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 if (!capable(CAP_SYS_NICE) && !capable(CAP_SYS_ADMIN)) 73 return -EPERM; 74 fallthrough; 75 /* rt has prio field too */ 76 case IOPRIO_CLASS_BE: 77 if (data >= IOPRIO_BE_NR || data < 0) 78 return -EINVAL; 79 80 break; 81 case IOPRIO_CLASS_IDLE: 82 break; 83 case IOPRIO_CLASS_NONE: 84 if (data) 85 return -EINVAL; 86 break; 87 default: 88 return -EINVAL; 89 } 90 91 return 0; 92 } 93 94 SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio) 95 { 96 struct task_struct *p, *g; 97 struct user_struct *user; 98 struct pid *pgrp; 99 kuid_t uid; 100 int ret; 101 102 ret = ioprio_check_cap(ioprio); 103 if (ret) 104 return ret; 105 106 ret = -ESRCH; 107 rcu_read_lock(); 108 switch (which) { 109 case IOPRIO_WHO_PROCESS: 110 if (!who) 111 p = current; 112 else 113 p = find_task_by_vpid(who); 114 if (p) 115 ret = set_task_ioprio(p, ioprio); 116 break; 117 case IOPRIO_WHO_PGRP: 118 if (!who) 119 pgrp = task_pgrp(current); 120 else 121 pgrp = find_vpid(who); 122 123 read_lock(&tasklist_lock); 124 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 125 ret = set_task_ioprio(p, ioprio); 126 if (ret) { 127 read_unlock(&tasklist_lock); 128 goto out; 129 } 130 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 131 read_unlock(&tasklist_lock); 132 133 break; 134 case IOPRIO_WHO_USER: 135 uid = make_kuid(current_user_ns(), who); 136 if (!uid_valid(uid)) 137 break; 138 if (!who) 139 user = current_user(); 140 else 141 user = find_user(uid); 142 143 if (!user) 144 break; 145 146 for_each_process_thread(g, p) { 147 if (!uid_eq(task_uid(p), uid) || 148 !task_pid_vnr(p)) 149 continue; 150 ret = set_task_ioprio(p, ioprio); 151 if (ret) 152 goto free_uid; 153 } 154 free_uid: 155 if (who) 156 free_uid(user); 157 break; 158 default: 159 ret = -EINVAL; 160 } 161 162 out: 163 rcu_read_unlock(); 164 return ret; 165 } 166 167 static int get_task_ioprio(struct task_struct *p) 168 { 169 int ret; 170 171 ret = security_task_getioprio(p); 172 if (ret) 173 goto out; 174 ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM); 175 task_lock(p); 176 if (p->io_context) 177 ret = p->io_context->ioprio; 178 task_unlock(p); 179 out: 180 return ret; 181 } 182 183 int ioprio_best(unsigned short aprio, unsigned short bprio) 184 { 185 if (!ioprio_valid(aprio)) 186 aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM); 187 if (!ioprio_valid(bprio)) 188 bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM); 189 190 return min(aprio, bprio); 191 } 192 193 SYSCALL_DEFINE2(ioprio_get, int, which, int, who) 194 { 195 struct task_struct *g, *p; 196 struct user_struct *user; 197 struct pid *pgrp; 198 kuid_t uid; 199 int ret = -ESRCH; 200 int tmpio; 201 202 rcu_read_lock(); 203 switch (which) { 204 case IOPRIO_WHO_PROCESS: 205 if (!who) 206 p = current; 207 else 208 p = find_task_by_vpid(who); 209 if (p) 210 ret = get_task_ioprio(p); 211 break; 212 case IOPRIO_WHO_PGRP: 213 if (!who) 214 pgrp = task_pgrp(current); 215 else 216 pgrp = find_vpid(who); 217 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { 218 tmpio = get_task_ioprio(p); 219 if (tmpio < 0) 220 continue; 221 if (ret == -ESRCH) 222 ret = tmpio; 223 else 224 ret = ioprio_best(ret, tmpio); 225 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); 226 break; 227 case IOPRIO_WHO_USER: 228 uid = make_kuid(current_user_ns(), who); 229 if (!who) 230 user = current_user(); 231 else 232 user = find_user(uid); 233 234 if (!user) 235 break; 236 237 for_each_process_thread(g, p) { 238 if (!uid_eq(task_uid(p), user->uid) || 239 !task_pid_vnr(p)) 240 continue; 241 tmpio = get_task_ioprio(p); 242 if (tmpio < 0) 243 continue; 244 if (ret == -ESRCH) 245 ret = tmpio; 246 else 247 ret = ioprio_best(ret, tmpio); 248 } 249 250 if (who) 251 free_uid(user); 252 break; 253 default: 254 ret = -EINVAL; 255 } 256 257 rcu_read_unlock(); 258 return ret; 259 } 260