aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/events
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/events')
-rw-r--r--kernel/events/uprobes.c323
1 files changed, 319 insertions, 4 deletions
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index e56e56aa7535..b807d1566b64 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -30,9 +30,12 @@
30#include <linux/rmap.h> /* anon_vma_prepare */ 30#include <linux/rmap.h> /* anon_vma_prepare */
31#include <linux/mmu_notifier.h> /* set_pte_at_notify */ 31#include <linux/mmu_notifier.h> /* set_pte_at_notify */
32#include <linux/swap.h> /* try_to_free_swap */ 32#include <linux/swap.h> /* try_to_free_swap */
33#include <linux/ptrace.h> /* user_enable_single_step */
34#include <linux/kdebug.h> /* notifier mechanism */
33 35
34#include <linux/uprobes.h> 36#include <linux/uprobes.h>
35 37
38static struct srcu_struct uprobes_srcu;
36static struct rb_root uprobes_tree = RB_ROOT; 39static struct rb_root uprobes_tree = RB_ROOT;
37 40
38static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ 41static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
@@ -486,6 +489,9 @@ static struct uprobe *insert_uprobe(struct uprobe *uprobe)
486 u = __insert_uprobe(uprobe); 489 u = __insert_uprobe(uprobe);
487 spin_unlock_irqrestore(&uprobes_treelock, flags); 490 spin_unlock_irqrestore(&uprobes_treelock, flags);
488 491
492 /* For now assume that the instruction need not be single-stepped */
493 uprobe->flags |= UPROBE_SKIP_SSTEP;
494
489 return u; 495 return u;
490} 496}
491 497
@@ -523,6 +529,21 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
523 return uprobe; 529 return uprobe;
524} 530}
525 531
532static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
533{
534 struct uprobe_consumer *uc;
535
536 if (!(uprobe->flags & UPROBE_RUN_HANDLER))
537 return;
538
539 down_read(&uprobe->consumer_rwsem);
540 for (uc = uprobe->consumers; uc; uc = uc->next) {
541 if (!uc->filter || uc->filter(uc, current))
542 uc->handler(uc, regs);
543 }
544 up_read(&uprobe->consumer_rwsem);
545}
546
526/* Returns the previous consumer */ 547/* Returns the previous consumer */
527static struct uprobe_consumer * 548static struct uprobe_consumer *
528consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) 549consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
@@ -645,7 +666,7 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
645 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn)) 666 if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
646 return -EEXIST; 667 return -EEXIST;
647 668
648 ret = arch_uprobes_analyze_insn(&uprobe->arch, mm); 669 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm);
649 if (ret) 670 if (ret)
650 return ret; 671 return ret;
651 672
@@ -662,10 +683,21 @@ remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, loff_t vaddr)
662 set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true); 683 set_orig_insn(&uprobe->arch, mm, (unsigned long)vaddr, true);
663} 684}
664 685
686/*
687 * There could be threads that have hit the breakpoint and are entering the
688 * notifier code and trying to acquire the uprobes_treelock. The thread
689 * calling delete_uprobe() that is removing the uprobe from the rb_tree can
690 * race with these threads and might acquire the uprobes_treelock compared
691 * to some of the breakpoint hit threads. In such a case, the breakpoint
692 * hit threads will not find the uprobe. The current unregistering thread
693 * waits till all other threads have hit a breakpoint, to acquire the
694 * uprobes_treelock before the uprobe is removed from the rbtree.
695 */
665static void delete_uprobe(struct uprobe *uprobe) 696static void delete_uprobe(struct uprobe *uprobe)
666{ 697{
667 unsigned long flags; 698 unsigned long flags;
668 699
700 synchronize_srcu(&uprobes_srcu);
669 spin_lock_irqsave(&uprobes_treelock, flags); 701 spin_lock_irqsave(&uprobes_treelock, flags);
670 rb_erase(&uprobe->rb_node, &uprobes_tree); 702 rb_erase(&uprobe->rb_node, &uprobes_tree);
671 spin_unlock_irqrestore(&uprobes_treelock, flags); 703 spin_unlock_irqrestore(&uprobes_treelock, flags);
@@ -1010,6 +1042,288 @@ int uprobe_mmap(struct vm_area_struct *vma)
1010 return ret; 1042 return ret;
1011} 1043}
1012 1044
1045/**
1046 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1047 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1048 * instruction.
1049 * Return the address of the breakpoint instruction.
1050 */
1051unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1052{
1053 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1054}
1055
1056/*
1057 * Called with no locks held.
1058 * Called in context of a exiting or a exec-ing thread.
1059 */
1060void uprobe_free_utask(struct task_struct *t)
1061{
1062 struct uprobe_task *utask = t->utask;
1063
1064 if (t->uprobe_srcu_id != -1)
1065 srcu_read_unlock_raw(&uprobes_srcu, t->uprobe_srcu_id);
1066
1067 if (!utask)
1068 return;
1069
1070 if (utask->active_uprobe)
1071 put_uprobe(utask->active_uprobe);
1072
1073 kfree(utask);
1074 t->utask = NULL;
1075}
1076
1077/*
1078 * Called in context of a new clone/fork from copy_process.
1079 */
1080void uprobe_copy_process(struct task_struct *t)
1081{
1082 t->utask = NULL;
1083 t->uprobe_srcu_id = -1;
1084}
1085
1086/*
1087 * Allocate a uprobe_task object for the task.
1088 * Called when the thread hits a breakpoint for the first time.
1089 *
1090 * Returns:
1091 * - pointer to new uprobe_task on success
1092 * - NULL otherwise
1093 */
1094static struct uprobe_task *add_utask(void)
1095{
1096 struct uprobe_task *utask;
1097
1098 utask = kzalloc(sizeof *utask, GFP_KERNEL);
1099 if (unlikely(!utask))
1100 return NULL;
1101
1102 utask->active_uprobe = NULL;
1103 current->utask = utask;
1104 return utask;
1105}
1106
1107/* Prepare to single-step probed instruction out of line. */
1108static int
1109pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1110{
1111 return -EFAULT;
1112}
1113
1114/*
1115 * If we are singlestepping, then ensure this thread is not connected to
1116 * non-fatal signals until completion of singlestep. When xol insn itself
1117 * triggers the signal, restart the original insn even if the task is
1118 * already SIGKILL'ed (since coredump should report the correct ip). This
1119 * is even more important if the task has a handler for SIGSEGV/etc, The
1120 * _same_ instruction should be repeated again after return from the signal
1121 * handler, and SSTEP can never finish in this case.
1122 */
1123bool uprobe_deny_signal(void)
1124{
1125 struct task_struct *t = current;
1126 struct uprobe_task *utask = t->utask;
1127
1128 if (likely(!utask || !utask->active_uprobe))
1129 return false;
1130
1131 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1132
1133 if (signal_pending(t)) {
1134 spin_lock_irq(&t->sighand->siglock);
1135 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1136 spin_unlock_irq(&t->sighand->siglock);
1137
1138 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1139 utask->state = UTASK_SSTEP_TRAPPED;
1140 set_tsk_thread_flag(t, TIF_UPROBE);
1141 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1142 }
1143 }
1144
1145 return true;
1146}
1147
1148/*
1149 * Avoid singlestepping the original instruction if the original instruction
1150 * is a NOP or can be emulated.
1151 */
1152static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1153{
1154 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1155 return true;
1156
1157 uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1158 return false;
1159}
1160
1161/*
1162 * Run handler and ask thread to singlestep.
1163 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1164 */
1165static void handle_swbp(struct pt_regs *regs)
1166{
1167 struct vm_area_struct *vma;
1168 struct uprobe_task *utask;
1169 struct uprobe *uprobe;
1170 struct mm_struct *mm;
1171 unsigned long bp_vaddr;
1172
1173 uprobe = NULL;
1174 bp_vaddr = uprobe_get_swbp_addr(regs);
1175 mm = current->mm;
1176 down_read(&mm->mmap_sem);
1177 vma = find_vma(mm, bp_vaddr);
1178
1179 if (vma && vma->vm_start <= bp_vaddr && valid_vma(vma, false)) {
1180 struct inode *inode;
1181 loff_t offset;
1182
1183 inode = vma->vm_file->f_mapping->host;
1184 offset = bp_vaddr - vma->vm_start;
1185 offset += (vma->vm_pgoff << PAGE_SHIFT);
1186 uprobe = find_uprobe(inode, offset);
1187 }
1188
1189 srcu_read_unlock_raw(&uprobes_srcu, current->uprobe_srcu_id);
1190 current->uprobe_srcu_id = -1;
1191 up_read(&mm->mmap_sem);
1192
1193 if (!uprobe) {
1194 /* No matching uprobe; signal SIGTRAP. */
1195 send_sig(SIGTRAP, current, 0);
1196 return;
1197 }
1198
1199 utask = current->utask;
1200 if (!utask) {
1201 utask = add_utask();
1202 /* Cannot allocate; re-execute the instruction. */
1203 if (!utask)
1204 goto cleanup_ret;
1205 }
1206 utask->active_uprobe = uprobe;
1207 handler_chain(uprobe, regs);
1208 if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1209 goto cleanup_ret;
1210
1211 utask->state = UTASK_SSTEP;
1212 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1213 user_enable_single_step(current);
1214 return;
1215 }
1216
1217cleanup_ret:
1218 if (utask) {
1219 utask->active_uprobe = NULL;
1220 utask->state = UTASK_RUNNING;
1221 }
1222 if (uprobe) {
1223 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1224
1225 /*
1226 * cannot singlestep; cannot skip instruction;
1227 * re-execute the instruction.
1228 */
1229 instruction_pointer_set(regs, bp_vaddr);
1230
1231 put_uprobe(uprobe);
1232 }
1233}
1234
1235/*
1236 * Perform required fix-ups and disable singlestep.
1237 * Allow pending signals to take effect.
1238 */
1239static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1240{
1241 struct uprobe *uprobe;
1242
1243 uprobe = utask->active_uprobe;
1244 if (utask->state == UTASK_SSTEP_ACK)
1245 arch_uprobe_post_xol(&uprobe->arch, regs);
1246 else if (utask->state == UTASK_SSTEP_TRAPPED)
1247 arch_uprobe_abort_xol(&uprobe->arch, regs);
1248 else
1249 WARN_ON_ONCE(1);
1250
1251 put_uprobe(uprobe);
1252 utask->active_uprobe = NULL;
1253 utask->state = UTASK_RUNNING;
1254 user_disable_single_step(current);
1255
1256 spin_lock_irq(&current->sighand->siglock);
1257 recalc_sigpending(); /* see uprobe_deny_signal() */
1258 spin_unlock_irq(&current->sighand->siglock);
1259}
1260
1261/*
1262 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag. (and on
1263 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1264 * allows the thread to return from interrupt.
1265 *
1266 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1267 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1268 * interrupt.
1269 *
1270 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1271 * uprobe_notify_resume().
1272 */
1273void uprobe_notify_resume(struct pt_regs *regs)
1274{
1275 struct uprobe_task *utask;
1276
1277 utask = current->utask;
1278 if (!utask || utask->state == UTASK_BP_HIT)
1279 handle_swbp(regs);
1280 else
1281 handle_singlestep(utask, regs);
1282}
1283
1284/*
1285 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1286 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1287 */
1288int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1289{
1290 struct uprobe_task *utask;
1291
1292 if (!current->mm)
1293 return 0;
1294
1295 utask = current->utask;
1296 if (utask)
1297 utask->state = UTASK_BP_HIT;
1298
1299 set_thread_flag(TIF_UPROBE);
1300 current->uprobe_srcu_id = srcu_read_lock_raw(&uprobes_srcu);
1301
1302 return 1;
1303}
1304
1305/*
1306 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1307 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1308 */
1309int uprobe_post_sstep_notifier(struct pt_regs *regs)
1310{
1311 struct uprobe_task *utask = current->utask;
1312
1313 if (!current->mm || !utask || !utask->active_uprobe)
1314 /* task is currently not uprobed */
1315 return 0;
1316
1317 utask->state = UTASK_SSTEP_ACK;
1318 set_thread_flag(TIF_UPROBE);
1319 return 1;
1320}
1321
1322static struct notifier_block uprobe_exception_nb = {
1323 .notifier_call = arch_uprobe_exception_notify,
1324 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
1325};
1326
1013static int __init init_uprobes(void) 1327static int __init init_uprobes(void)
1014{ 1328{
1015 int i; 1329 int i;
@@ -1018,12 +1332,13 @@ static int __init init_uprobes(void)
1018 mutex_init(&uprobes_mutex[i]); 1332 mutex_init(&uprobes_mutex[i]);
1019 mutex_init(&uprobes_mmap_mutex[i]); 1333 mutex_init(&uprobes_mmap_mutex[i]);
1020 } 1334 }
1021 return 0; 1335 init_srcu_struct(&uprobes_srcu);
1336
1337 return register_die_notifier(&uprobe_exception_nb);
1022} 1338}
1339module_init(init_uprobes);
1023 1340
1024static void __exit exit_uprobes(void) 1341static void __exit exit_uprobes(void)
1025{ 1342{
1026} 1343}
1027
1028module_init(init_uprobes);
1029module_exit(exit_uprobes); 1344module_exit(exit_uprobes);