In my case, with a faulty new SSD:
tl;dr for the test I gave the entire device (not a partition) to UFS for the stress test.
You can not. Disks can and will fail spontaneously. Sometimes they give explicit warnings (if you have software running that looks at their PFA = predictive fault analysis, which can be done on SCSI, it's harder on SATA). Sometimes smartctl data will give you a hint or clue to that might be failing. But often enough, they fail without prior warning.
When it arrived the packaging said 'Hard Drive Case (M.2)' so I thought I had been cheated but when plugged in gpart identified it as a 2TB device.
www.samsung.com
I guess you get what you pay for. Mine seemed like a bargain at the time.Portable SSD T9 USB 3.2 Gen2x2 4TB (Black) | Samsung US
Actually, there is a cleaner test to do that. Use camcontrol (or a SCSI "read capacity" command through sgutils) to ask the device itself what its capacity is. If the capacity advertises itself as 64GiB, you are done. If the capacity advertises itself as 2TiB, it is pretty easy to test the actual capacity: Write a small script or program that writes 65 GiB worth of random sectors (each sector with different data) to the device, in random places spread over the whole 2TiB address space. Mark in each sector's data what sector number it was written to. Remember (for example in a relatively small file) where you wrote them. After writing them, read them all back. If all the 65GiB's worth of sectors are readable and correct, then you know that your device has more than 64GiB worth of capacity. And since you wrote random data, any dedup or compression trick won't help the device. You can use this trick to measure the actual capacity of a device, by interleaving reads and writes. I used to have a pretty fancy program that did this (and lots of other performance/functionality tests), but the source code is with an ex-employer.
