1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/ioprio.c 4 * 5 * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk> 6 * 7 * Helper functions for setting/querying io priorities of processes. The 8 * system calls closely mimmick getpriority/setpriority, see the man page for 9 * those. The prio argument is a composite of prio class and prio data, where 10 * the data argument has meaning within that class. The standard scheduling 11 * classes have 8 distinct prio levels, with 0 being the highest prio and 7 12 * being the lowest. 13 * 14 * IOW, setting BE scheduling class with prio 2 is done ala: 15 * 16 * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2; 17 * 18 * ioprio_set(PRIO_PROCESS, pid, prio); 19 * 20 * See also Documentation/block/ioprio.rst 21 * 22 */ 23 #include <linux/gfp.h> 24 #include <linux/kernel.h> 25 #include <linux/export.h> 26 #include <linux/ioprio.h> 27 #include <linux/cred.h> 28 #include <linux/blkdev.h> 29 #include <linux/capability.h> 30 #include <linux/sched/user.h> 31 #include <linux/sched/task.h> 32 #include <linux/syscalls.h> 33 #include <linux/security.h> 34 #include <linux/pid_namespace.h> 35 36 int set_task_ioprio(struct task_struct *task, int ioprio) 37 { 38 int err; 39 struct io_context *ioc; 40 const struct cred *cred = current_cred(), *tcred; 41 42 rcu_read_lock(); 43 tcred = __task_cred(task); 44 if (!uid_eq(tcred->uid, cred->euid) && 45 !uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) { 46 rcu_read_unlock(); 47 return -EPERM; 48 } 49 rcu_read_unlock(); 50 51 err = security_task_setioprio(task, ioprio); 52 if (err) 53 return err; 54 55 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE); 56 if (ioc) { 57 ioc->ioprio = ioprio; 58 put_io_context(ioc); 59 } 60 61 return err; 62 } 63 EXPORT_SYMBOL_GPL(set_task_ioprio); 64 65 int ioprio_check_cap(int ioprio) 66 { 67 int class = IOPRIO_PRIO_CLASS(ioprio); 68 int data = IOPRIO_PRIO_DATA(ioprio); 69 70 switch (class) { 71 case IOPRIO_CLASS_RT: 72 /* 73 * Originally this only checked for CAP_SYS_ADMIN, 74 * which was implicitly allowed for pid 0 by security 75 * modules such as SELinux. Make sure we check 76 * CAP_SYS_ADMIN first to avoid a denial/avc for 77 * possibly missing CAP_SYS_NICE permission. 78 */ 79 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE)) 80 return -EPERM; 81 fallthrough; 82 /* rt has prio field too */ 83 case IOPRIO_CLASS_BE: 84 if (data >= IOPRIO_NR_LEVELS || data < 0) 85 return -EINVAL; 86 break; 87 case IOPRIO_CLASS_IDLE: 88 break; 89 case IOPRIO_CLASS_NONE: 90 if (data) 91 return -EINVAL; 92 break; 93 default: 94 return -EINVAL; 95 } 96 97 return 0; 98 } 99 100 SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio) 101 { 102 struct task_struct *p, *g; 103 struct user_struct *user; 104 struct pid *pgrp; 105 kuid_t uid; 106 int ret; 107 108 ret = ioprio_check_cap(ioprio); 109 if (ret) 110 return ret; 111 112 ret = -ESRCH; 113 rcu_read_lock(); 114 switch (which) { 115 case IOPRIO_WHO_PROCESS: 116 if (!who) 117 p = current; 118 else 119 p = find_task_by_vpid(who); 120 if (p) 121 ret = set_task_ioprio(p, ioprio); 122 break; 123 case IOPRIO_WHO_PGRP: 124 if (!who) 125 pgrp = task_pgrp(current); 126 else 127 pgrp = find_vpid(who); 128 129 read_lock(&tasklist_lock); 130 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 131 ret = set_task_ioprio(p, ioprio); 132 if (ret) { 133 read_unlock(&tasklist_lock); 134 goto out; 135 } 136 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 137 read_unlock(&tasklist_lock); 138 139 break; 140 case IOPRIO_WHO_USER: 141 uid = make_kuid(current_user_ns(), who); 142 if (!uid_valid(uid)) 143 break; 144 if (!who) 145 user = current_user(); 146 else 147 user = find_user(uid); 148 149 if (!user) 150 break; 151 152 for_each_process_thread(g, p) { 153 if (!uid_eq(task_uid(p), uid) || 154 !task_pid_vnr(p)) 155 continue; 156 ret = set_task_ioprio(p, ioprio); 157 if (ret) 158 goto free_uid; 159 } 160 free_uid: 161 if (who) 162 free_uid(user); 163 break; 164 default: 165 ret = -EINVAL; 166 } 167 168 out: 169 rcu_read_unlock(); 170 return ret; 171 } 172 173 static int get_task_ioprio(struct task_struct *p) 174 { 175 int ret; 176 177 ret = security_task_getioprio(p); 178 if (ret) 179 goto out; 180 ret = IOPRIO_DEFAULT; 181 task_lock(p); 182 if (p->io_context) 183 ret = p->io_context->ioprio; 184 task_unlock(p); 185 out: 186 return ret; 187 } 188 189 int ioprio_best(unsigned short aprio, unsigned short bprio) 190 { 191 if (!ioprio_valid(aprio)) 192 aprio = IOPRIO_DEFAULT; 193 if (!ioprio_valid(bprio)) 194 bprio = IOPRIO_DEFAULT; 195 196 return min(aprio, bprio); 197 } 198 199 SYSCALL_DEFINE2(ioprio_get, int, which, int, who) 200 { 201 struct task_struct *g, *p; 202 struct user_struct *user; 203 struct pid *pgrp; 204 kuid_t uid; 205 int ret = -ESRCH; 206 int tmpio; 207 208 rcu_read_lock(); 209 switch (which) { 210 case IOPRIO_WHO_PROCESS: 211 if (!who) 212 p = current; 213 else 214 p = find_task_by_vpid(who); 215 if (p) 216 ret = get_task_ioprio(p); 217 break; 218 case IOPRIO_WHO_PGRP: 219 if (!who) 220 pgrp = task_pgrp(current); 221 else 222 pgrp = find_vpid(who); 223 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 224 tmpio = get_task_ioprio(p); 225 if (tmpio < 0) 226 continue; 227 if (ret == -ESRCH) 228 ret = tmpio; 229 else 230 ret = ioprio_best(ret, tmpio); 231 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 232 break; 233 case IOPRIO_WHO_USER: 234 uid = make_kuid(current_user_ns(), who); 235 if (!who) 236 user = current_user(); 237 else 238 user = find_user(uid); 239 240 if (!user) 241 break; 242 243 for_each_process_thread(g, p) { 244 if (!uid_eq(task_uid(p), user->uid) || 245 !task_pid_vnr(p)) 246 continue; 247 tmpio = get_task_ioprio(p); 248 if (tmpio < 0) 249 continue; 250 if (ret == -ESRCH) 251 ret = tmpio; 252 else 253 ret = ioprio_best(ret, tmpio); 254 } 255 256 if (who) 257 free_uid(user); 258 break; 259 default: 260 ret = -EINVAL; 261 } 262 263 rcu_read_unlock(); 264 return ret; 265 } 266