Last week I got my grubby mitts on a clean, white and shiny new copy of The Linux Programming Interface ("A Linux and UNIX System Programming Handbook") by Michael Kerrisk.
So far I've read only a few chapters in detail, skimmed over one or two others, and dived around in it to look some things up. Just a week after receiving it is probably too soon for a fair review of such an enormous book, but Michael asked me if I would consider writing one (after I contacted him -- I don't know him personally), and it's the least I can do to thank him for what I can already tell is going to be a stupendously useful book.
From its uncluttered no-nonsense title and cover design, right down to its nitty-gritty technical details, it's obvious that this is a clear, well thought out, and well written book by someone who knows their subject matter inside and out. That's no surprise: Michael Kerrisk is maintainer and a major contributor to the Linux man-pages project. But let's be clear, this book isn't just a bunch of man pages glued together and given a cover!
Despite its large size (over 60 chapters, 1500+ pages) and level of detail, this is a surprisingly readable guidebook for UNIX system calls with a particular focus on Linux. The chapters are arranged such that they can be read in order with minimal forward-referencing. Topics covered range from the history of UNIX and fundamental systems concepts, through file i/o and file systems, processes & threads, IPC, up to advanced socket techniques and alternative I/O models. The book has a fairly comprehensive (although not exhaustive) index. The chapter list at man7.org provides a complete picture. The IPC chapters by themselves look as good as any other book I've seen on the subject.
The topics are covered in enough detail to make the book useful as a reference, but retaining a readable style throughout. It does a good job of pointing out some of the UNIX and C library quirks that might get lost in the detail of pure reference material. The book doesn't stray too far from the main focus -- the system calls themselves -- and provides references to other sources that cover related topics in more detail. There are plenty of diagrams and examples, including source code (available from man7.org). Source examples tend to be fairly simple illustrations of individual or related system calls. The book is not too code-heavy and doesn't just go through the motions of listing every possible call and parameter. After all, you already have the freely-available man pages for that. (It's occasionally useful to have a BASH prompt to hand while reading it.)
This isn't a beginner's book: a certain level of basic UNIX/Linux knowledge and a good grasp of C programming is a pre-requisite to a book like this. There is a good balance of introductory and advanced material without "dumbing down" or sacrificing of important detail.
The chapters on sockets, while quite short, cover a surprising amount of detailed ground, although a discussion of some of the more important TCP options available with setsockopt() is notably lacking. To be fair, this isn't a socket-programming book, and references to other sources of information are provided. However, I did think this was an odd omission considering the level of detail provided elsewhere in the book.
There are some exercises at the end of chapters with selected answers provided. I think this is one area that could be expanded and improved for readers who might want to treat the book like a `self-study' course or use it in an educational environment.
Any book this size with this level of technical detail inevitably has some errors, and it was after reporting one (minor) mistake that the author asked if I would consider posting a review. However, I was apparently only the second person to report one! Considering how many eyes must have glanced over this book already, that surely says something about the book's accuracy. The book has its own web page on the author's site (man7.org) where there is an errata section and a list of kernel changes since the book was released. It looks like this book will be well supported, even as the Linux kernel continues to march on its evolutionary path.
As you can probably tell from the overall tone of this review, I'm very pleased with my purchase. I plan to read this cover-to-cover over the next couple of months, and I will probably get a second copy on expenses to sit on my desk at work! I may post a follow-up review later, once I've had more time to read and reflect.
I would recommend this book to any Linux programmer.