How to check a download against its checksum
- Drop the file on the page. Several files at once work too, and each gets its own block.
- Leave MD5 and SHA-256 ticked, or tick the one the download page names.
- Paste the published checksum into the compare box. You can paste a single value, a
value with a prefix such as
sha256:, or the whole SHA256SUMS file that many projects publish next to their releases. - Read the verdict. Match means every byte is the same. Anything else tells you the character where the two values first differ.
The algorithm is worked out from the length of the value you paste: 8 hexadecimal
characters is CRC-32, 32 is MD5, 40 is SHA-1, 64 is SHA-256 and 128 is SHA-512. If you
paste a SHA-1 value while only SHA-256 is ticked, the page adds SHA-1 by itself. When the
pasted text is a list, each file is compared with the line that carries its name, in
either the sha256sum layout or the BSD one, SHA256 (name) = value.
No file size limit
The browser's built-in digest function takes the whole input in one call, so a page built on it has to hold the entire file in memory. That is where the caps come from: at least one of the pages that rank for this search states a 50 MB limit for files. This page does not use it. MD5, SHA-1, SHA-256, SHA-512 and CRC-32 are written out here as incremental hashes, and the file is read 8 MB at a time and fed through as it goes. Memory use stays at one piece whatever the size. A 320 MiB test file was hashed this way with MD5 and SHA-256 together in about 3.5 seconds on the machine these pages are built on, and the SHA-256 matched Python's.
How it was checked
Hand-written hash code is easy to get subtly wrong, and the errors cluster at the edges.
Every algorithm here was compared with Python's hashlib and zlib
on inputs of 23 different lengths, including the ones on either side of each block
boundary — 55, 56, 57, 63, 64, 65 bytes for the 64-byte algorithms and 111 to 129
for SHA-512, whose blocks are 128 bytes — plus the empty input. All 115 comparisons
agree. (The first version of SHA-512 did not: one intermediate value was treated as a
signed number, and every SHA-512 result was wrong until that test caught it.)
On the same machine, in Chrome, the speeds came out at roughly 280 MiB/s for MD5, 230 for SHA-1, 155 for SHA-256, 80 for CRC-32 and 75 for SHA-512. SHA-512 is slower because JavaScript has no 64-bit integer arithmetic on ordinary numbers, so each 64-bit word is handled as two 32-bit halves. Your device will differ; the order will not.
Why a text hash differs from the command line
The most common surprise is a line break. The SHA-256 of hello is
2cf24dba…9824, and the SHA-256 of hello followed by a
newline is 5891b5b5…be03 — completely different. The shell
command echo hello prints the newline, so hashing its output gives the
second value; echo -n hello or printf hello gives the first.
The page tells you when your text ends with a line break so you can see which one you
are hashing.
Text is hashed as UTF-8. An é is two bytes and a Chinese character is
three, so the same visible text saved in another encoding produces a different hash.
The size line above the result shows the byte count, not the character count.
Which algorithm to pick
| Algorithm | Length | Use it for |
|---|---|---|
| SHA-256 | 64 characters | Verifying downloads; the default choice |
| SHA-512 | 128 characters | When a project publishes it; not stronger in practice for this job |
| SHA-1 | 40 characters | Older release lists, Git object names |
| MD5 | 32 characters | Catching accidental corruption only |
| CRC-32 | 8 characters | Comparing with a ZIP or PNG checksum; accidental errors only |
MD5 and SHA-1 are fine for noticing that a download was cut short. They are not fine for proving that nobody swapped the file, because collisions can be manufactured for both. A checksum only protects you if it comes from somewhere the attacker could not also change — a hash printed on the same page as the file adds nothing against someone who controls that page.
Related tools
Packing files together before you send them? The ZIP maker checks each file's CRC-32 after packing; the same CRC-32 option is here for comparing a file with an archive listing. To see what changed between two versions of a text file rather than whether they differ at all, use text diff. Nothing you drop on this page is uploaded.