Check: Linux, filesystem deep dive

The first character of an ls -l line is the type. l means symbolic link. A regular file starts with -, a directory with d, a block device with b. The full table of seven types is in the file types lesson, section "The seven file types".

A hard link is just another name for the same inode. All names are equal, and the data stays alive as long as the inode has at least one name left. A symlink is a separate small file with its own inode, and inside it stores a path. Remove the target, and the symlink goes dangling: the path is still there, but there's no file at that path anymore. It's symlinks that work across filesystems, a hard link can't reach that far.

The kernel assembles the contents of /proc at the moment you read it, there are no such files on disk. The numbered directories are processes: /proc/<PID>/cmdline has the launch command, /proc/<PID>/status has memory and state, /proc/<PID>/fd has open files. That's why you can find processes even on a server with no ps. This comes in handy in the mini-project.

df -h / only tells you about the root filesystem. A directory inside the tree can be a mount point for another filesystem, with its own amount of space. A tmpfs with a size=16m limit runs out at 16 MB, no matter how much is free on disk. df -h <directory> and findmnt -T <directory> show what the directory is actually sitting on. Swap has nothing to do with disk space: it's a reserve for RAM.

/dev/sdb1 in fstab does work, but it depends on the order the kernel found the disks in. Add a controller or move the VM, and sdb can become sdc. Then at boot the wrong partition gets mounted, or nothing mounts at all. A UUID is written into the filesystem itself and survives that kind of reshuffle. You can look up the UUID with lsblk -o NAME,UUID,MOUNTPOINTS or blkid.

Swap is a partition or file on disk where the kernel pushes out rarely used memory pages, to hand the freed RAM to active processes. A program doesn't run from swap: when a page is needed again, the kernel brings it back into RAM. Disk is far slower than memory, so a large swap doesn't make the system faster. It only gives you a reserve to survive a spike in memory use without the OOM killer.

RAID 5 spreads data with parity across all disks. At least three disks, usable capacity (n-1)/n, it survives one disk failure. RAID 0 survives no failures at all. RAID 1 survives a failure, but is built from just two disks. RAID 10 needs at least four. After the first failure, RAID 5's reserve is spent: the next failure before the rebuild finishes takes down the whole array.