
189 			io_uring_show_cred(m, index, cred);
190 	}
191 
192 	seq_puts(m, "PollList:\n");
193 	for (i = 0; i < (1U << ctx->cancel_table.hash_bits); i++) {
194 		struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
195 		struct io_hash_bucket *hbl = &ctx->cancel_table_locked.hbs[i];
196 		struct io_kiocb *req;
197 
198 		spin_lock(&hb->lock);
199 		hlist_for_each_entry(req, &hb->list, hash_node)
200 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
201 					task_work_pending(req->task));
202 		spin_unlock(&hb->lock);
203 
204 		if (!has_lock)
205 			continue;
206 		hlist_for_each_entry(req, &hbl->list, hash_node)
207 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
208 					task_work_pending(req->task));
209 	}
210 
211 	if (has_lock)
212 		mutex_unlock(&ctx->uring_lock);
213 
214 	seq_puts(m, "CqOverflowList:\n");
215 	spin_lock(&ctx->completion_lock);
216 	list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
217 		struct io_uring_cqe *cqe = &ocqe->cqe;
218 
219 		seq_printf(m, "  user_data=%llu, res=%d, flags=%x\n",
220 			   cqe->user_data, cqe->res, cqe->flags);
221 
222 	}
223 	spin_unlock(&ctx->completion_lock);
224 
225 #ifdef CONFIG_NET_RX_BUSY_POLL
226 	if (ctx->napi_enabled) {
227 		seq_puts(m, "NAPI:\tenabled\n");
228 		seq_printf(m, "napi_busy_poll_dt:\t%llu\n", ctx->napi_busy_poll_dt);
229 		if (ctx->napi_prefer_busy_poll)
230 			seq_puts(m, "napi_prefer_busy_poll:\ttrue\n");
231 		else
232 			seq_puts(m, "napi_prefer_busy_poll:\tfalse\n");
233 	} else {
234 		seq_puts(m, "NAPI:\tdisabled\n");
235 	}
236 #endif
237 }
238 #endif
239