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 	unsigned __capi;
26 	int g;
27 
28 	seq_printf(m, "%5d\n", id);
29 	seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
30 	seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
31 	seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
32 	seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
33 	seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
34 	seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
35 	seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
36 	seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
37 	seq_puts(m, "\n\tGroups:\t");
38 	gi = cred->group_info;
39 	for (g = 0; g < gi->ngroups; g++) {
40 		seq_put_decimal_ull(m, g ? " " : "",
41 					from_kgid_munged(uns, gi->gid[g]));
42 	}
43 	seq_puts(m, "\n\tCapEff:\t");
44 	cap = cred->cap_effective;
45 	CAP_FOR_EACH_U32(__capi)
46 		seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
47 	seq_putc(m, '\n');
48 	return 0;
49 }
50 
__io_uring_show_fdinfo(struct io_ring_ctx * ctx,struct seq_file * m)51 static __cold void __io_uring_show_fdinfo(struct io_ring_ctx *ctx,
52 					  struct seq_file *m)
53 {
54 	struct io_sq_data *sq = NULL;
55 	struct io_overflow_cqe *ocqe;
56 	struct io_rings *r = ctx->rings;
57 	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
58 	unsigned int sq_head = READ_ONCE(r->sq.head);
59 	unsigned int sq_tail = READ_ONCE(r->sq.tail);
60 	unsigned int cq_head = READ_ONCE(r->cq.head);
61 	unsigned int cq_tail = READ_ONCE(r->cq.tail);
62 	unsigned int cq_shift = 0;
63 	unsigned int sq_shift = 0;
64 	unsigned int sq_entries, cq_entries;
65 	bool has_lock;
66 	unsigned int i;
67 
68 	if (ctx->flags & IORING_SETUP_CQE32)
69 		cq_shift = 1;
70 	if (ctx->flags & IORING_SETUP_SQE128)
71 		sq_shift = 1;
72 
73 	/*
74 	 * we may get imprecise sqe and cqe info if uring is actively running
75 	 * since we get cached_sq_head and cached_cq_tail without uring_lock
76 	 * and sq_tail and cq_head are changed by userspace. But it's ok since
77 	 * we usually use these info when it is stuck.
78 	 */
79 	seq_printf(m, "SqMask:\t0x%x\n", sq_mask);
80 	seq_printf(m, "SqHead:\t%u\n", sq_head);
81 	seq_printf(m, "SqTail:\t%u\n", sq_tail);
82 	seq_printf(m, "CachedSqHead:\t%u\n", ctx->cached_sq_head);
83 	seq_printf(m, "CqMask:\t0x%x\n", cq_mask);
84 	seq_printf(m, "CqHead:\t%u\n", cq_head);
85 	seq_printf(m, "CqTail:\t%u\n", cq_tail);
86 	seq_printf(m, "CachedCqTail:\t%u\n", ctx->cached_cq_tail);
87 	seq_printf(m, "SQEs:\t%u\n", sq_tail - sq_head);
88 	sq_entries = min(sq_tail - sq_head, ctx->sq_entries);
89 	for (i = 0; i < sq_entries; i++) {
90 		unsigned int entry = i + sq_head;
91 		struct io_uring_sqe *sqe;
92 		unsigned int sq_idx;
93 
94 		sq_idx = READ_ONCE(ctx->sq_array[entry & sq_mask]);
95 		if (sq_idx > sq_mask)
96 			continue;
97 		sqe = &ctx->sq_sqes[sq_idx << sq_shift];
98 		seq_printf(m, "%5u: opcode:%s, fd:%d, flags:%x, off:%llu, "
99 			      "addr:0x%llx, rw_flags:0x%x, buf_index:%d "
100 			      "user_data:%llu",
101 			   sq_idx, io_uring_get_opcode(sqe->opcode), sqe->fd,
102 			   sqe->flags, (unsigned long long) sqe->off,
103 			   (unsigned long long) sqe->addr, sqe->rw_flags,
104 			   sqe->buf_index, sqe->user_data);
105 		if (sq_shift) {
106 			u64 *sqeb = (void *) (sqe + 1);
107 			int size = sizeof(struct io_uring_sqe) / sizeof(u64);
108 			int j;
109 
110 			for (j = 0; j < size; j++) {
111 				seq_printf(m, ", e%d:0x%llx", j,
112 						(unsigned long long) *sqeb);
113 				sqeb++;
114 			}
115 		}
116 		seq_printf(m, "\n");
117 	}
118 	seq_printf(m, "CQEs:\t%u\n", cq_tail - cq_head);
119 	cq_entries = min(cq_tail - cq_head, ctx->cq_entries);
120 	for (i = 0; i < cq_entries; i++) {
121 		unsigned int entry = i + cq_head;
122 		struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift];
123 
124 		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x",
125 			   entry & cq_mask, cqe->user_data, cqe->res,
126 			   cqe->flags);
127 		if (cq_shift)
128 			seq_printf(m, ", extra1:%llu, extra2:%llu\n",
129 					cqe->big_cqe[0], cqe->big_cqe[1]);
130 		seq_printf(m, "\n");
131 	}
132 
133 	/*
134 	 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
135 	 * since fdinfo case grabs it in the opposite direction of normal use
136 	 * cases. If we fail to get the lock, we just don't iterate any
137 	 * structures that could be going away outside the io_uring mutex.
138 	 */
139 	has_lock = mutex_trylock(&ctx->uring_lock);
140 
141 	if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
142 		sq = ctx->sq_data;
143 		if (!sq->thread)
144 			sq = NULL;
145 	}
146 
147 	seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
148 	seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
149 	seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
150 	for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
151 		struct file *f = io_file_from_index(&ctx->file_table, i);
152 
153 		if (f)
154 			seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
155 		else
156 			seq_printf(m, "%5u: <none>\n", i);
157 	}
158 	seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
159 	for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
160 		struct io_mapped_ubuf *buf = ctx->user_bufs[i];
161 		unsigned int len = buf->ubuf_end - buf->ubuf;
162 
163 		seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len);
164 	}
165 	if (has_lock && !xa_empty(&ctx->personalities)) {
166 		unsigned long index;
167 		const struct cred *cred;
168 
169 		seq_printf(m, "Personalities:\n");
170 		xa_for_each(&ctx->personalities, index, cred)
171 			io_uring_show_cred(m, index, cred);
172 	}
173 
174 	seq_puts(m, "PollList:\n");
175 	for (i = 0; i < (1U << ctx->cancel_table.hash_bits); i++) {
176 		struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
177 		struct io_hash_bucket *hbl = &ctx->cancel_table_locked.hbs[i];
178 		struct io_kiocb *req;
179 
180 		spin_lock(&hb->lock);
181 		hlist_for_each_entry(req, &hb->list, hash_node)
182 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
183 					task_work_pending(req->task));
184 		spin_unlock(&hb->lock);
185 
186 		if (!has_lock)
187 			continue;
188 		hlist_for_each_entry(req, &hbl->list, hash_node)
189 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
190 					task_work_pending(req->task));
191 	}
192 
193 	if (has_lock)
194 		mutex_unlock(&ctx->uring_lock);
195 
196 	seq_puts(m, "CqOverflowList:\n");
197 	spin_lock(&ctx->completion_lock);
198 	list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
199 		struct io_uring_cqe *cqe = &ocqe->cqe;
200 
201 		seq_printf(m, "  user_data=%llu, res=%d, flags=%x\n",
202 			   cqe->user_data, cqe->res, cqe->flags);
203 
204 	}
205 
206 	spin_unlock(&ctx->completion_lock);
207 }
208 
io_uring_show_fdinfo(struct seq_file * m,struct file * f)209 __cold void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
210 {
211 	struct io_ring_ctx *ctx = f->private_data;
212 
213 	if (percpu_ref_tryget(&ctx->refs)) {
214 		__io_uring_show_fdinfo(ctx, m);
215 		percpu_ref_put(&ctx->refs);
216 	}
217 }
218 #endif
219