CSC 539: Operating Systems Structure and Design Spring 2005

Process deadlock
deadlock prevention deadlock avoidance deadlock detection recovery from deadlock

Process deadlock
in general, can partition system resources into equivalence types or classes
e.g, CPUs, RAM, files, printers, tape drives,

when a process requests a resource from a particular resource class, any available resource in the class is allocated to the process
generally, resources go through the sequence: 1. request 2. use 3. release

deadlock occurs when a set of processes is waiting for resources that are held by other processes in the set
2

Deadlock examples
real-world example: single-lane bridge
each section of a bridge can be viewed as a resource if a deadlock occurs, it can be resolved if one car backs up (preempt & rollback) several cars may have to be backed up if a deadlock occurs starvation is possible

system example: 2 tape drives

P1 and P2 each hold one tape drive and each needs another one

system example: semaphores A and B, initialized to 1

P0: wait(A); wait(B); P1: wait(B); wait(A);
3

Deadlock characterization
deadlock can arise if four conditions hold simultaneously:
Mutual Exclusion: only one process at a time can use a resource. Hold and Wait: a process holding at least one resource is waiting to acquire additional resources held by other processes. No Preemption: a resource can be released only voluntarily by the process holding it, after that process has completed its task. Circular Wait: there exists a set {P0, P1, , Pn} of waiting processes such that: P0 is waiting for a resource held by P1 P1 is waiting for a resource held by P2 Pn is waiting for a resource held by P0

consider single-lane bridge example: 4 necessary conditions?

4

System resource allocation graph

we can use a directed graph to represent the state of the system as it allocates resources and deals with requests for resources
nodes of graph represent processes or resources process nodes will have circles around them resource nodes will have rectangles around them edges of graph represent allocations or requests process to resource edge is a request resource to process edge is an allocation

in this example: P1 has R2, has requested R1 P2 has R1 & R2, has requested R3 P3 has R3

cycle = deadlock?
If graph contains no cycles If graph contains a cycle no deadlock

if only one instance per resource type, then deadlock. if several instances per resource type, deadlock is possible

no deadlock

deadlock

no deadlock

Methods for dealing with deadlock

Prevention or Avoidance: ensure system never enters a deadlocked state
deadlock prevention scheme ensure that at least one of the necessary conditions cannot hold deadlock avoidance scheme require the OS to know the resource usage requirements of all processes for each request, the OS decides if it could lead to a deadlock before granting

Detection & Recovery:

allow deadlocks to occur system implements an algorithm for deadlock detection, if so then recover

Ignorance:

assume/pretend deadlocks never occur in the system used by most operating systems, including UNIX & Windows
7

Deadlock prevention
to prevent deadlock, must eliminate one of the four necessary conditions
Mutual Exclusion: can't eliminate since some resources are not shareable Hold and Wait: options exist, but wasteful and may lead to starvation 1. grant all resources when process starts, or 2. allocate and deallocate resources in complete groups No Preemption: options exist, but taking away resources can be tricky Circular Wait: most likely candidate impose an ordering on resources [R0, R1, , Rn], and require that each process requests resources in increasing order e.g., P0 wants R0, R1, and R2 P1 wants R1, R2, and R3

informal proof that resource ordering ensures no circular wait?

8

Deadlock avoidance
if have some knowledge of future resource demands (e.g., max number of resources needed per process), can avoid deadlock situations
safe state: a state is safe if the system can allocate resources to each process (up to its maximum need) in some sequence and still avoid a deadlock safe sequence: a sequence of processes {P0, P1, , Pn} is safe for the current allocation state if, for each Pi, the resources that Pi can still request are either free or else held by {P0, P1, , Pi-1} a safe sequence implies that each process need only wait for lower numbered processes to finish before it can finish

Note: unsafe deadlock

a safe state is guaranteed no deadlock an unsafe state may lead to deadlock
9

Example: safe or unsafe?

suppose there are 12 instances of a given resource (e.g., tape drives) in the current state, resources are allocated to 3 processes
Process P: max need = 10, currently allocated = 5 currently allocated = 2 currently allocated = 2

Process Q: max need = 4, Process R: max need = 9,

is this a safe state? if so, identify a safe sequence.

10

Resource allocation graph algorithm

if there is only one instance of each resource class, then unsafe states can be avoided by considering a generalized resource allocation graph
allocation edge Ri Pj request edge Pi Rj claim edge Pi Rj if Ri is currently held by Pj (as before) if Pi has requested Rj (as before) if Pi may eventually request Rj

unsafe avoidance: do not allocate resource if it creates a cycle

allocate R2 to P1: OK allocate R2 to P2: NOT ALLOWED

leads to cycle in graph (unsafe state)

11

Banker's algorithm
the banker's algorithm can avoid deadlocks even with multiple instances
looks at each request for resources and tests if the request moves the system into an unsafe state if the system is still safe, then the request is granted if the system would become unsafe, then the request is denied

utilizes the following data structures:

Available[m] Max[n][m] Allocation[n][m] Need[n][m] number of resources of Rm that are unallocated max demand of Pn for Rm number of Rm that are allocated to Pn number of Rm that may be needed by Pn
Note: Need[n, m] = Max[n, m] Allocation[n, m] as in C++, will treat a matrix as a vector of vectors (Max[i] is max resources demand for Pi) for vectors X & Y, X Y if and only if X[i] Y[i] for all i
12

Example: banker's algorithm

suppose have: 3 types of resources (6 of type A, 2 of type B, 2 of type C) 3 processes (P, Q, and R) Max A B C P 2 1 0 Q 3 1 1 R 5 1 2 Allocation A B C 0 1 0 2 0 0 3 0 2 Available A B C 1 1 0 Need A B C 2 0 0 1 1 1 2 1 0

is the system in a safe state?

13

Banker's: safety algorithm

to find out whether a system is in a safe state:
1. define vector: Finish[k] = false (for all k n) 2. find an i such that (Finish[i] == false && Need[i] Available) if no such i, then go to STEP 4 3. update: Available = Available + Allocation[i], Finish[i] = true go to STEP 2 4. if (Finish[i] == true) for all i, then the system is in a safe state otherwise, the system is in an unsafe state

Max A B C P 2 1 0 Q 3 1 1 R 5 1 2

Allocation A B C 0 1 0 2 0 0 2 0 1

Available A B C 2 1 1

Need A B C 2 0 0 1 1 1 3 1 1

is this state safe?

14

Banker's: resource request algorithm

let Request[i] be the request vector from Pi
1. if Request[i] > Need[i], then ERROR (asked for too many resources) & EXIT 2. if Request[i] > Available[i], then deny request and Pi must wait (the requested resources are not currently available) 3. otherwise, pretend to allocate the resources Available = Available Request[i] Allocation[i] = Allocation[i] + Request[i] Need[i] = Need[i] Request[i] 4. call the Safety algorithm, if safe then allocation is OK otherwise, the request is denied and the pretended allocation is undone Max A B C P 2 1 0 Q 3 1 1 R 5 1 2 Allocation A B C 0 1 0 2 0 0 2 0 1 Available A B C 2 1 1 Need A B C 2 0 0 1 1 1 3 1 1

what if R asks for [1 0 1] ? 15

Example: banker's algorithm

suppose 5 processes: P0 through P4 3 resource types: A (10), B (5), and C (7) snapshot at time T0:
Max A B C 7 5 3 3 2 2 9 0 2 2 2 2 4 3 3 Allocation A B C 0 1 0 2 0 0 3 0 2 2 1 1 0 0 2 Available A B C 3 3 2 Need A B C 7 4 3 1 2 2 6 0 0 0 1 1 4 3 1

P0 P1 P2 P3 P4

the system is in a safe state since the sequence P1, P3, P4, P2, P0 satisfies the safety criteria
16

Example: banker's algorithm (cont.)

P0 P1 P2 P3 P4 Max A B C 7 5 3 3 2 2 9 0 2 2 2 2 4 3 3 Allocation A B C 0 1 0 2 0 0 3 0 2 2 1 1 0 0 2 Available A B C 3 3 2 Need A B C 7 4 3 1 2 2 6 0 0 0 1 1 4 3 1

P1 requests [1 0 2]

1. Request[i] Need[i]: [1 0 2] [1 2 2], so no ERROR 2. Request[i] Available[i]: [1 0 2] [3 3 2], so resources are available 3. pretend to allocate the resources Allocation A B C 0 1 0 3 2 2 3 0 2 2 1 1 0 0 2 Available A B C 2 1 0 Need A B C 7 4 3 0 0 0 6 0 0 0 1 1 4 3 1

P0 P1 P2 P3 P4

Max A B C 7 5 3 3 2 2 9 0 2 2 2 2 4 3 3

the sequence P1, P3, P4, P0, P2 satisfies the safety 17 requirement

Example: banker's algorithm (cont.)

Max A B C 7 5 3 3 2 2 9 0 2 2 2 2 4 3 3 Allocation A B C 0 1 0 2 0 0 3 0 2 2 1 1 0 0 2 Available A B C 3 3 2 Need A B C 7 4 3 1 2 2 6 0 0 0 1 1 4 3 1

P0 P1 P2 P3 P4

P4 requests [3 3 0] ? P0 requests [0 2 0] ?

18

Deadlock detection
if there is only a single instance of each resource, can construct a wait-for graph out of the resource allocation graph
remove all resource nodes and collapse the edges

wait-for graph

a deadlock has occurred if and only if there is a cycle in the wait-for graph
to detect deadlock, need to periodically check for cycles can be accomplished in O(n2) operations, where n is number of processes
19

Deadlock detection algorithm

if multiple instances of resources, need test similar to safety algorithm
1. define vector: Finish[k] = (Allocation[k] == 0) 2. find an i such that (Finish[i] == false && Request[i] Available) if no such i, then go to STEP 4 3. update: Available = Available + Allocation[i], Finish[i] = true go to STEP 2 4. if (Finish[i] == false) for any i, then Pi is deadlocked otherwise, no deadlock Allocation A B C P0 0 1 0 P1 2 0 0 P2 3 0 3 P4 2 1 1 P5 0 0 2 Request A B C 0 0 0 2 0 2 0 0 0 1 0 0 0 0 2 Available A B C 0 0 0

not deadlocked P0, P2, P3, P1, P4 results in Finish[i] == true for all i
20

Deadlock detection algorithm (cont.)

1. define vector: Finish[k] = (Allocation[k] == 0) 2. find an i such that (Finish[i] == false && Request[i] Available) if no such i, then go to STEP 4 3. update: Available = Available + Allocation[i], Finish[i] = true go to STEP 2 4. if (Finish[i] == false) for any i, then Pi is deadlocked otherwise, no deadlock Allocation A B C P0 0 1 0 P1 2 0 0 P2 3 0 3 P4 2 1 1 P5 0 0 2 Request A B C 0 0 0 2 0 2 0 0 1 1 0 0 0 0 2 Available A B C 0 0 0

deadlock! can reclaim P0's resources, but not enough to meet requests of other processes
21

When to detect deadlock?

checking for deadlock is time consuming
given n processes, m resource types, deadlock detection is O(m*n2)

testing at every resource request would greatly slow down all requests

if deadlocks are rare or only affect a small number of processes, then may resort to testing periodically (e.g., once a day, when CPU use < 40%)
note that if testing is sporadic, there may be many cycles in the resource graph will not be able to tell which of the deadlocked processes caused the deadlock

22

Deadlock recovery
to recover from deadlock, must eliminate one of the necessary conditions
this involves preempting a resource or aborting a deadlocked process

preempting a resource
How do you select a victim? When preempt a resource, how do you roll back the process? How do you prevent starvation?

aborting a process
Do you abort all deadlocked processes or one at a time?

Note: it is possible to combine the three basic approaches

prevention + avoidance + detection

allowing the use of the optimal approach for each class of resources

23