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