Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:100 |
2 | * This is the Launcher code, a simple program which lays out the "physical" | |
3 | * memory for the new Guest by mapping the kernel image and the virtual | |
4 | * devices, then opens /dev/lguest to tell the kernel about the Guest and | |
5 | * control it. | |
6 | :*/ | |
8ca47e00 RR |
7 | #define _LARGEFILE64_SOURCE |
8 | #define _GNU_SOURCE | |
9 | #include <stdio.h> | |
10 | #include <string.h> | |
11 | #include <unistd.h> | |
12 | #include <err.h> | |
13 | #include <stdint.h> | |
14 | #include <stdlib.h> | |
15 | #include <elf.h> | |
16 | #include <sys/mman.h> | |
6649bb7a | 17 | #include <sys/param.h> |
8ca47e00 RR |
18 | #include <sys/types.h> |
19 | #include <sys/stat.h> | |
20 | #include <sys/wait.h> | |
659a0e66 | 21 | #include <sys/eventfd.h> |
8ca47e00 RR |
22 | #include <fcntl.h> |
23 | #include <stdbool.h> | |
24 | #include <errno.h> | |
25 | #include <ctype.h> | |
26 | #include <sys/socket.h> | |
27 | #include <sys/ioctl.h> | |
28 | #include <sys/time.h> | |
29 | #include <time.h> | |
30 | #include <netinet/in.h> | |
31 | #include <net/if.h> | |
32 | #include <linux/sockios.h> | |
33 | #include <linux/if_tun.h> | |
34 | #include <sys/uio.h> | |
35 | #include <termios.h> | |
36 | #include <getopt.h> | |
17cbca2b RR |
37 | #include <assert.h> |
38 | #include <sched.h> | |
a586d4f6 RR |
39 | #include <limits.h> |
40 | #include <stddef.h> | |
a161883a | 41 | #include <signal.h> |
8aeb36e8 PS |
42 | #include <pwd.h> |
43 | #include <grp.h> | |
c565650b | 44 | #include <sys/user.h> |
d7fbf6e9 | 45 | #include <linux/pci_regs.h> |
8aeb36e8 | 46 | |
927cfb97 RR |
47 | #ifndef VIRTIO_F_ANY_LAYOUT |
48 | #define VIRTIO_F_ANY_LAYOUT 27 | |
49 | #endif | |
50 | ||
2e04ef76 | 51 | /*L:110 |
9f54288d | 52 | * We can ignore the 43 include files we need for this program, but I do want |
2e04ef76 | 53 | * to draw attention to the use of kernel-style types. |
db24e8c2 RR |
54 | * |
55 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
56 | * like these abbreviations, so we define them here. Note that u64 is always | |
57 | * unsigned long long, which works on all Linux systems: this means that we can | |
2e04ef76 RR |
58 | * use %llu in printf for any u64. |
59 | */ | |
db24e8c2 RR |
60 | typedef unsigned long long u64; |
61 | typedef uint32_t u32; | |
62 | typedef uint16_t u16; | |
63 | typedef uint8_t u8; | |
dde79789 | 64 | /*:*/ |
8ca47e00 | 65 | |
eb39f833 | 66 | #define VIRTIO_CONFIG_NO_LEGACY |
93153077 | 67 | #define VIRTIO_PCI_NO_LEGACY |
50516547 | 68 | #define VIRTIO_BLK_NO_LEGACY |
206ad06b | 69 | #define VIRTIO_NET_NO_LEGACY |
93153077 RR |
70 | |
71 | /* Use in-kernel ones, which defines VIRTIO_F_VERSION_1 */ | |
72 | #include "../../include/uapi/linux/virtio_config.h" | |
bf6d4034 | 73 | #include "../../include/uapi/linux/virtio_net.h" |
50516547 | 74 | #include "../../include/uapi/linux/virtio_blk.h" |
e8330d9b | 75 | #include "../../include/uapi/linux/virtio_console.h" |
0d5b5d39 | 76 | #include "../../include/uapi/linux/virtio_rng.h" |
e6dc0418 | 77 | #include <linux/virtio_ring.h> |
93153077 | 78 | #include "../../include/uapi/linux/virtio_pci.h" |
e6dc0418 RR |
79 | #include <asm/bootparam.h> |
80 | #include "../../include/linux/lguest_launcher.h" | |
81 | ||
8ca47e00 RR |
82 | #define BRIDGE_PFX "bridge:" |
83 | #ifndef SIOCBRADDIF | |
84 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
85 | #endif | |
3c6b5bfa RR |
86 | /* We can have up to 256 pages for devices. */ |
87 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
88 | /* This will occupy 3 pages: it must be a power of 2. */ |
89 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 90 | |
2e04ef76 RR |
91 | /*L:120 |
92 | * verbose is both a global flag and a macro. The C preprocessor allows | |
93 | * this, and although I wouldn't recommend it, it works quite nicely here. | |
94 | */ | |
8ca47e00 RR |
95 | static bool verbose; |
96 | #define verbose(args...) \ | |
97 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
98 | /*:*/ |
99 | ||
3c6b5bfa RR |
100 | /* The pointer to the start of guest memory. */ |
101 | static void *guest_base; | |
102 | /* The maximum guest physical address allowed, and maximum possible. */ | |
0a6bcc18 | 103 | static unsigned long guest_limit, guest_max, guest_mmio; |
56739c80 RR |
104 | /* The /dev/lguest file descriptor. */ |
105 | static int lguest_fd; | |
8ca47e00 | 106 | |
e3283fa0 GOC |
107 | /* a per-cpu variable indicating whose vcpu is currently running */ |
108 | static unsigned int __thread cpu_id; | |
109 | ||
6a54f9ab RR |
110 | /* 5 bit device number in the PCI_CONFIG_ADDR => 32 only */ |
111 | #define MAX_PCI_DEVICES 32 | |
112 | ||
dde79789 | 113 | /* This is our list of devices. */ |
1842f23c | 114 | struct device_list { |
17cbca2b RR |
115 | /* Counter to assign interrupt numbers. */ |
116 | unsigned int next_irq; | |
117 | ||
118 | /* Counter to print out convenient device numbers. */ | |
119 | unsigned int device_num; | |
120 | ||
6a54f9ab RR |
121 | /* PCI devices. */ |
122 | struct device *pci[MAX_PCI_DEVICES]; | |
8ca47e00 RR |
123 | }; |
124 | ||
17cbca2b RR |
125 | /* The list of Guest devices, based on command line arguments. */ |
126 | static struct device_list devices; | |
127 | ||
93153077 RR |
128 | struct virtio_pci_cfg_cap { |
129 | struct virtio_pci_cap cap; | |
b2ce1ea4 | 130 | u32 pci_cfg_data; /* Data for BAR access. */ |
93153077 RR |
131 | }; |
132 | ||
133 | struct virtio_pci_mmio { | |
134 | struct virtio_pci_common_cfg cfg; | |
135 | u16 notify; | |
136 | u8 isr; | |
137 | u8 padding; | |
138 | /* Device-specific configuration follows this. */ | |
139 | }; | |
140 | ||
d7fbf6e9 RR |
141 | /* This is the layout (little-endian) of the PCI config space. */ |
142 | struct pci_config { | |
143 | u16 vendor_id, device_id; | |
144 | u16 command, status; | |
145 | u8 revid, prog_if, subclass, class; | |
146 | u8 cacheline_size, lat_timer, header_type, bist; | |
147 | u32 bar[6]; | |
148 | u32 cardbus_cis_ptr; | |
149 | u16 subsystem_vendor_id, subsystem_device_id; | |
150 | u32 expansion_rom_addr; | |
151 | u8 capabilities, reserved1[3]; | |
152 | u32 reserved2; | |
153 | u8 irq_line, irq_pin, min_grant, max_latency; | |
93153077 RR |
154 | |
155 | /* Now, this is the linked capability list. */ | |
156 | struct virtio_pci_cap common; | |
157 | struct virtio_pci_notify_cap notify; | |
158 | struct virtio_pci_cap isr; | |
159 | struct virtio_pci_cap device; | |
93153077 | 160 | struct virtio_pci_cfg_cap cfg_access; |
d7fbf6e9 RR |
161 | }; |
162 | ||
dde79789 | 163 | /* The device structure describes a single device. */ |
1842f23c | 164 | struct device { |
17cbca2b RR |
165 | /* The name of this device, for --verbose. */ |
166 | const char *name; | |
8ca47e00 | 167 | |
17cbca2b RR |
168 | /* Any queues attached to this device */ |
169 | struct virtqueue *vq; | |
8ca47e00 | 170 | |
659a0e66 RR |
171 | /* Is it operational */ |
172 | bool running; | |
a007a751 | 173 | |
d39a6785 RR |
174 | /* Has it written FEATURES_OK but not re-checked it? */ |
175 | bool wrote_features_ok; | |
176 | ||
d7fbf6e9 RR |
177 | /* PCI configuration */ |
178 | union { | |
179 | struct pci_config config; | |
180 | u32 config_words[sizeof(struct pci_config) / sizeof(u32)]; | |
181 | }; | |
182 | ||
93153077 RR |
183 | /* Features we offer, and those accepted. */ |
184 | u64 features, features_accepted; | |
185 | ||
d7fbf6e9 RR |
186 | /* Device-specific config hangs off the end of this. */ |
187 | struct virtio_pci_mmio *mmio; | |
188 | ||
6a54f9ab RR |
189 | /* PCI MMIO resources (all in BAR0) */ |
190 | size_t mmio_size; | |
191 | u32 mmio_addr; | |
192 | ||
8ca47e00 RR |
193 | /* Device-specific data. */ |
194 | void *priv; | |
195 | }; | |
196 | ||
17cbca2b | 197 | /* The virtqueue structure describes a queue attached to a device. */ |
1842f23c | 198 | struct virtqueue { |
17cbca2b RR |
199 | struct virtqueue *next; |
200 | ||
201 | /* Which device owns me. */ | |
202 | struct device *dev; | |
203 | ||
17c56d6d RR |
204 | /* Name for printing errors. */ |
205 | const char *name; | |
206 | ||
17cbca2b RR |
207 | /* The actual ring of buffers. */ |
208 | struct vring vring; | |
209 | ||
93153077 RR |
210 | /* The information about this virtqueue (we only use queue_size on) */ |
211 | struct virtio_pci_common_cfg pci_config; | |
212 | ||
17cbca2b RR |
213 | /* Last available index we saw. */ |
214 | u16 last_avail_idx; | |
215 | ||
95c517c0 RR |
216 | /* How many are used since we sent last irq? */ |
217 | unsigned int pending_used; | |
218 | ||
659a0e66 RR |
219 | /* Eventfd where Guest notifications arrive. */ |
220 | int eventfd; | |
20887611 | 221 | |
659a0e66 RR |
222 | /* Function for the thread which is servicing this virtqueue. */ |
223 | void (*service)(struct virtqueue *vq); | |
224 | pid_t thread; | |
17cbca2b RR |
225 | }; |
226 | ||
ec04b13f BR |
227 | /* Remember the arguments to the program so we can "reboot" */ |
228 | static char **main_args; | |
229 | ||
659a0e66 RR |
230 | /* The original tty settings to restore on exit. */ |
231 | static struct termios orig_term; | |
232 | ||
2e04ef76 RR |
233 | /* |
234 | * We have to be careful with barriers: our devices are all run in separate | |
f7027c63 | 235 | * threads and so we need to make sure that changes visible to the Guest happen |
2e04ef76 RR |
236 | * in precise order. |
237 | */ | |
f7027c63 | 238 | #define wmb() __asm__ __volatile__("" : : : "memory") |
0d69a65e RR |
239 | #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") |
240 | #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") | |
17cbca2b | 241 | |
b5111790 RR |
242 | /* Wrapper for the last available index. Makes it easier to change. */ |
243 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
244 | ||
2e04ef76 RR |
245 | /* |
246 | * The virtio configuration space is defined to be little-endian. x86 is | |
247 | * little-endian too, but it's nice to be explicit so we have these helpers. | |
248 | */ | |
17cbca2b RR |
249 | #define cpu_to_le16(v16) (v16) |
250 | #define cpu_to_le32(v32) (v32) | |
251 | #define cpu_to_le64(v64) (v64) | |
252 | #define le16_to_cpu(v16) (v16) | |
253 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 254 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 255 | |
1e1c17a7 RR |
256 | /* |
257 | * A real device would ignore weird/non-compliant driver behaviour. We | |
258 | * stop and flag it, to help debugging Linux problems. | |
259 | */ | |
260 | #define bad_driver(d, fmt, ...) \ | |
261 | errx(1, "%s: bad driver: " fmt, (d)->name, ## __VA_ARGS__) | |
262 | #define bad_driver_vq(vq, fmt, ...) \ | |
263 | errx(1, "%s vq %s: bad driver: " fmt, (vq)->dev->name, \ | |
264 | vq->name, ## __VA_ARGS__) | |
265 | ||
28fd6d7f RR |
266 | /* Is this iovec empty? */ |
267 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
268 | { | |
269 | unsigned int i; | |
270 | ||
271 | for (i = 0; i < num_iov; i++) | |
272 | if (iov[i].iov_len) | |
273 | return false; | |
274 | return true; | |
275 | } | |
276 | ||
277 | /* Take len bytes from the front of this iovec. */ | |
1e1c17a7 RR |
278 | static void iov_consume(struct device *d, |
279 | struct iovec iov[], unsigned num_iov, | |
c0316a94 | 280 | void *dest, unsigned len) |
28fd6d7f RR |
281 | { |
282 | unsigned int i; | |
283 | ||
284 | for (i = 0; i < num_iov; i++) { | |
285 | unsigned int used; | |
286 | ||
287 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
c0316a94 RR |
288 | if (dest) { |
289 | memcpy(dest, iov[i].iov_base, used); | |
290 | dest += used; | |
291 | } | |
28fd6d7f RR |
292 | iov[i].iov_base += used; |
293 | iov[i].iov_len -= used; | |
294 | len -= used; | |
295 | } | |
c0316a94 | 296 | if (len != 0) |
1e1c17a7 | 297 | bad_driver(d, "iovec too short!"); |
28fd6d7f RR |
298 | } |
299 | ||
2e04ef76 RR |
300 | /*L:100 |
301 | * The Launcher code itself takes us out into userspace, that scary place where | |
302 | * pointers run wild and free! Unfortunately, like most userspace programs, | |
303 | * it's quite boring (which is why everyone likes to hack on the kernel!). | |
304 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get | |
305 | * you through this section. Or, maybe not. | |
3c6b5bfa RR |
306 | * |
307 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
308 | * memory and stores it in "guest_base". In other words, Guest physical == | |
309 | * Launcher virtual with an offset. | |
310 | * | |
311 | * This can be tough to get your head around, but usually it just means that we | |
a33f3224 | 312 | * use these trivial conversion functions when the Guest gives us its |
2e04ef76 RR |
313 | * "physical" addresses: |
314 | */ | |
3c6b5bfa RR |
315 | static void *from_guest_phys(unsigned long addr) |
316 | { | |
317 | return guest_base + addr; | |
318 | } | |
319 | ||
320 | static unsigned long to_guest_phys(const void *addr) | |
321 | { | |
322 | return (addr - guest_base); | |
323 | } | |
324 | ||
dde79789 RR |
325 | /*L:130 |
326 | * Loading the Kernel. | |
327 | * | |
328 | * We start with couple of simple helper routines. open_or_die() avoids | |
2e04ef76 RR |
329 | * error-checking code cluttering the callers: |
330 | */ | |
8ca47e00 RR |
331 | static int open_or_die(const char *name, int flags) |
332 | { | |
333 | int fd = open(name, flags); | |
334 | if (fd < 0) | |
335 | err(1, "Failed to open %s", name); | |
336 | return fd; | |
337 | } | |
338 | ||
3c6b5bfa RR |
339 | /* map_zeroed_pages() takes a number of pages. */ |
340 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 341 | { |
3c6b5bfa RR |
342 | int fd = open_or_die("/dev/zero", O_RDONLY); |
343 | void *addr; | |
8ca47e00 | 344 | |
2e04ef76 RR |
345 | /* |
346 | * We use a private mapping (ie. if we write to the page, it will be | |
5230ff0c PS |
347 | * copied). We allocate an extra two pages PROT_NONE to act as guard |
348 | * pages against read/write attempts that exceed allocated space. | |
2e04ef76 | 349 | */ |
5230ff0c PS |
350 | addr = mmap(NULL, getpagesize() * (num+2), |
351 | PROT_NONE, MAP_PRIVATE, fd, 0); | |
352 | ||
3c6b5bfa | 353 | if (addr == MAP_FAILED) |
af901ca1 | 354 | err(1, "Mmapping %u pages of /dev/zero", num); |
a91d74a3 | 355 | |
5230ff0c PS |
356 | if (mprotect(addr + getpagesize(), getpagesize() * num, |
357 | PROT_READ|PROT_WRITE) == -1) | |
358 | err(1, "mprotect rw %u pages failed", num); | |
359 | ||
a91d74a3 RR |
360 | /* |
361 | * One neat mmap feature is that you can close the fd, and it | |
362 | * stays mapped. | |
363 | */ | |
34bdaab4 | 364 | close(fd); |
3c6b5bfa | 365 | |
5230ff0c PS |
366 | /* Return address after PROT_NONE page */ |
367 | return addr + getpagesize(); | |
3c6b5bfa RR |
368 | } |
369 | ||
0a6bcc18 RR |
370 | /* Get some bytes which won't be mapped into the guest. */ |
371 | static unsigned long get_mmio_region(size_t size) | |
372 | { | |
373 | unsigned long addr = guest_mmio; | |
374 | size_t i; | |
375 | ||
376 | if (!size) | |
377 | return addr; | |
378 | ||
379 | /* Size has to be a power of 2 (and multiple of 16) */ | |
380 | for (i = 1; i < size; i <<= 1); | |
381 | ||
382 | guest_mmio += i; | |
383 | ||
384 | return addr; | |
385 | } | |
386 | ||
2e04ef76 RR |
387 | /* |
388 | * This routine is used to load the kernel or initrd. It tries mmap, but if | |
6649bb7a | 389 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
2e04ef76 RR |
390 | * it falls back to reading the memory in. |
391 | */ | |
6649bb7a RM |
392 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
393 | { | |
394 | ssize_t r; | |
395 | ||
2e04ef76 RR |
396 | /* |
397 | * We map writable even though for some segments are marked read-only. | |
6649bb7a RM |
398 | * The kernel really wants to be writable: it patches its own |
399 | * instructions. | |
400 | * | |
401 | * MAP_PRIVATE means that the page won't be copied until a write is | |
402 | * done to it. This allows us to share untouched memory between | |
2e04ef76 RR |
403 | * Guests. |
404 | */ | |
5230ff0c | 405 | if (mmap(addr, len, PROT_READ|PROT_WRITE, |
6649bb7a RM |
406 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) |
407 | return; | |
408 | ||
409 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
410 | r = pread(fd, addr, len, offset); | |
411 | if (r != len) | |
412 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
413 | } | |
414 | ||
2e04ef76 RR |
415 | /* |
416 | * This routine takes an open vmlinux image, which is in ELF, and maps it into | |
dde79789 RR |
417 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
418 | * by all modern binaries on Linux including the kernel. | |
419 | * | |
420 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
421 | * address. We use the physical address; the Guest will map itself to the |
422 | * virtual address. | |
dde79789 | 423 | * |
2e04ef76 RR |
424 | * We return the starting address. |
425 | */ | |
47436aa4 | 426 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 427 | { |
8ca47e00 RR |
428 | Elf32_Phdr phdr[ehdr->e_phnum]; |
429 | unsigned int i; | |
8ca47e00 | 430 | |
2e04ef76 RR |
431 | /* |
432 | * Sanity checks on the main ELF header: an x86 executable with a | |
433 | * reasonable number of correctly-sized program headers. | |
434 | */ | |
8ca47e00 RR |
435 | if (ehdr->e_type != ET_EXEC |
436 | || ehdr->e_machine != EM_386 | |
437 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
438 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
439 | errx(1, "Malformed elf header"); | |
440 | ||
2e04ef76 RR |
441 | /* |
442 | * An ELF executable contains an ELF header and a number of "program" | |
dde79789 | 443 | * headers which indicate which parts ("segments") of the program to |
2e04ef76 RR |
444 | * load where. |
445 | */ | |
dde79789 RR |
446 | |
447 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
448 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
449 | err(1, "Seeking to program headers"); | |
450 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
451 | err(1, "Reading program headers"); | |
452 | ||
2e04ef76 RR |
453 | /* |
454 | * Try all the headers: there are usually only three. A read-only one, | |
455 | * a read-write one, and a "note" section which we don't load. | |
456 | */ | |
8ca47e00 | 457 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 458 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
459 | if (phdr[i].p_type != PT_LOAD) |
460 | continue; | |
461 | ||
462 | verbose("Section %i: size %i addr %p\n", | |
463 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
464 | ||
6649bb7a | 465 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 466 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 467 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
468 | } |
469 | ||
814a0e5c RR |
470 | /* The entry point is given in the ELF header. */ |
471 | return ehdr->e_entry; | |
8ca47e00 RR |
472 | } |
473 | ||
2e04ef76 RR |
474 | /*L:150 |
475 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed | |
476 | * to jump into it and it will unpack itself. We used to have to perform some | |
477 | * hairy magic because the unpacking code scared me. | |
dde79789 | 478 | * |
5bbf89fc RR |
479 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
480 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
2e04ef76 RR |
481 | * the funky header so we know where in the file to load, and away we go! |
482 | */ | |
47436aa4 | 483 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 484 | { |
43d33b21 | 485 | struct boot_params boot; |
5bbf89fc RR |
486 | int r; |
487 | /* Modern bzImages get loaded at 1M. */ | |
488 | void *p = from_guest_phys(0x100000); | |
489 | ||
2e04ef76 RR |
490 | /* |
491 | * Go back to the start of the file and read the header. It should be | |
395cf969 | 492 | * a Linux boot header (see Documentation/x86/boot.txt) |
2e04ef76 | 493 | */ |
5bbf89fc | 494 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 495 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 496 | |
43d33b21 RR |
497 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
498 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
499 | errx(1, "This doesn't look like a bzImage to me"); |
500 | ||
43d33b21 RR |
501 | /* Skip over the extra sectors of the header. */ |
502 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
503 | |
504 | /* Now read everything into memory. in nice big chunks. */ | |
505 | while ((r = read(fd, p, 65536)) > 0) | |
506 | p += r; | |
507 | ||
43d33b21 RR |
508 | /* Finally, code32_start tells us where to enter the kernel. */ |
509 | return boot.hdr.code32_start; | |
8ca47e00 RR |
510 | } |
511 | ||
2e04ef76 RR |
512 | /*L:140 |
513 | * Loading the kernel is easy when it's a "vmlinux", but most kernels | |
e1e72965 | 514 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
2e04ef76 RR |
515 | * work, we can load those, too. |
516 | */ | |
47436aa4 | 517 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
518 | { |
519 | Elf32_Ehdr hdr; | |
520 | ||
dde79789 | 521 | /* Read in the first few bytes. */ |
8ca47e00 RR |
522 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
523 | err(1, "Reading kernel"); | |
524 | ||
dde79789 | 525 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 526 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 527 | return map_elf(fd, &hdr); |
8ca47e00 | 528 | |
a6bd8e13 | 529 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 530 | return load_bzimage(fd); |
8ca47e00 RR |
531 | } |
532 | ||
2e04ef76 RR |
533 | /* |
534 | * This is a trivial little helper to align pages. Andi Kleen hated it because | |
dde79789 RR |
535 | * it calls getpagesize() twice: "it's dumb code." |
536 | * | |
537 | * Kernel guys get really het up about optimization, even when it's not | |
2e04ef76 RR |
538 | * necessary. I leave this code as a reaction against that. |
539 | */ | |
8ca47e00 RR |
540 | static inline unsigned long page_align(unsigned long addr) |
541 | { | |
dde79789 | 542 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
543 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
544 | } | |
545 | ||
2e04ef76 RR |
546 | /*L:180 |
547 | * An "initial ram disk" is a disk image loaded into memory along with the | |
548 | * kernel which the kernel can use to boot from without needing any drivers. | |
549 | * Most distributions now use this as standard: the initrd contains the code to | |
550 | * load the appropriate driver modules for the current machine. | |
dde79789 RR |
551 | * |
552 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
2e04ef76 RR |
553 | * kernels. He sent me this (and tells me when I break it). |
554 | */ | |
8ca47e00 RR |
555 | static unsigned long load_initrd(const char *name, unsigned long mem) |
556 | { | |
557 | int ifd; | |
558 | struct stat st; | |
559 | unsigned long len; | |
8ca47e00 RR |
560 | |
561 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 562 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
563 | if (fstat(ifd, &st) < 0) |
564 | err(1, "fstat() on initrd '%s'", name); | |
565 | ||
2e04ef76 RR |
566 | /* |
567 | * We map the initrd at the top of memory, but mmap wants it to be | |
568 | * page-aligned, so we round the size up for that. | |
569 | */ | |
8ca47e00 | 570 | len = page_align(st.st_size); |
3c6b5bfa | 571 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
2e04ef76 RR |
572 | /* |
573 | * Once a file is mapped, you can close the file descriptor. It's a | |
574 | * little odd, but quite useful. | |
575 | */ | |
8ca47e00 | 576 | close(ifd); |
6649bb7a | 577 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
578 | |
579 | /* We return the initrd size. */ | |
8ca47e00 RR |
580 | return len; |
581 | } | |
e1e72965 | 582 | /*:*/ |
8ca47e00 | 583 | |
2e04ef76 RR |
584 | /* |
585 | * Simple routine to roll all the commandline arguments together with spaces | |
586 | * between them. | |
587 | */ | |
8ca47e00 RR |
588 | static void concat(char *dst, char *args[]) |
589 | { | |
590 | unsigned int i, len = 0; | |
591 | ||
592 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
593 | if (i) { |
594 | strcat(dst+len, " "); | |
595 | len++; | |
596 | } | |
8ca47e00 | 597 | strcpy(dst+len, args[i]); |
1ef36fa6 | 598 | len += strlen(args[i]); |
8ca47e00 RR |
599 | } |
600 | /* In case it's empty. */ | |
601 | dst[len] = '\0'; | |
602 | } | |
603 | ||
2e04ef76 RR |
604 | /*L:185 |
605 | * This is where we actually tell the kernel to initialize the Guest. We | |
e1e72965 | 606 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
58a24566 | 607 | * the base of Guest "physical" memory, the top physical page to allow and the |
2e04ef76 RR |
608 | * entry point for the Guest. |
609 | */ | |
56739c80 | 610 | static void tell_kernel(unsigned long start) |
8ca47e00 | 611 | { |
511801dc JS |
612 | unsigned long args[] = { LHREQ_INITIALIZE, |
613 | (unsigned long)guest_base, | |
7313d521 | 614 | guest_limit / getpagesize(), start, |
0a6bcc18 RR |
615 | (guest_mmio+getpagesize()-1) / getpagesize() }; |
616 | verbose("Guest: %p - %p (%#lx, MMIO %#lx)\n", | |
617 | guest_base, guest_base + guest_limit, | |
618 | guest_limit, guest_mmio); | |
56739c80 RR |
619 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
620 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 621 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 622 | } |
dde79789 | 623 | /*:*/ |
8ca47e00 | 624 | |
a91d74a3 | 625 | /*L:200 |
dde79789 RR |
626 | * Device Handling. |
627 | * | |
e1e72965 | 628 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 629 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 630 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
631 | * if something funny is going on: |
632 | */ | |
1e1c17a7 RR |
633 | static void *_check_pointer(struct device *d, |
634 | unsigned long addr, unsigned int size, | |
8ca47e00 RR |
635 | unsigned int line) |
636 | { | |
2e04ef76 | 637 | /* |
5230ff0c PS |
638 | * Check if the requested address and size exceeds the allocated memory, |
639 | * or addr + size wraps around. | |
2e04ef76 | 640 | */ |
5230ff0c | 641 | if ((addr + size) > guest_limit || (addr + size) < addr) |
1e1c17a7 RR |
642 | bad_driver(d, "%s:%i: Invalid address %#lx", |
643 | __FILE__, line, addr); | |
2e04ef76 RR |
644 | /* |
645 | * We return a pointer for the caller's convenience, now we know it's | |
646 | * safe to use. | |
647 | */ | |
3c6b5bfa | 648 | return from_guest_phys(addr); |
8ca47e00 | 649 | } |
dde79789 | 650 | /* A macro which transparently hands the line number to the real function. */ |
1e1c17a7 | 651 | #define check_pointer(d,addr,size) _check_pointer(d, addr, size, __LINE__) |
8ca47e00 | 652 | |
2e04ef76 RR |
653 | /* |
654 | * Each buffer in the virtqueues is actually a chain of descriptors. This | |
e1e72965 | 655 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
2e04ef76 RR |
656 | * at the end. |
657 | */ | |
1e1c17a7 | 658 | static unsigned next_desc(struct device *d, struct vring_desc *desc, |
d1f0132e | 659 | unsigned int i, unsigned int max) |
17cbca2b RR |
660 | { |
661 | unsigned int next; | |
662 | ||
663 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
664 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
665 | return max; | |
17cbca2b RR |
666 | |
667 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 668 | next = desc[i].next; |
17cbca2b RR |
669 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
670 | wmb(); | |
671 | ||
d1f0132e | 672 | if (next >= max) |
1e1c17a7 | 673 | bad_driver(d, "Desc next is %u", next); |
17cbca2b RR |
674 | |
675 | return next; | |
676 | } | |
677 | ||
a91d74a3 RR |
678 | /* |
679 | * This actually sends the interrupt for this virtqueue, if we've used a | |
680 | * buffer. | |
681 | */ | |
38bc2b8c RR |
682 | static void trigger_irq(struct virtqueue *vq) |
683 | { | |
d9028eda | 684 | unsigned long buf[] = { LHREQ_IRQ, vq->dev->config.irq_line }; |
38bc2b8c | 685 | |
95c517c0 RR |
686 | /* Don't inform them if nothing used. */ |
687 | if (!vq->pending_used) | |
688 | return; | |
689 | vq->pending_used = 0; | |
690 | ||
d39a6785 RR |
691 | /* |
692 | * 2.4.7.1: | |
693 | * | |
694 | * If the VIRTIO_F_EVENT_IDX feature bit is not negotiated: | |
695 | * The driver MUST set flags to 0 or 1. | |
696 | */ | |
697 | if (vq->vring.avail->flags > 1) | |
1e1c17a7 | 698 | bad_driver_vq(vq, "avail->flags = %u\n", vq->vring.avail->flags); |
d39a6785 RR |
699 | |
700 | /* | |
701 | * 2.4.7.2: | |
702 | * | |
703 | * If the VIRTIO_F_EVENT_IDX feature bit is not negotiated: | |
704 | * | |
705 | * - The device MUST ignore the used_event value. | |
706 | * - After the device writes a descriptor index into the used ring: | |
707 | * - If flags is 1, the device SHOULD NOT send an interrupt. | |
708 | * - If flags is 0, the device MUST send an interrupt. | |
709 | */ | |
ca60a42c | 710 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) { |
990c91f0 | 711 | return; |
ca60a42c | 712 | } |
38bc2b8c | 713 | |
8dc425ff RR |
714 | /* |
715 | * 4.1.4.5.1: | |
716 | * | |
717 | * If MSI-X capability is disabled, the device MUST set the Queue | |
718 | * Interrupt bit in ISR status before sending a virtqueue notification | |
719 | * to the driver. | |
720 | */ | |
d9028eda | 721 | vq->dev->mmio->isr = 0x1; |
93153077 | 722 | |
38bc2b8c RR |
723 | /* Send the Guest an interrupt tell them we used something up. */ |
724 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
d9028eda | 725 | err(1, "Triggering irq %i", vq->dev->config.irq_line); |
38bc2b8c RR |
726 | } |
727 | ||
2e04ef76 | 728 | /* |
a91d74a3 | 729 | * This looks in the virtqueue for the first available buffer, and converts |
17cbca2b RR |
730 | * it to an iovec for convenient access. Since descriptors consist of some |
731 | * number of output then some number of input descriptors, it's actually two | |
732 | * iovecs, but we pack them into one and note how many of each there were. | |
733 | * | |
a91d74a3 | 734 | * This function waits if necessary, and returns the descriptor number found. |
2e04ef76 | 735 | */ |
659a0e66 RR |
736 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
737 | struct iovec iov[], | |
738 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 739 | { |
d1f0132e MM |
740 | unsigned int i, head, max; |
741 | struct vring_desc *desc; | |
659a0e66 RR |
742 | u16 last_avail = lg_last_avail(vq); |
743 | ||
d39a6785 RR |
744 | /* |
745 | * 2.4.7.1: | |
746 | * | |
747 | * The driver MUST handle spurious interrupts from the device. | |
748 | * | |
749 | * That's why this is a while loop. | |
750 | */ | |
751 | ||
a91d74a3 | 752 | /* There's nothing available? */ |
659a0e66 RR |
753 | while (last_avail == vq->vring.avail->idx) { |
754 | u64 event; | |
755 | ||
a91d74a3 RR |
756 | /* |
757 | * Since we're about to sleep, now is a good time to tell the | |
758 | * Guest about what we've used up to now. | |
759 | */ | |
38bc2b8c RR |
760 | trigger_irq(vq); |
761 | ||
b60da13f RR |
762 | /* OK, now we need to know about added descriptors. */ |
763 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
764 | ||
2e04ef76 RR |
765 | /* |
766 | * They could have slipped one in as we were doing that: make | |
767 | * sure it's written, then check again. | |
768 | */ | |
b60da13f RR |
769 | mb(); |
770 | if (last_avail != vq->vring.avail->idx) { | |
771 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
772 | break; | |
773 | } | |
774 | ||
659a0e66 RR |
775 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
776 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
777 | errx(1, "Event read failed?"); | |
b60da13f RR |
778 | |
779 | /* We don't need to be notified again. */ | |
780 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 781 | } |
17cbca2b RR |
782 | |
783 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 784 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
1e1c17a7 RR |
785 | bad_driver_vq(vq, "Guest moved used index from %u to %u", |
786 | last_avail, vq->vring.avail->idx); | |
17cbca2b | 787 | |
8fd9a636 RR |
788 | /* |
789 | * Make sure we read the descriptor number *after* we read the ring | |
790 | * update; don't let the cpu or compiler change the order. | |
791 | */ | |
792 | rmb(); | |
793 | ||
2e04ef76 RR |
794 | /* |
795 | * Grab the next descriptor number they're advertising, and increment | |
796 | * the index we've seen. | |
797 | */ | |
b5111790 RR |
798 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
799 | lg_last_avail(vq)++; | |
17cbca2b RR |
800 | |
801 | /* If their number is silly, that's a fatal mistake. */ | |
802 | if (head >= vq->vring.num) | |
1e1c17a7 | 803 | bad_driver_vq(vq, "Guest says index %u is available", head); |
17cbca2b RR |
804 | |
805 | /* When we start there are none of either input nor output. */ | |
806 | *out_num = *in_num = 0; | |
807 | ||
d1f0132e MM |
808 | max = vq->vring.num; |
809 | desc = vq->vring.desc; | |
17cbca2b | 810 | i = head; |
d1f0132e | 811 | |
8fd9a636 RR |
812 | /* |
813 | * We have to read the descriptor after we read the descriptor number, | |
814 | * but there's a data dependency there so the CPU shouldn't reorder | |
815 | * that: no rmb() required. | |
816 | */ | |
817 | ||
3afe3e0f RR |
818 | do { |
819 | /* | |
820 | * If this is an indirect entry, then this buffer contains a | |
821 | * descriptor table which we handle as if it's any normal | |
822 | * descriptor chain. | |
823 | */ | |
824 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { | |
d39a6785 RR |
825 | /* 2.4.5.3.1: |
826 | * | |
827 | * The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT | |
828 | * flag unless the VIRTIO_F_INDIRECT_DESC feature was | |
829 | * negotiated. | |
830 | */ | |
831 | if (!(vq->dev->features_accepted & | |
832 | (1<<VIRTIO_RING_F_INDIRECT_DESC))) | |
1e1c17a7 | 833 | bad_driver_vq(vq, "vq indirect not negotiated"); |
d39a6785 RR |
834 | |
835 | /* | |
836 | * 2.4.5.3.1: | |
837 | * | |
838 | * The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT | |
839 | * flag within an indirect descriptor (ie. only one | |
840 | * table per descriptor). | |
841 | */ | |
842 | if (desc != vq->vring.desc) | |
1e1c17a7 | 843 | bad_driver_vq(vq, "Indirect within indirect"); |
d39a6785 RR |
844 | |
845 | /* | |
846 | * Proposed update VIRTIO-134 spells this out: | |
847 | * | |
848 | * A driver MUST NOT set both VIRTQ_DESC_F_INDIRECT | |
849 | * and VIRTQ_DESC_F_NEXT in flags. | |
850 | */ | |
851 | if (desc[i].flags & VRING_DESC_F_NEXT) | |
1e1c17a7 | 852 | bad_driver_vq(vq, "indirect and next together"); |
d39a6785 | 853 | |
3afe3e0f | 854 | if (desc[i].len % sizeof(struct vring_desc)) |
1e1c17a7 RR |
855 | bad_driver_vq(vq, |
856 | "Invalid size for indirect table"); | |
d39a6785 RR |
857 | /* |
858 | * 2.4.5.3.2: | |
859 | * | |
860 | * The device MUST ignore the write-only flag | |
861 | * (flags&VIRTQ_DESC_F_WRITE) in the descriptor that | |
862 | * refers to an indirect table. | |
863 | * | |
864 | * We ignore it here: :) | |
865 | */ | |
d1f0132e | 866 | |
3afe3e0f | 867 | max = desc[i].len / sizeof(struct vring_desc); |
1e1c17a7 | 868 | desc = check_pointer(vq->dev, desc[i].addr, desc[i].len); |
3afe3e0f | 869 | i = 0; |
d39a6785 RR |
870 | |
871 | /* 2.4.5.3.1: | |
872 | * | |
873 | * A driver MUST NOT create a descriptor chain longer | |
874 | * than the Queue Size of the device. | |
875 | */ | |
876 | if (max > vq->pci_config.queue_size) | |
1e1c17a7 RR |
877 | bad_driver_vq(vq, |
878 | "indirect has too many entries"); | |
3afe3e0f | 879 | } |
d1f0132e | 880 | |
17cbca2b | 881 | /* Grab the first descriptor, and check it's OK. */ |
d1f0132e | 882 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 883 | iov[*out_num + *in_num].iov_base |
1e1c17a7 | 884 | = check_pointer(vq->dev, desc[i].addr, desc[i].len); |
17cbca2b | 885 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 886 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
887 | (*in_num)++; |
888 | else { | |
2e04ef76 RR |
889 | /* |
890 | * If it's an output descriptor, they're all supposed | |
891 | * to come before any input descriptors. | |
892 | */ | |
17cbca2b | 893 | if (*in_num) |
1e1c17a7 RR |
894 | bad_driver_vq(vq, |
895 | "Descriptor has out after in"); | |
17cbca2b RR |
896 | (*out_num)++; |
897 | } | |
898 | ||
899 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 900 | if (*out_num + *in_num > max) |
1e1c17a7 RR |
901 | bad_driver_vq(vq, "Looped descriptor"); |
902 | } while ((i = next_desc(vq->dev, desc, i, max)) != max); | |
dde79789 | 903 | |
17cbca2b | 904 | return head; |
8ca47e00 RR |
905 | } |
906 | ||
2e04ef76 | 907 | /* |
a91d74a3 RR |
908 | * After we've used one of their buffers, we tell the Guest about it. Sometime |
909 | * later we'll want to send them an interrupt using trigger_irq(); note that | |
910 | * wait_for_vq_desc() does that for us if it has to wait. | |
2e04ef76 | 911 | */ |
17cbca2b | 912 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
8ca47e00 | 913 | { |
17cbca2b RR |
914 | struct vring_used_elem *used; |
915 | ||
2e04ef76 RR |
916 | /* |
917 | * The virtqueue contains a ring of used buffers. Get a pointer to the | |
918 | * next entry in that used ring. | |
919 | */ | |
17cbca2b RR |
920 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
921 | used->id = head; | |
922 | used->len = len; | |
923 | /* Make sure buffer is written before we update index. */ | |
924 | wmb(); | |
925 | vq->vring.used->idx++; | |
95c517c0 | 926 | vq->pending_used++; |
8ca47e00 RR |
927 | } |
928 | ||
17cbca2b | 929 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 930 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 931 | { |
17cbca2b | 932 | add_used(vq, head, len); |
56739c80 | 933 | trigger_irq(vq); |
8ca47e00 RR |
934 | } |
935 | ||
e1e72965 RR |
936 | /* |
937 | * The Console | |
938 | * | |
2e04ef76 RR |
939 | * We associate some data with the console for our exit hack. |
940 | */ | |
1842f23c | 941 | struct console_abort { |
dde79789 | 942 | /* How many times have they hit ^C? */ |
8ca47e00 | 943 | int count; |
dde79789 | 944 | /* When did they start? */ |
8ca47e00 RR |
945 | struct timeval start; |
946 | }; | |
947 | ||
dde79789 | 948 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 949 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 950 | { |
8ca47e00 | 951 | int len; |
17cbca2b | 952 | unsigned int head, in_num, out_num; |
659a0e66 RR |
953 | struct console_abort *abort = vq->dev->priv; |
954 | struct iovec iov[vq->vring.num]; | |
56ae43df | 955 | |
a91d74a3 | 956 | /* Make sure there's a descriptor available. */ |
659a0e66 | 957 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
56ae43df | 958 | if (out_num) |
1e1c17a7 | 959 | bad_driver_vq(vq, "Output buffers in console in queue?"); |
8ca47e00 | 960 | |
a91d74a3 | 961 | /* Read into it. This is where we usually wait. */ |
659a0e66 | 962 | len = readv(STDIN_FILENO, iov, in_num); |
8ca47e00 | 963 | if (len <= 0) { |
659a0e66 | 964 | /* Ran out of input? */ |
8ca47e00 | 965 | warnx("Failed to get console input, ignoring console."); |
2e04ef76 RR |
966 | /* |
967 | * For simplicity, dying threads kill the whole Launcher. So | |
968 | * just nap here. | |
969 | */ | |
659a0e66 RR |
970 | for (;;) |
971 | pause(); | |
8ca47e00 RR |
972 | } |
973 | ||
a91d74a3 | 974 | /* Tell the Guest we used a buffer. */ |
659a0e66 | 975 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 976 | |
2e04ef76 RR |
977 | /* |
978 | * Three ^C within one second? Exit. | |
dde79789 | 979 | * |
659a0e66 RR |
980 | * This is such a hack, but works surprisingly well. Each ^C has to |
981 | * be in a buffer by itself, so they can't be too fast. But we check | |
982 | * that we get three within about a second, so they can't be too | |
2e04ef76 RR |
983 | * slow. |
984 | */ | |
659a0e66 | 985 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
8ca47e00 | 986 | abort->count = 0; |
659a0e66 RR |
987 | return; |
988 | } | |
8ca47e00 | 989 | |
659a0e66 RR |
990 | abort->count++; |
991 | if (abort->count == 1) | |
992 | gettimeofday(&abort->start, NULL); | |
993 | else if (abort->count == 3) { | |
994 | struct timeval now; | |
995 | gettimeofday(&now, NULL); | |
996 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
997 | if (now.tv_sec <= abort->start.tv_sec+1) | |
998 | kill(0, SIGINT); | |
999 | abort->count = 0; | |
1000 | } | |
8ca47e00 RR |
1001 | } |
1002 | ||
659a0e66 RR |
1003 | /* This is the routine which handles console output (ie. stdout). */ |
1004 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 1005 | { |
17cbca2b | 1006 | unsigned int head, out, in; |
17cbca2b RR |
1007 | struct iovec iov[vq->vring.num]; |
1008 | ||
a91d74a3 | 1009 | /* We usually wait in here, for the Guest to give us something. */ |
659a0e66 RR |
1010 | head = wait_for_vq_desc(vq, iov, &out, &in); |
1011 | if (in) | |
1e1c17a7 | 1012 | bad_driver_vq(vq, "Input buffers in console output queue?"); |
a91d74a3 RR |
1013 | |
1014 | /* writev can return a partial write, so we loop here. */ | |
659a0e66 RR |
1015 | while (!iov_empty(iov, out)) { |
1016 | int len = writev(STDOUT_FILENO, iov, out); | |
e0377e25 SA |
1017 | if (len <= 0) { |
1018 | warn("Write to stdout gave %i (%d)", len, errno); | |
1019 | break; | |
1020 | } | |
1e1c17a7 | 1021 | iov_consume(vq->dev, iov, out, NULL, len); |
17cbca2b | 1022 | } |
a91d74a3 RR |
1023 | |
1024 | /* | |
1025 | * We're finished with that buffer: if we're going to sleep, | |
1026 | * wait_for_vq_desc() will prod the Guest with an interrupt. | |
1027 | */ | |
38bc2b8c | 1028 | add_used(vq, head, 0); |
a161883a RR |
1029 | } |
1030 | ||
e1e72965 RR |
1031 | /* |
1032 | * The Network | |
1033 | * | |
1034 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 1035 | * and write them to /dev/net/tun. |
a6bd8e13 | 1036 | */ |
659a0e66 RR |
1037 | struct net_info { |
1038 | int tunfd; | |
1039 | }; | |
1040 | ||
1041 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 1042 | { |
659a0e66 RR |
1043 | struct net_info *net_info = vq->dev->priv; |
1044 | unsigned int head, out, in; | |
17cbca2b | 1045 | struct iovec iov[vq->vring.num]; |
a161883a | 1046 | |
a91d74a3 | 1047 | /* We usually wait in here for the Guest to give us a packet. */ |
659a0e66 RR |
1048 | head = wait_for_vq_desc(vq, iov, &out, &in); |
1049 | if (in) | |
1e1c17a7 | 1050 | bad_driver_vq(vq, "Input buffers in net output queue?"); |
a91d74a3 RR |
1051 | /* |
1052 | * Send the whole thing through to /dev/net/tun. It expects the exact | |
1053 | * same format: what a coincidence! | |
1054 | */ | |
659a0e66 | 1055 | if (writev(net_info->tunfd, iov, out) < 0) |
e0377e25 | 1056 | warnx("Write to tun failed (%d)?", errno); |
a91d74a3 RR |
1057 | |
1058 | /* | |
1059 | * Done with that one; wait_for_vq_desc() will send the interrupt if | |
1060 | * all packets are processed. | |
1061 | */ | |
38bc2b8c | 1062 | add_used(vq, head, 0); |
8ca47e00 RR |
1063 | } |
1064 | ||
a91d74a3 RR |
1065 | /* |
1066 | * Handling network input is a bit trickier, because I've tried to optimize it. | |
1067 | * | |
1068 | * First we have a helper routine which tells is if from this file descriptor | |
1069 | * (ie. the /dev/net/tun device) will block: | |
1070 | */ | |
4a8962e2 RR |
1071 | static bool will_block(int fd) |
1072 | { | |
1073 | fd_set fdset; | |
1074 | struct timeval zero = { 0, 0 }; | |
1075 | FD_ZERO(&fdset); | |
1076 | FD_SET(fd, &fdset); | |
1077 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
1078 | } | |
1079 | ||
a91d74a3 RR |
1080 | /* |
1081 | * This handles packets coming in from the tun device to our Guest. Like all | |
1082 | * service routines, it gets called again as soon as it returns, so you don't | |
1083 | * see a while(1) loop here. | |
1084 | */ | |
659a0e66 | 1085 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 1086 | { |
8ca47e00 | 1087 | int len; |
659a0e66 RR |
1088 | unsigned int head, out, in; |
1089 | struct iovec iov[vq->vring.num]; | |
1090 | struct net_info *net_info = vq->dev->priv; | |
1091 | ||
a91d74a3 RR |
1092 | /* |
1093 | * Get a descriptor to write an incoming packet into. This will also | |
1094 | * send an interrupt if they're out of descriptors. | |
1095 | */ | |
659a0e66 RR |
1096 | head = wait_for_vq_desc(vq, iov, &out, &in); |
1097 | if (out) | |
1e1c17a7 | 1098 | bad_driver_vq(vq, "Output buffers in net input queue?"); |
4a8962e2 | 1099 | |
a91d74a3 RR |
1100 | /* |
1101 | * If it looks like we'll block reading from the tun device, send them | |
1102 | * an interrupt. | |
1103 | */ | |
4a8962e2 RR |
1104 | if (vq->pending_used && will_block(net_info->tunfd)) |
1105 | trigger_irq(vq); | |
1106 | ||
a91d74a3 RR |
1107 | /* |
1108 | * Read in the packet. This is where we normally wait (when there's no | |
1109 | * incoming network traffic). | |
1110 | */ | |
659a0e66 | 1111 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 1112 | if (len <= 0) |
e0377e25 | 1113 | warn("Failed to read from tun (%d).", errno); |
a91d74a3 RR |
1114 | |
1115 | /* | |
1116 | * Mark that packet buffer as used, but don't interrupt here. We want | |
1117 | * to wait until we've done as much work as we can. | |
1118 | */ | |
4a8962e2 | 1119 | add_used(vq, head, len); |
659a0e66 | 1120 | } |
a91d74a3 | 1121 | /*:*/ |
dde79789 | 1122 | |
a91d74a3 | 1123 | /* This is the helper to create threads: run the service routine in a loop. */ |
659a0e66 RR |
1124 | static int do_thread(void *_vq) |
1125 | { | |
1126 | struct virtqueue *vq = _vq; | |
17cbca2b | 1127 | |
659a0e66 RR |
1128 | for (;;) |
1129 | vq->service(vq); | |
1130 | return 0; | |
1131 | } | |
17cbca2b | 1132 | |
2e04ef76 RR |
1133 | /* |
1134 | * When a child dies, we kill our entire process group with SIGTERM. This | |
1135 | * also has the side effect that the shell restores the console for us! | |
1136 | */ | |
659a0e66 RR |
1137 | static void kill_launcher(int signal) |
1138 | { | |
1139 | kill(0, SIGTERM); | |
8ca47e00 RR |
1140 | } |
1141 | ||
d2dbdac3 RR |
1142 | static void reset_vq_pci_config(struct virtqueue *vq) |
1143 | { | |
1144 | vq->pci_config.queue_size = VIRTQUEUE_NUM; | |
1145 | vq->pci_config.queue_enable = 0; | |
1146 | } | |
1147 | ||
659a0e66 | 1148 | static void reset_device(struct device *dev) |
56ae43df | 1149 | { |
659a0e66 RR |
1150 | struct virtqueue *vq; |
1151 | ||
1152 | verbose("Resetting device %s\n", dev->name); | |
1153 | ||
1154 | /* Clear any features they've acked. */ | |
d9028eda | 1155 | dev->features_accepted = 0; |
659a0e66 RR |
1156 | |
1157 | /* We're going to be explicitly killing threads, so ignore them. */ | |
1158 | signal(SIGCHLD, SIG_IGN); | |
1159 | ||
d2dbdac3 RR |
1160 | /* |
1161 | * 4.1.4.3.1: | |
1162 | * | |
1163 | * The device MUST present a 0 in queue_enable on reset. | |
1164 | * | |
1165 | * This means we set it here, and reset the saved ones in every vq. | |
1166 | */ | |
1167 | dev->mmio->cfg.queue_enable = 0; | |
1168 | ||
d9028eda | 1169 | /* Get rid of the virtqueue threads */ |
659a0e66 | 1170 | for (vq = dev->vq; vq; vq = vq->next) { |
d2dbdac3 RR |
1171 | vq->last_avail_idx = 0; |
1172 | reset_vq_pci_config(vq); | |
659a0e66 RR |
1173 | if (vq->thread != (pid_t)-1) { |
1174 | kill(vq->thread, SIGTERM); | |
1175 | waitpid(vq->thread, NULL, 0); | |
1176 | vq->thread = (pid_t)-1; | |
1177 | } | |
659a0e66 RR |
1178 | } |
1179 | dev->running = false; | |
d39a6785 | 1180 | dev->wrote_features_ok = false; |
659a0e66 RR |
1181 | |
1182 | /* Now we care if threads die. */ | |
1183 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
1184 | } |
1185 | ||
d9028eda | 1186 | static void cleanup_devices(void) |
6e5aa7ef | 1187 | { |
659a0e66 | 1188 | unsigned int i; |
659a0e66 | 1189 | |
d9028eda RR |
1190 | for (i = 1; i < MAX_PCI_DEVICES; i++) { |
1191 | struct device *d = devices.pci[i]; | |
1192 | if (!d) | |
1193 | continue; | |
1194 | reset_device(d); | |
659a0e66 | 1195 | } |
6e5aa7ef | 1196 | |
659a0e66 RR |
1197 | /* If we saved off the original terminal settings, restore them now. */ |
1198 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
1199 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
1200 | } | |
6e5aa7ef | 1201 | |
d7fbf6e9 RR |
1202 | /*L:217 |
1203 | * We do PCI. This is mainly done to let us test the kernel virtio PCI | |
1204 | * code. | |
1205 | */ | |
1206 | ||
8e709469 RR |
1207 | /* Linux expects a PCI host bridge: ours is a dummy, and first on the bus. */ |
1208 | static struct device pci_host_bridge; | |
1209 | ||
1210 | static void init_pci_host_bridge(void) | |
1211 | { | |
1212 | pci_host_bridge.name = "PCI Host Bridge"; | |
1213 | pci_host_bridge.config.class = 0x06; /* bridge */ | |
1214 | pci_host_bridge.config.subclass = 0; /* host bridge */ | |
1215 | devices.pci[0] = &pci_host_bridge; | |
1216 | } | |
1217 | ||
d7fbf6e9 RR |
1218 | /* The IO ports used to read the PCI config space. */ |
1219 | #define PCI_CONFIG_ADDR 0xCF8 | |
1220 | #define PCI_CONFIG_DATA 0xCFC | |
1221 | ||
1222 | /* | |
1223 | * Not really portable, but does help readability: this is what the Guest | |
1224 | * writes to the PCI_CONFIG_ADDR IO port. | |
1225 | */ | |
1226 | union pci_config_addr { | |
1227 | struct { | |
1228 | unsigned mbz: 2; | |
1229 | unsigned offset: 6; | |
1230 | unsigned funcnum: 3; | |
1231 | unsigned devnum: 5; | |
1232 | unsigned busnum: 8; | |
1233 | unsigned reserved: 7; | |
1234 | unsigned enabled : 1; | |
1235 | } bits; | |
1236 | u32 val; | |
1237 | }; | |
1238 | ||
1239 | /* | |
1240 | * We cache what they wrote to the address port, so we know what they're | |
1241 | * talking about when they access the data port. | |
1242 | */ | |
1243 | static union pci_config_addr pci_config_addr; | |
1244 | ||
1245 | static struct device *find_pci_device(unsigned int index) | |
1246 | { | |
1247 | return devices.pci[index]; | |
1248 | } | |
1249 | ||
1250 | /* PCI can do 1, 2 and 4 byte reads; we handle that here. */ | |
1251 | static void ioread(u16 off, u32 v, u32 mask, u32 *val) | |
1252 | { | |
1253 | assert(off < 4); | |
1254 | assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF); | |
1255 | *val = (v >> (off * 8)) & mask; | |
1256 | } | |
1257 | ||
1258 | /* PCI can do 1, 2 and 4 byte writes; we handle that here. */ | |
1259 | static void iowrite(u16 off, u32 v, u32 mask, u32 *dst) | |
1260 | { | |
1261 | assert(off < 4); | |
1262 | assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF); | |
1263 | *dst &= ~(mask << (off * 8)); | |
1264 | *dst |= (v & mask) << (off * 8); | |
1265 | } | |
1266 | ||
1267 | /* | |
1268 | * Where PCI_CONFIG_DATA accesses depends on the previous write to | |
1269 | * PCI_CONFIG_ADDR. | |
1270 | */ | |
1271 | static struct device *dev_and_reg(u32 *reg) | |
1272 | { | |
1273 | if (!pci_config_addr.bits.enabled) | |
1274 | return NULL; | |
1275 | ||
1276 | if (pci_config_addr.bits.funcnum != 0) | |
1277 | return NULL; | |
1278 | ||
1279 | if (pci_config_addr.bits.busnum != 0) | |
1280 | return NULL; | |
1281 | ||
1282 | if (pci_config_addr.bits.offset * 4 >= sizeof(struct pci_config)) | |
1283 | return NULL; | |
1284 | ||
1285 | *reg = pci_config_addr.bits.offset; | |
1286 | return find_pci_device(pci_config_addr.bits.devnum); | |
1287 | } | |
1288 | ||
59eba788 RR |
1289 | /* |
1290 | * We can get invalid combinations of values while they're writing, so we | |
1291 | * only fault if they try to write with some invalid bar/offset/length. | |
1292 | */ | |
1293 | static bool valid_bar_access(struct device *d, | |
1294 | struct virtio_pci_cfg_cap *cfg_access) | |
1295 | { | |
1296 | /* We only have 1 bar (BAR0) */ | |
1297 | if (cfg_access->cap.bar != 0) | |
1298 | return false; | |
1299 | ||
1300 | /* Check it's within BAR0. */ | |
1301 | if (cfg_access->cap.offset >= d->mmio_size | |
1302 | || cfg_access->cap.offset + cfg_access->cap.length > d->mmio_size) | |
1303 | return false; | |
1304 | ||
1305 | /* Check length is 1, 2 or 4. */ | |
1306 | if (cfg_access->cap.length != 1 | |
1307 | && cfg_access->cap.length != 2 | |
1308 | && cfg_access->cap.length != 4) | |
1309 | return false; | |
1310 | ||
c97eb679 RR |
1311 | /* |
1312 | * 4.1.4.7.2: | |
1313 | * | |
1314 | * The driver MUST NOT write a cap.offset which is not a multiple of | |
1315 | * cap.length (ie. all accesses MUST be aligned). | |
1316 | */ | |
59eba788 RR |
1317 | if (cfg_access->cap.offset % cfg_access->cap.length != 0) |
1318 | return false; | |
1319 | ||
1320 | /* Return pointer into word in BAR0. */ | |
1321 | return true; | |
1322 | } | |
1323 | ||
d7fbf6e9 RR |
1324 | /* Is this accessing the PCI config address port?. */ |
1325 | static bool is_pci_addr_port(u16 port) | |
1326 | { | |
1327 | return port >= PCI_CONFIG_ADDR && port < PCI_CONFIG_ADDR + 4; | |
1328 | } | |
1329 | ||
1330 | static bool pci_addr_iowrite(u16 port, u32 mask, u32 val) | |
1331 | { | |
1332 | iowrite(port - PCI_CONFIG_ADDR, val, mask, | |
1333 | &pci_config_addr.val); | |
1334 | verbose("PCI%s: %#x/%x: bus %u dev %u func %u reg %u\n", | |
1335 | pci_config_addr.bits.enabled ? "" : " DISABLED", | |
1336 | val, mask, | |
1337 | pci_config_addr.bits.busnum, | |
1338 | pci_config_addr.bits.devnum, | |
1339 | pci_config_addr.bits.funcnum, | |
1340 | pci_config_addr.bits.offset); | |
1341 | return true; | |
1342 | } | |
1343 | ||
1344 | static void pci_addr_ioread(u16 port, u32 mask, u32 *val) | |
1345 | { | |
1346 | ioread(port - PCI_CONFIG_ADDR, pci_config_addr.val, mask, val); | |
1347 | } | |
1348 | ||
1349 | /* Is this accessing the PCI config data port?. */ | |
1350 | static bool is_pci_data_port(u16 port) | |
1351 | { | |
1352 | return port >= PCI_CONFIG_DATA && port < PCI_CONFIG_DATA + 4; | |
1353 | } | |
1354 | ||
59eba788 RR |
1355 | static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask); |
1356 | ||
d7fbf6e9 RR |
1357 | static bool pci_data_iowrite(u16 port, u32 mask, u32 val) |
1358 | { | |
1359 | u32 reg, portoff; | |
1360 | struct device *d = dev_and_reg(®); | |
1361 | ||
1362 | /* Complain if they don't belong to a device. */ | |
1363 | if (!d) | |
1364 | return false; | |
1365 | ||
1366 | /* They can do 1 byte writes, etc. */ | |
1367 | portoff = port - PCI_CONFIG_DATA; | |
1368 | ||
1369 | /* | |
1370 | * PCI uses a weird way to determine the BAR size: the OS | |
1371 | * writes all 1's, and sees which ones stick. | |
1372 | */ | |
1373 | if (&d->config_words[reg] == &d->config.bar[0]) { | |
1374 | int i; | |
1375 | ||
1376 | iowrite(portoff, val, mask, &d->config.bar[0]); | |
1377 | for (i = 0; (1 << i) < d->mmio_size; i++) | |
1378 | d->config.bar[0] &= ~(1 << i); | |
1379 | return true; | |
1380 | } else if ((&d->config_words[reg] > &d->config.bar[0] | |
1381 | && &d->config_words[reg] <= &d->config.bar[6]) | |
1382 | || &d->config_words[reg] == &d->config.expansion_rom_addr) { | |
1383 | /* Allow writing to any other BAR, or expansion ROM */ | |
1384 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1385 | return true; | |
1386 | /* We let them overide latency timer and cacheline size */ | |
1387 | } else if (&d->config_words[reg] == (void *)&d->config.cacheline_size) { | |
1388 | /* Only let them change the first two fields. */ | |
1389 | if (mask == 0xFFFFFFFF) | |
1390 | mask = 0xFFFF; | |
1391 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1392 | return true; | |
1393 | } else if (&d->config_words[reg] == (void *)&d->config.command | |
1394 | && mask == 0xFFFF) { | |
1395 | /* Ignore command writes. */ | |
1396 | return true; | |
59eba788 RR |
1397 | } else if (&d->config_words[reg] |
1398 | == (void *)&d->config.cfg_access.cap.bar | |
1399 | || &d->config_words[reg] | |
1400 | == &d->config.cfg_access.cap.length | |
1401 | || &d->config_words[reg] | |
1402 | == &d->config.cfg_access.cap.offset) { | |
1403 | ||
1404 | /* | |
1405 | * The VIRTIO_PCI_CAP_PCI_CFG capability | |
1406 | * provides a backdoor to access the MMIO | |
1407 | * regions without mapping them. Weird, but | |
1408 | * useful. | |
1409 | */ | |
1410 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1411 | return true; | |
b2ce1ea4 | 1412 | } else if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) { |
59eba788 RR |
1413 | u32 write_mask; |
1414 | ||
8dc425ff RR |
1415 | /* |
1416 | * 4.1.4.7.1: | |
1417 | * | |
1418 | * Upon detecting driver write access to pci_cfg_data, the | |
1419 | * device MUST execute a write access at offset cap.offset at | |
1420 | * BAR selected by cap.bar using the first cap.length bytes | |
1421 | * from pci_cfg_data. | |
1422 | */ | |
1423 | ||
59eba788 RR |
1424 | /* Must be bar 0 */ |
1425 | if (!valid_bar_access(d, &d->config.cfg_access)) | |
1426 | return false; | |
1427 | ||
b2ce1ea4 | 1428 | iowrite(portoff, val, mask, &d->config.cfg_access.pci_cfg_data); |
59eba788 RR |
1429 | |
1430 | /* | |
1431 | * Now emulate a write. The mask we use is set by | |
1432 | * len, *not* this write! | |
1433 | */ | |
1434 | write_mask = (1ULL<<(8*d->config.cfg_access.cap.length)) - 1; | |
1435 | verbose("Window writing %#x/%#x to bar %u, offset %u len %u\n", | |
b2ce1ea4 | 1436 | d->config.cfg_access.pci_cfg_data, write_mask, |
59eba788 RR |
1437 | d->config.cfg_access.cap.bar, |
1438 | d->config.cfg_access.cap.offset, | |
1439 | d->config.cfg_access.cap.length); | |
1440 | ||
1441 | emulate_mmio_write(d, d->config.cfg_access.cap.offset, | |
b2ce1ea4 RR |
1442 | d->config.cfg_access.pci_cfg_data, |
1443 | write_mask); | |
59eba788 | 1444 | return true; |
d7fbf6e9 RR |
1445 | } |
1446 | ||
c97eb679 RR |
1447 | /* |
1448 | * 4.1.4.1: | |
1449 | * | |
1450 | * The driver MUST NOT write into any field of the capability | |
1451 | * structure, with the exception of those with cap_type | |
1452 | * VIRTIO_PCI_CAP_PCI_CFG... | |
1453 | */ | |
d7fbf6e9 RR |
1454 | return false; |
1455 | } | |
1456 | ||
59eba788 RR |
1457 | static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask); |
1458 | ||
d7fbf6e9 RR |
1459 | static void pci_data_ioread(u16 port, u32 mask, u32 *val) |
1460 | { | |
1461 | u32 reg; | |
1462 | struct device *d = dev_and_reg(®); | |
1463 | ||
1464 | if (!d) | |
1465 | return; | |
59eba788 RR |
1466 | |
1467 | /* Read through the PCI MMIO access window is special */ | |
b2ce1ea4 | 1468 | if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) { |
59eba788 RR |
1469 | u32 read_mask; |
1470 | ||
8dc425ff RR |
1471 | /* |
1472 | * 4.1.4.7.1: | |
1473 | * | |
1474 | * Upon detecting driver read access to pci_cfg_data, the | |
1475 | * device MUST execute a read access of length cap.length at | |
1476 | * offset cap.offset at BAR selected by cap.bar and store the | |
1477 | * first cap.length bytes in pci_cfg_data. | |
1478 | */ | |
59eba788 RR |
1479 | /* Must be bar 0 */ |
1480 | if (!valid_bar_access(d, &d->config.cfg_access)) | |
1e1c17a7 RR |
1481 | bad_driver(d, |
1482 | "Invalid cfg_access to bar%u, offset %u len %u", | |
59eba788 RR |
1483 | d->config.cfg_access.cap.bar, |
1484 | d->config.cfg_access.cap.offset, | |
1485 | d->config.cfg_access.cap.length); | |
1486 | ||
1487 | /* | |
1488 | * Read into the window. The mask we use is set by | |
1489 | * len, *not* this read! | |
1490 | */ | |
1491 | read_mask = (1ULL<<(8*d->config.cfg_access.cap.length))-1; | |
b2ce1ea4 | 1492 | d->config.cfg_access.pci_cfg_data |
59eba788 RR |
1493 | = emulate_mmio_read(d, |
1494 | d->config.cfg_access.cap.offset, | |
1495 | read_mask); | |
1496 | verbose("Window read %#x/%#x from bar %u, offset %u len %u\n", | |
b2ce1ea4 | 1497 | d->config.cfg_access.pci_cfg_data, read_mask, |
59eba788 RR |
1498 | d->config.cfg_access.cap.bar, |
1499 | d->config.cfg_access.cap.offset, | |
1500 | d->config.cfg_access.cap.length); | |
1501 | } | |
d7fbf6e9 RR |
1502 | ioread(port - PCI_CONFIG_DATA, d->config_words[reg], mask, val); |
1503 | } | |
1504 | ||
c565650b RR |
1505 | /*L:216 |
1506 | * This is where we emulate a handful of Guest instructions. It's ugly | |
1507 | * and we used to do it in the kernel but it grew over time. | |
1508 | */ | |
1509 | ||
1510 | /* | |
1511 | * We use the ptrace syscall's pt_regs struct to talk about registers | |
1512 | * to lguest: these macros convert the names to the offsets. | |
1513 | */ | |
1514 | #define getreg(name) getreg_off(offsetof(struct user_regs_struct, name)) | |
1515 | #define setreg(name, val) \ | |
1516 | setreg_off(offsetof(struct user_regs_struct, name), (val)) | |
1517 | ||
1518 | static u32 getreg_off(size_t offset) | |
1519 | { | |
1520 | u32 r; | |
1521 | unsigned long args[] = { LHREQ_GETREG, offset }; | |
1522 | ||
1523 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) | |
1524 | err(1, "Getting register %u", offset); | |
1525 | if (pread(lguest_fd, &r, sizeof(r), cpu_id) != sizeof(r)) | |
1526 | err(1, "Reading register %u", offset); | |
1527 | ||
1528 | return r; | |
1529 | } | |
1530 | ||
1531 | static void setreg_off(size_t offset, u32 val) | |
1532 | { | |
1533 | unsigned long args[] = { LHREQ_SETREG, offset, val }; | |
1534 | ||
1535 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) | |
1536 | err(1, "Setting register %u", offset); | |
1537 | } | |
1538 | ||
6a54f9ab RR |
1539 | /* Get register by instruction encoding */ |
1540 | static u32 getreg_num(unsigned regnum, u32 mask) | |
1541 | { | |
1542 | /* 8 bit ops use regnums 4-7 for high parts of word */ | |
1543 | if (mask == 0xFF && (regnum & 0x4)) | |
1544 | return getreg_num(regnum & 0x3, 0xFFFF) >> 8; | |
1545 | ||
1546 | switch (regnum) { | |
1547 | case 0: return getreg(eax) & mask; | |
1548 | case 1: return getreg(ecx) & mask; | |
1549 | case 2: return getreg(edx) & mask; | |
1550 | case 3: return getreg(ebx) & mask; | |
1551 | case 4: return getreg(esp) & mask; | |
1552 | case 5: return getreg(ebp) & mask; | |
1553 | case 6: return getreg(esi) & mask; | |
1554 | case 7: return getreg(edi) & mask; | |
1555 | } | |
1556 | abort(); | |
1557 | } | |
1558 | ||
1559 | /* Set register by instruction encoding */ | |
1560 | static void setreg_num(unsigned regnum, u32 val, u32 mask) | |
1561 | { | |
1562 | /* Don't try to set bits out of range */ | |
1563 | assert(~(val & ~mask)); | |
1564 | ||
1565 | /* 8 bit ops use regnums 4-7 for high parts of word */ | |
1566 | if (mask == 0xFF && (regnum & 0x4)) { | |
1567 | /* Construct the 16 bits we want. */ | |
1568 | val = (val << 8) | getreg_num(regnum & 0x3, 0xFF); | |
1569 | setreg_num(regnum & 0x3, val, 0xFFFF); | |
1570 | return; | |
1571 | } | |
1572 | ||
1573 | switch (regnum) { | |
1574 | case 0: setreg(eax, val | (getreg(eax) & ~mask)); return; | |
1575 | case 1: setreg(ecx, val | (getreg(ecx) & ~mask)); return; | |
1576 | case 2: setreg(edx, val | (getreg(edx) & ~mask)); return; | |
1577 | case 3: setreg(ebx, val | (getreg(ebx) & ~mask)); return; | |
1578 | case 4: setreg(esp, val | (getreg(esp) & ~mask)); return; | |
1579 | case 5: setreg(ebp, val | (getreg(ebp) & ~mask)); return; | |
1580 | case 6: setreg(esi, val | (getreg(esi) & ~mask)); return; | |
1581 | case 7: setreg(edi, val | (getreg(edi) & ~mask)); return; | |
1582 | } | |
1583 | abort(); | |
1584 | } | |
1585 | ||
1586 | /* Get bytes of displacement appended to instruction, from r/m encoding */ | |
1587 | static u32 insn_displacement_len(u8 mod_reg_rm) | |
1588 | { | |
1589 | /* Switch on the mod bits */ | |
1590 | switch (mod_reg_rm >> 6) { | |
1591 | case 0: | |
1592 | /* If mod == 0, and r/m == 101, 16-bit displacement follows */ | |
1593 | if ((mod_reg_rm & 0x7) == 0x5) | |
1594 | return 2; | |
1595 | /* Normally, mod == 0 means no literal displacement */ | |
1596 | return 0; | |
1597 | case 1: | |
1598 | /* One byte displacement */ | |
1599 | return 1; | |
1600 | case 2: | |
1601 | /* Four byte displacement */ | |
1602 | return 4; | |
1603 | case 3: | |
1604 | /* Register mode */ | |
1605 | return 0; | |
1606 | } | |
1607 | abort(); | |
1608 | } | |
1609 | ||
c565650b RR |
1610 | static void emulate_insn(const u8 insn[]) |
1611 | { | |
1612 | unsigned long args[] = { LHREQ_TRAP, 13 }; | |
1613 | unsigned int insnlen = 0, in = 0, small_operand = 0, byte_access; | |
1614 | unsigned int eax, port, mask; | |
1615 | /* | |
d7fbf6e9 | 1616 | * Default is to return all-ones on IO port reads, which traditionally |
c565650b RR |
1617 | * means "there's nothing there". |
1618 | */ | |
1619 | u32 val = 0xFFFFFFFF; | |
1620 | ||
1621 | /* | |
1622 | * This must be the Guest kernel trying to do something, not userspace! | |
1623 | * The bottom two bits of the CS segment register are the privilege | |
1624 | * level. | |
1625 | */ | |
1626 | if ((getreg(xcs) & 3) != 0x1) | |
1627 | goto no_emulate; | |
1628 | ||
1629 | /* Decoding x86 instructions is icky. */ | |
1630 | ||
1631 | /* | |
1632 | * Around 2.6.33, the kernel started using an emulation for the | |
1633 | * cmpxchg8b instruction in early boot on many configurations. This | |
1634 | * code isn't paravirtualized, and it tries to disable interrupts. | |
1635 | * Ignore it, which will Mostly Work. | |
1636 | */ | |
1637 | if (insn[insnlen] == 0xfa) { | |
1638 | /* "cli", or Clear Interrupt Enable instruction. Skip it. */ | |
1639 | insnlen = 1; | |
1640 | goto skip_insn; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out. | |
1645 | */ | |
1646 | if (insn[insnlen] == 0x66) { | |
1647 | small_operand = 1; | |
1648 | /* The instruction is 1 byte so far, read the next byte. */ | |
1649 | insnlen = 1; | |
1650 | } | |
1651 | ||
1652 | /* If the lower bit isn't set, it's a single byte access */ | |
1653 | byte_access = !(insn[insnlen] & 1); | |
1654 | ||
1655 | /* | |
1656 | * Now we can ignore the lower bit and decode the 4 opcodes | |
1657 | * we need to emulate. | |
1658 | */ | |
1659 | switch (insn[insnlen] & 0xFE) { | |
1660 | case 0xE4: /* in <next byte>,%al */ | |
1661 | port = insn[insnlen+1]; | |
1662 | insnlen += 2; | |
1663 | in = 1; | |
1664 | break; | |
1665 | case 0xEC: /* in (%dx),%al */ | |
1666 | port = getreg(edx) & 0xFFFF; | |
1667 | insnlen += 1; | |
1668 | in = 1; | |
1669 | break; | |
1670 | case 0xE6: /* out %al,<next byte> */ | |
1671 | port = insn[insnlen+1]; | |
1672 | insnlen += 2; | |
1673 | break; | |
1674 | case 0xEE: /* out %al,(%dx) */ | |
1675 | port = getreg(edx) & 0xFFFF; | |
1676 | insnlen += 1; | |
1677 | break; | |
1678 | default: | |
1679 | /* OK, we don't know what this is, can't emulate. */ | |
1680 | goto no_emulate; | |
1681 | } | |
1682 | ||
1683 | /* Set a mask of the 1, 2 or 4 bytes, depending on size of IO */ | |
1684 | if (byte_access) | |
1685 | mask = 0xFF; | |
1686 | else if (small_operand) | |
1687 | mask = 0xFFFF; | |
1688 | else | |
1689 | mask = 0xFFFFFFFF; | |
1690 | ||
1691 | /* | |
1692 | * If it was an "IN" instruction, they expect the result to be read | |
1693 | * into %eax, so we change %eax. | |
1694 | */ | |
1695 | eax = getreg(eax); | |
1696 | ||
1697 | if (in) { | |
d7fbf6e9 RR |
1698 | /* This is the PS/2 keyboard status; 1 means ready for output */ |
1699 | if (port == 0x64) | |
1700 | val = 1; | |
1701 | else if (is_pci_addr_port(port)) | |
1702 | pci_addr_ioread(port, mask, &val); | |
1703 | else if (is_pci_data_port(port)) | |
1704 | pci_data_ioread(port, mask, &val); | |
1705 | ||
c565650b RR |
1706 | /* Clear the bits we're about to read */ |
1707 | eax &= ~mask; | |
1708 | /* Copy bits in from val. */ | |
1709 | eax |= val & mask; | |
1710 | /* Now update the register. */ | |
1711 | setreg(eax, eax); | |
d7fbf6e9 RR |
1712 | } else { |
1713 | if (is_pci_addr_port(port)) { | |
1714 | if (!pci_addr_iowrite(port, mask, eax)) | |
1715 | goto bad_io; | |
1716 | } else if (is_pci_data_port(port)) { | |
1717 | if (!pci_data_iowrite(port, mask, eax)) | |
1718 | goto bad_io; | |
1719 | } | |
1720 | /* There are many other ports, eg. CMOS clock, serial | |
1721 | * and parallel ports, so we ignore them all. */ | |
c565650b RR |
1722 | } |
1723 | ||
1724 | verbose("IO %s of %x to %u: %#08x\n", | |
1725 | in ? "IN" : "OUT", mask, port, eax); | |
1726 | skip_insn: | |
1727 | /* Finally, we've "done" the instruction, so move past it. */ | |
1728 | setreg(eip, getreg(eip) + insnlen); | |
1729 | return; | |
1730 | ||
d7fbf6e9 RR |
1731 | bad_io: |
1732 | warnx("Attempt to %s port %u (%#x mask)", | |
1733 | in ? "read from" : "write to", port, mask); | |
1734 | ||
c565650b RR |
1735 | no_emulate: |
1736 | /* Inject trap into Guest. */ | |
1737 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
1738 | err(1, "Reinjecting trap 13 for fault at %#x", getreg(eip)); | |
1739 | } | |
1740 | ||
6a54f9ab RR |
1741 | static struct device *find_mmio_region(unsigned long paddr, u32 *off) |
1742 | { | |
1743 | unsigned int i; | |
1744 | ||
1745 | for (i = 1; i < MAX_PCI_DEVICES; i++) { | |
1746 | struct device *d = devices.pci[i]; | |
1747 | ||
1748 | if (!d) | |
1749 | continue; | |
1750 | if (paddr < d->mmio_addr) | |
1751 | continue; | |
1752 | if (paddr >= d->mmio_addr + d->mmio_size) | |
1753 | continue; | |
1754 | *off = paddr - d->mmio_addr; | |
1755 | return d; | |
1756 | } | |
1757 | return NULL; | |
1758 | } | |
1759 | ||
93153077 RR |
1760 | /* FIXME: Use vq array. */ |
1761 | static struct virtqueue *vq_by_num(struct device *d, u32 num) | |
1762 | { | |
1763 | struct virtqueue *vq = d->vq; | |
1764 | ||
1765 | while (num-- && vq) | |
1766 | vq = vq->next; | |
1767 | ||
1768 | return vq; | |
1769 | } | |
1770 | ||
1771 | static void save_vq_config(const struct virtio_pci_common_cfg *cfg, | |
1772 | struct virtqueue *vq) | |
1773 | { | |
1774 | vq->pci_config = *cfg; | |
1775 | } | |
1776 | ||
1777 | static void restore_vq_config(struct virtio_pci_common_cfg *cfg, | |
1778 | struct virtqueue *vq) | |
1779 | { | |
1780 | /* Only restore the per-vq part */ | |
1781 | size_t off = offsetof(struct virtio_pci_common_cfg, queue_size); | |
1782 | ||
1783 | memcpy((void *)cfg + off, (void *)&vq->pci_config + off, | |
1784 | sizeof(*cfg) - off); | |
1785 | } | |
1786 | ||
1787 | /* | |
d761b032 RR |
1788 | * 4.1.4.3.2: |
1789 | * | |
1790 | * The driver MUST configure the other virtqueue fields before | |
1791 | * enabling the virtqueue with queue_enable. | |
1792 | * | |
93153077 RR |
1793 | * When they enable the virtqueue, we check that their setup is valid. |
1794 | */ | |
d761b032 | 1795 | static void check_virtqueue(struct device *d, struct virtqueue *vq) |
93153077 | 1796 | { |
93153077 RR |
1797 | /* Because lguest is 32 bit, all the descriptor high bits must be 0 */ |
1798 | if (vq->pci_config.queue_desc_hi | |
1799 | || vq->pci_config.queue_avail_hi | |
1800 | || vq->pci_config.queue_used_hi) | |
1e1c17a7 | 1801 | bad_driver_vq(vq, "invalid 64-bit queue address"); |
93153077 | 1802 | |
d39a6785 RR |
1803 | /* |
1804 | * 2.4.1: | |
1805 | * | |
1806 | * The driver MUST ensure that the physical address of the first byte | |
1807 | * of each virtqueue part is a multiple of the specified alignment | |
1808 | * value in the above table. | |
1809 | */ | |
1810 | if (vq->pci_config.queue_desc_lo % 16 | |
1811 | || vq->pci_config.queue_avail_lo % 2 | |
1812 | || vq->pci_config.queue_used_lo % 4) | |
1e1c17a7 | 1813 | bad_driver_vq(vq, "invalid alignment in queue addresses"); |
d39a6785 | 1814 | |
93153077 RR |
1815 | /* Initialize the virtqueue and check they're all in range. */ |
1816 | vq->vring.num = vq->pci_config.queue_size; | |
1e1c17a7 RR |
1817 | vq->vring.desc = check_pointer(vq->dev, |
1818 | vq->pci_config.queue_desc_lo, | |
93153077 | 1819 | sizeof(*vq->vring.desc) * vq->vring.num); |
1e1c17a7 RR |
1820 | vq->vring.avail = check_pointer(vq->dev, |
1821 | vq->pci_config.queue_avail_lo, | |
93153077 RR |
1822 | sizeof(*vq->vring.avail) |
1823 | + (sizeof(vq->vring.avail->ring[0]) | |
1824 | * vq->vring.num)); | |
1e1c17a7 RR |
1825 | vq->vring.used = check_pointer(vq->dev, |
1826 | vq->pci_config.queue_used_lo, | |
93153077 RR |
1827 | sizeof(*vq->vring.used) |
1828 | + (sizeof(vq->vring.used->ring[0]) | |
1829 | * vq->vring.num)); | |
d39a6785 RR |
1830 | |
1831 | /* | |
1832 | * 2.4.9.1: | |
1833 | * | |
1834 | * The driver MUST initialize flags in the used ring to 0 | |
1835 | * when allocating the used ring. | |
1836 | */ | |
1837 | if (vq->vring.used->flags != 0) | |
1e1c17a7 RR |
1838 | bad_driver_vq(vq, "invalid initial used.flags %#x", |
1839 | vq->vring.used->flags); | |
d761b032 | 1840 | } |
93153077 | 1841 | |
d761b032 RR |
1842 | static void start_virtqueue(struct virtqueue *vq) |
1843 | { | |
1844 | /* | |
1845 | * Create stack for thread. Since the stack grows upwards, we point | |
1846 | * the stack pointer to the end of this region. | |
1847 | */ | |
1848 | char *stack = malloc(32768); | |
93153077 RR |
1849 | |
1850 | /* Create a zero-initialized eventfd. */ | |
1851 | vq->eventfd = eventfd(0, 0); | |
1852 | if (vq->eventfd < 0) | |
1853 | err(1, "Creating eventfd"); | |
1854 | ||
1855 | /* | |
1856 | * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so | |
1857 | * we get a signal if it dies. | |
1858 | */ | |
1859 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); | |
1860 | if (vq->thread == (pid_t)-1) | |
1861 | err(1, "Creating clone"); | |
1862 | } | |
1863 | ||
d761b032 RR |
1864 | static void start_virtqueues(struct device *d) |
1865 | { | |
1866 | struct virtqueue *vq; | |
1867 | ||
1868 | for (vq = d->vq; vq; vq = vq->next) { | |
1869 | if (vq->pci_config.queue_enable) | |
1870 | start_virtqueue(vq); | |
1871 | } | |
1872 | } | |
1873 | ||
6a54f9ab RR |
1874 | static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask) |
1875 | { | |
93153077 RR |
1876 | struct virtqueue *vq; |
1877 | ||
1878 | switch (off) { | |
1879 | case offsetof(struct virtio_pci_mmio, cfg.device_feature_select): | |
8dc425ff RR |
1880 | /* |
1881 | * 4.1.4.3.1: | |
1882 | * | |
1883 | * The device MUST present the feature bits it is offering in | |
1884 | * device_feature, starting at bit device_feature_select ∗ 32 | |
1885 | * for any device_feature_select written by the driver | |
1886 | */ | |
93153077 RR |
1887 | if (val == 0) |
1888 | d->mmio->cfg.device_feature = d->features; | |
1889 | else if (val == 1) | |
1890 | d->mmio->cfg.device_feature = (d->features >> 32); | |
1891 | else | |
1892 | d->mmio->cfg.device_feature = 0; | |
d39a6785 | 1893 | goto feature_write_through32; |
93153077 RR |
1894 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select): |
1895 | if (val > 1) | |
1e1c17a7 | 1896 | bad_driver(d, "Unexpected driver select %u", val); |
d39a6785 | 1897 | goto feature_write_through32; |
93153077 RR |
1898 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature): |
1899 | if (d->mmio->cfg.guest_feature_select == 0) { | |
1900 | d->features_accepted &= ~((u64)0xFFFFFFFF); | |
1901 | d->features_accepted |= val; | |
1902 | } else { | |
1903 | assert(d->mmio->cfg.guest_feature_select == 1); | |
53aceb49 | 1904 | d->features_accepted &= 0xFFFFFFFF; |
93153077 RR |
1905 | d->features_accepted |= ((u64)val) << 32; |
1906 | } | |
d39a6785 RR |
1907 | /* |
1908 | * 2.2.1: | |
1909 | * | |
1910 | * The driver MUST NOT accept a feature which the device did | |
1911 | * not offer | |
1912 | */ | |
93153077 | 1913 | if (d->features_accepted & ~d->features) |
1e1c17a7 RR |
1914 | bad_driver(d, "over-accepted features %#llx of %#llx", |
1915 | d->features_accepted, d->features); | |
d39a6785 RR |
1916 | goto feature_write_through32; |
1917 | case offsetof(struct virtio_pci_mmio, cfg.device_status): { | |
1918 | u8 prev; | |
1919 | ||
93153077 | 1920 | verbose("%s: device status -> %#x\n", d->name, val); |
8dc425ff RR |
1921 | /* |
1922 | * 4.1.4.3.1: | |
1923 | * | |
1924 | * The device MUST reset when 0 is written to device_status, | |
1925 | * and present a 0 in device_status once that is done. | |
1926 | */ | |
d39a6785 | 1927 | if (val == 0) { |
d9028eda | 1928 | reset_device(d); |
d39a6785 RR |
1929 | goto write_through8; |
1930 | } | |
1931 | ||
1932 | /* 2.1.1: The driver MUST NOT clear a device status bit. */ | |
1933 | if (d->mmio->cfg.device_status & ~val) | |
1e1c17a7 RR |
1934 | bad_driver(d, "unset of device status bit %#x -> %#x", |
1935 | d->mmio->cfg.device_status, val); | |
d761b032 RR |
1936 | |
1937 | /* | |
1938 | * 2.1.2: | |
1939 | * | |
1940 | * The device MUST NOT consume buffers or notify the driver | |
1941 | * before DRIVER_OK. | |
1942 | */ | |
1943 | if (val & VIRTIO_CONFIG_S_DRIVER_OK | |
1944 | && !(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK)) | |
1945 | start_virtqueues(d); | |
1946 | ||
d39a6785 RR |
1947 | /* |
1948 | * 3.1.1: | |
1949 | * | |
1950 | * The driver MUST follow this sequence to initialize a device: | |
1951 | * - Reset the device. | |
1952 | * - Set the ACKNOWLEDGE status bit: the guest OS has | |
1953 | * notice the device. | |
1954 | * - Set the DRIVER status bit: the guest OS knows how | |
1955 | * to drive the device. | |
1956 | * - Read device feature bits, and write the subset | |
1957 | * of feature bits understood by the OS and driver | |
1958 | * to the device. During this step the driver MAY | |
1959 | * read (but MUST NOT write) the device-specific | |
1960 | * configuration fields to check that it can | |
1961 | * support the device before accepting it. | |
1962 | * - Set the FEATURES_OK status bit. The driver | |
1963 | * MUST not accept new feature bits after this | |
1964 | * step. | |
1965 | * - Re-read device status to ensure the FEATURES_OK | |
1966 | * bit is still set: otherwise, the device does | |
1967 | * not support our subset of features and the | |
1968 | * device is unusable. | |
1969 | * - Perform device-specific setup, including | |
1970 | * discovery of virtqueues for the device, | |
1971 | * optional per-bus setup, reading and possibly | |
1972 | * writing the device’s virtio configuration | |
1973 | * space, and population of virtqueues. | |
1974 | * - Set the DRIVER_OK status bit. At this point the | |
1975 | * device is “live”. | |
1976 | */ | |
1977 | prev = 0; | |
1978 | switch (val & ~d->mmio->cfg.device_status) { | |
1979 | case VIRTIO_CONFIG_S_DRIVER_OK: | |
1980 | prev |= VIRTIO_CONFIG_S_FEATURES_OK; /* fall thru */ | |
1981 | case VIRTIO_CONFIG_S_FEATURES_OK: | |
1982 | prev |= VIRTIO_CONFIG_S_DRIVER; /* fall thru */ | |
1983 | case VIRTIO_CONFIG_S_DRIVER: | |
1984 | prev |= VIRTIO_CONFIG_S_ACKNOWLEDGE; /* fall thru */ | |
1985 | case VIRTIO_CONFIG_S_ACKNOWLEDGE: | |
1986 | break; | |
1987 | default: | |
1e1c17a7 RR |
1988 | bad_driver(d, "unknown device status bit %#x -> %#x", |
1989 | d->mmio->cfg.device_status, val); | |
d39a6785 RR |
1990 | } |
1991 | if (d->mmio->cfg.device_status != prev) | |
1e1c17a7 RR |
1992 | bad_driver(d, "unexpected status transition %#x -> %#x", |
1993 | d->mmio->cfg.device_status, val); | |
d39a6785 RR |
1994 | |
1995 | /* If they just wrote FEATURES_OK, we make sure they read */ | |
1996 | switch (val & ~d->mmio->cfg.device_status) { | |
1997 | case VIRTIO_CONFIG_S_FEATURES_OK: | |
1998 | d->wrote_features_ok = true; | |
1999 | break; | |
2000 | case VIRTIO_CONFIG_S_DRIVER_OK: | |
2001 | if (d->wrote_features_ok) | |
1e1c17a7 | 2002 | bad_driver(d, "did not re-read FEATURES_OK"); |
d39a6785 RR |
2003 | break; |
2004 | } | |
93153077 | 2005 | goto write_through8; |
d39a6785 | 2006 | } |
93153077 RR |
2007 | case offsetof(struct virtio_pci_mmio, cfg.queue_select): |
2008 | vq = vq_by_num(d, val); | |
8dc425ff RR |
2009 | /* |
2010 | * 4.1.4.3.1: | |
2011 | * | |
2012 | * The device MUST present a 0 in queue_size if the virtqueue | |
2013 | * corresponding to the current queue_select is unavailable. | |
2014 | */ | |
93153077 RR |
2015 | if (!vq) { |
2016 | d->mmio->cfg.queue_size = 0; | |
2017 | goto write_through16; | |
2018 | } | |
2019 | /* Save registers for old vq, if it was a valid vq */ | |
2020 | if (d->mmio->cfg.queue_size) | |
2021 | save_vq_config(&d->mmio->cfg, | |
2022 | vq_by_num(d, d->mmio->cfg.queue_select)); | |
2023 | /* Restore the registers for the queue they asked for */ | |
2024 | restore_vq_config(&d->mmio->cfg, vq); | |
2025 | goto write_through16; | |
2026 | case offsetof(struct virtio_pci_mmio, cfg.queue_size): | |
c97eb679 RR |
2027 | /* |
2028 | * 4.1.4.3.2: | |
2029 | * | |
2030 | * The driver MUST NOT write a value which is not a power of 2 | |
2031 | * to queue_size. | |
2032 | */ | |
93153077 | 2033 | if (val & (val-1)) |
1e1c17a7 | 2034 | bad_driver(d, "invalid queue size %u", val); |
93153077 | 2035 | if (d->mmio->cfg.queue_enable) |
1e1c17a7 | 2036 | bad_driver(d, "changing queue size on live device"); |
93153077 RR |
2037 | goto write_through16; |
2038 | case offsetof(struct virtio_pci_mmio, cfg.queue_msix_vector): | |
1e1c17a7 | 2039 | bad_driver(d, "attempt to set MSIX vector to %u", val); |
d39a6785 RR |
2040 | case offsetof(struct virtio_pci_mmio, cfg.queue_enable): { |
2041 | struct virtqueue *vq = vq_by_num(d, d->mmio->cfg.queue_select); | |
2042 | ||
c97eb679 RR |
2043 | /* |
2044 | * 4.1.4.3.2: | |
2045 | * | |
2046 | * The driver MUST NOT write a 0 to queue_enable. | |
2047 | */ | |
93153077 | 2048 | if (val != 1) |
1e1c17a7 | 2049 | bad_driver(d, "setting queue_enable to %u", val); |
d39a6785 | 2050 | |
c97eb679 | 2051 | /* |
d39a6785 | 2052 | * 3.1.1: |
c97eb679 | 2053 | * |
d39a6785 RR |
2054 | * 7. Perform device-specific setup, including discovery of |
2055 | * virtqueues for the device, optional per-bus setup, | |
2056 | * reading and possibly writing the device’s virtio | |
2057 | * configuration space, and population of virtqueues. | |
2058 | * 8. Set the DRIVER_OK status bit. | |
2059 | * | |
2060 | * All our devices require all virtqueues to be enabled, so | |
2061 | * they should have done that before setting DRIVER_OK. | |
c97eb679 | 2062 | */ |
d39a6785 | 2063 | if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK) |
1e1c17a7 | 2064 | bad_driver(d, "enabling vq after DRIVER_OK"); |
d39a6785 RR |
2065 | |
2066 | d->mmio->cfg.queue_enable = val; | |
2067 | save_vq_config(&d->mmio->cfg, vq); | |
2068 | check_virtqueue(d, vq); | |
93153077 | 2069 | goto write_through16; |
d39a6785 | 2070 | } |
93153077 | 2071 | case offsetof(struct virtio_pci_mmio, cfg.queue_notify_off): |
1e1c17a7 | 2072 | bad_driver(d, "attempt to write to queue_notify_off"); |
93153077 RR |
2073 | case offsetof(struct virtio_pci_mmio, cfg.queue_desc_lo): |
2074 | case offsetof(struct virtio_pci_mmio, cfg.queue_desc_hi): | |
2075 | case offsetof(struct virtio_pci_mmio, cfg.queue_avail_lo): | |
2076 | case offsetof(struct virtio_pci_mmio, cfg.queue_avail_hi): | |
2077 | case offsetof(struct virtio_pci_mmio, cfg.queue_used_lo): | |
2078 | case offsetof(struct virtio_pci_mmio, cfg.queue_used_hi): | |
c97eb679 RR |
2079 | /* |
2080 | * 4.1.4.3.2: | |
2081 | * | |
2082 | * The driver MUST configure the other virtqueue fields before | |
2083 | * enabling the virtqueue with queue_enable. | |
2084 | */ | |
93153077 | 2085 | if (d->mmio->cfg.queue_enable) |
1e1c17a7 | 2086 | bad_driver(d, "changing queue on live device"); |
d39a6785 RR |
2087 | |
2088 | /* | |
2089 | * 3.1.1: | |
2090 | * | |
2091 | * The driver MUST follow this sequence to initialize a device: | |
2092 | *... | |
2093 | * 5. Set the FEATURES_OK status bit. The driver MUST not | |
2094 | * accept new feature bits after this step. | |
2095 | */ | |
2096 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK)) | |
1e1c17a7 | 2097 | bad_driver(d, "setting up vq before FEATURES_OK"); |
d39a6785 RR |
2098 | |
2099 | /* | |
2100 | * 6. Re-read device status to ensure the FEATURES_OK bit is | |
2101 | * still set... | |
2102 | */ | |
2103 | if (d->wrote_features_ok) | |
1e1c17a7 | 2104 | bad_driver(d, "didn't re-read FEATURES_OK before setup"); |
d39a6785 | 2105 | |
93153077 RR |
2106 | goto write_through32; |
2107 | case offsetof(struct virtio_pci_mmio, notify): | |
2108 | vq = vq_by_num(d, val); | |
2109 | if (!vq) | |
1e1c17a7 | 2110 | bad_driver(d, "Invalid vq notification on %u", val); |
93153077 RR |
2111 | /* Notify the process handling this vq by adding 1 to eventfd */ |
2112 | write(vq->eventfd, "\1\0\0\0\0\0\0\0", 8); | |
2113 | goto write_through16; | |
2114 | case offsetof(struct virtio_pci_mmio, isr): | |
1e1c17a7 | 2115 | bad_driver(d, "Unexpected write to isr"); |
e8330d9b RR |
2116 | /* Weird corner case: write to emerg_wr of console */ |
2117 | case sizeof(struct virtio_pci_mmio) | |
2118 | + offsetof(struct virtio_console_config, emerg_wr): | |
2119 | if (strcmp(d->name, "console") == 0) { | |
2120 | char c = val; | |
2121 | write(STDOUT_FILENO, &c, 1); | |
2122 | goto write_through32; | |
2123 | } | |
2124 | /* Fall through... */ | |
93153077 | 2125 | default: |
c97eb679 RR |
2126 | /* |
2127 | * 4.1.4.3.2: | |
2128 | * | |
2129 | * The driver MUST NOT write to device_feature, num_queues, | |
2130 | * config_generation or queue_notify_off. | |
2131 | */ | |
1e1c17a7 | 2132 | bad_driver(d, "Unexpected write to offset %u", off); |
93153077 RR |
2133 | } |
2134 | ||
d39a6785 RR |
2135 | feature_write_through32: |
2136 | /* | |
2137 | * 3.1.1: | |
2138 | * | |
2139 | * The driver MUST follow this sequence to initialize a device: | |
2140 | *... | |
2141 | * - Set the DRIVER status bit: the guest OS knows how | |
2142 | * to drive the device. | |
2143 | * - Read device feature bits, and write the subset | |
2144 | * of feature bits understood by the OS and driver | |
2145 | * to the device. | |
2146 | *... | |
2147 | * - Set the FEATURES_OK status bit. The driver MUST not | |
2148 | * accept new feature bits after this step. | |
2149 | */ | |
2150 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
1e1c17a7 | 2151 | bad_driver(d, "feature write before VIRTIO_CONFIG_S_DRIVER"); |
d39a6785 | 2152 | if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK) |
1e1c17a7 | 2153 | bad_driver(d, "feature write after VIRTIO_CONFIG_S_FEATURES_OK"); |
c97eb679 RR |
2154 | |
2155 | /* | |
2156 | * 4.1.3.1: | |
2157 | * | |
2158 | * The driver MUST access each field using the “natural” access | |
2159 | * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for | |
2160 | * 16-bit fields and 8-bit accesses for 8-bit fields. | |
2161 | */ | |
93153077 RR |
2162 | write_through32: |
2163 | if (mask != 0xFFFFFFFF) { | |
1e1c17a7 RR |
2164 | bad_driver(d, "non-32-bit write to offset %u (%#x)", |
2165 | off, getreg(eip)); | |
93153077 RR |
2166 | return; |
2167 | } | |
2168 | memcpy((char *)d->mmio + off, &val, 4); | |
2169 | return; | |
2170 | ||
2171 | write_through16: | |
2172 | if (mask != 0xFFFF) | |
1e1c17a7 RR |
2173 | bad_driver(d, "non-16-bit write to offset %u (%#x)", |
2174 | off, getreg(eip)); | |
93153077 RR |
2175 | memcpy((char *)d->mmio + off, &val, 2); |
2176 | return; | |
2177 | ||
2178 | write_through8: | |
2179 | if (mask != 0xFF) | |
1e1c17a7 RR |
2180 | bad_driver(d, "non-8-bit write to offset %u (%#x)", |
2181 | off, getreg(eip)); | |
93153077 RR |
2182 | memcpy((char *)d->mmio + off, &val, 1); |
2183 | return; | |
6a54f9ab RR |
2184 | } |
2185 | ||
2186 | static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask) | |
2187 | { | |
93153077 RR |
2188 | u8 isr; |
2189 | u32 val = 0; | |
2190 | ||
2191 | switch (off) { | |
2192 | case offsetof(struct virtio_pci_mmio, cfg.device_feature_select): | |
2193 | case offsetof(struct virtio_pci_mmio, cfg.device_feature): | |
2194 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select): | |
2195 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature): | |
d39a6785 RR |
2196 | /* |
2197 | * 3.1.1: | |
2198 | * | |
2199 | * The driver MUST follow this sequence to initialize a device: | |
2200 | *... | |
2201 | * - Set the DRIVER status bit: the guest OS knows how | |
2202 | * to drive the device. | |
2203 | * - Read device feature bits, and write the subset | |
2204 | * of feature bits understood by the OS and driver | |
2205 | * to the device. | |
2206 | */ | |
2207 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
1e1c17a7 RR |
2208 | bad_driver(d, |
2209 | "feature read before VIRTIO_CONFIG_S_DRIVER"); | |
93153077 RR |
2210 | goto read_through32; |
2211 | case offsetof(struct virtio_pci_mmio, cfg.msix_config): | |
1e1c17a7 | 2212 | bad_driver(d, "read of msix_config"); |
93153077 RR |
2213 | case offsetof(struct virtio_pci_mmio, cfg.num_queues): |
2214 | goto read_through16; | |
2215 | case offsetof(struct virtio_pci_mmio, cfg.device_status): | |
d39a6785 RR |
2216 | /* As they did read, any write of FEATURES_OK is now fine. */ |
2217 | d->wrote_features_ok = false; | |
2218 | goto read_through8; | |
93153077 | 2219 | case offsetof(struct virtio_pci_mmio, cfg.config_generation): |
8dc425ff RR |
2220 | /* |
2221 | * 4.1.4.3.1: | |
2222 | * | |
2223 | * The device MUST present a changed config_generation after | |
2224 | * the driver has read a device-specific configuration value | |
2225 | * which has changed since any part of the device-specific | |
2226 | * configuration was last read. | |
2227 | * | |
2228 | * This is simple: none of our devices change config, so this | |
2229 | * is always 0. | |
2230 | */ | |
93153077 RR |
2231 | goto read_through8; |
2232 | case offsetof(struct virtio_pci_mmio, notify): | |
d39a6785 RR |
2233 | /* |
2234 | * 3.1.1: | |
2235 | * | |
2236 | * The driver MUST NOT notify the device before setting | |
2237 | * DRIVER_OK. | |
2238 | */ | |
2239 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK)) | |
1e1c17a7 | 2240 | bad_driver(d, "notify before VIRTIO_CONFIG_S_DRIVER_OK"); |
93153077 RR |
2241 | goto read_through16; |
2242 | case offsetof(struct virtio_pci_mmio, isr): | |
2243 | if (mask != 0xFF) | |
1e1c17a7 RR |
2244 | bad_driver(d, "non-8-bit read from offset %u (%#x)", |
2245 | off, getreg(eip)); | |
93153077 | 2246 | isr = d->mmio->isr; |
8dc425ff RR |
2247 | /* |
2248 | * 4.1.4.5.1: | |
2249 | * | |
2250 | * The device MUST reset ISR status to 0 on driver read. | |
2251 | */ | |
93153077 RR |
2252 | d->mmio->isr = 0; |
2253 | return isr; | |
2254 | case offsetof(struct virtio_pci_mmio, padding): | |
1e1c17a7 | 2255 | bad_driver(d, "read from padding (%#x)", getreg(eip)); |
93153077 RR |
2256 | default: |
2257 | /* Read from device config space, beware unaligned overflow */ | |
2258 | if (off > d->mmio_size - 4) | |
1e1c17a7 | 2259 | bad_driver(d, "read past end (%#x)", getreg(eip)); |
d39a6785 RR |
2260 | |
2261 | /* | |
2262 | * 3.1.1: | |
2263 | * The driver MUST follow this sequence to initialize a device: | |
2264 | *... | |
2265 | * 3. Set the DRIVER status bit: the guest OS knows how to | |
2266 | * drive the device. | |
2267 | * 4. Read device feature bits, and write the subset of | |
2268 | * feature bits understood by the OS and driver to the | |
2269 | * device. During this step the driver MAY read (but MUST NOT | |
2270 | * write) the device-specific configuration fields to check | |
2271 | * that it can support the device before accepting it. | |
2272 | */ | |
2273 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
1e1c17a7 RR |
2274 | bad_driver(d, |
2275 | "config read before VIRTIO_CONFIG_S_DRIVER"); | |
d39a6785 | 2276 | |
93153077 RR |
2277 | if (mask == 0xFFFFFFFF) |
2278 | goto read_through32; | |
2279 | else if (mask == 0xFFFF) | |
2280 | goto read_through16; | |
2281 | else | |
2282 | goto read_through8; | |
2283 | } | |
2284 | ||
c97eb679 RR |
2285 | /* |
2286 | * 4.1.3.1: | |
2287 | * | |
2288 | * The driver MUST access each field using the “natural” access | |
2289 | * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for | |
2290 | * 16-bit fields and 8-bit accesses for 8-bit fields. | |
2291 | */ | |
93153077 RR |
2292 | read_through32: |
2293 | if (mask != 0xFFFFFFFF) | |
1e1c17a7 RR |
2294 | bad_driver(d, "non-32-bit read to offset %u (%#x)", |
2295 | off, getreg(eip)); | |
93153077 RR |
2296 | memcpy(&val, (char *)d->mmio + off, 4); |
2297 | return val; | |
2298 | ||
2299 | read_through16: | |
2300 | if (mask != 0xFFFF) | |
1e1c17a7 RR |
2301 | bad_driver(d, "non-16-bit read to offset %u (%#x)", |
2302 | off, getreg(eip)); | |
93153077 RR |
2303 | memcpy(&val, (char *)d->mmio + off, 2); |
2304 | return val; | |
2305 | ||
2306 | read_through8: | |
2307 | if (mask != 0xFF) | |
1e1c17a7 RR |
2308 | bad_driver(d, "non-8-bit read to offset %u (%#x)", |
2309 | off, getreg(eip)); | |
93153077 RR |
2310 | memcpy(&val, (char *)d->mmio + off, 1); |
2311 | return val; | |
6a54f9ab RR |
2312 | } |
2313 | ||
2314 | static void emulate_mmio(unsigned long paddr, const u8 *insn) | |
2315 | { | |
2316 | u32 val, off, mask = 0xFFFFFFFF, insnlen = 0; | |
2317 | struct device *d = find_mmio_region(paddr, &off); | |
2318 | unsigned long args[] = { LHREQ_TRAP, 14 }; | |
2319 | ||
2320 | if (!d) { | |
2321 | warnx("MMIO touching %#08lx (not a device)", paddr); | |
2322 | goto reinject; | |
2323 | } | |
2324 | ||
2325 | /* Prefix makes it a 16 bit op */ | |
2326 | if (insn[0] == 0x66) { | |
2327 | mask = 0xFFFF; | |
2328 | insnlen++; | |
2329 | } | |
2330 | ||
2331 | /* iowrite */ | |
2332 | if (insn[insnlen] == 0x89) { | |
2333 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2334 | val = getreg_num((insn[insnlen+1] >> 3) & 0x7, mask); | |
2335 | emulate_mmio_write(d, off, val, mask); | |
2336 | insnlen += 2 + insn_displacement_len(insn[insnlen+1]); | |
2337 | } else if (insn[insnlen] == 0x8b) { /* ioread */ | |
2338 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2339 | val = emulate_mmio_read(d, off, mask); | |
2340 | setreg_num((insn[insnlen+1] >> 3) & 0x7, val, mask); | |
2341 | insnlen += 2 + insn_displacement_len(insn[insnlen+1]); | |
2342 | } else if (insn[0] == 0x88) { /* 8-bit iowrite */ | |
2343 | mask = 0xff; | |
2344 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2345 | val = getreg_num((insn[1] >> 3) & 0x7, mask); | |
2346 | emulate_mmio_write(d, off, val, mask); | |
2347 | insnlen = 2 + insn_displacement_len(insn[1]); | |
2348 | } else if (insn[0] == 0x8a) { /* 8-bit ioread */ | |
2349 | mask = 0xff; | |
2350 | val = emulate_mmio_read(d, off, mask); | |
2351 | setreg_num((insn[1] >> 3) & 0x7, val, mask); | |
2352 | insnlen = 2 + insn_displacement_len(insn[1]); | |
2353 | } else { | |
2354 | warnx("Unknown MMIO instruction touching %#08lx:" | |
2355 | " %02x %02x %02x %02x at %u", | |
2356 | paddr, insn[0], insn[1], insn[2], insn[3], getreg(eip)); | |
2357 | reinject: | |
2358 | /* Inject trap into Guest. */ | |
2359 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
2360 | err(1, "Reinjecting trap 14 for fault at %#x", | |
2361 | getreg(eip)); | |
2362 | return; | |
2363 | } | |
2364 | ||
2365 | /* Finally, we've "done" the instruction, so move past it. */ | |
2366 | setreg(eip, getreg(eip) + insnlen); | |
2367 | } | |
c565650b | 2368 | |
dde79789 RR |
2369 | /*L:190 |
2370 | * Device Setup | |
2371 | * | |
2372 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
2373 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
2374 | * routines to allocate and manage them. |
2375 | */ | |
93153077 | 2376 | static void add_pci_virtqueue(struct device *dev, |
17c56d6d RR |
2377 | void (*service)(struct virtqueue *), |
2378 | const char *name) | |
93153077 RR |
2379 | { |
2380 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
2381 | ||
2382 | /* Initialize the virtqueue */ | |
2383 | vq->next = NULL; | |
2384 | vq->last_avail_idx = 0; | |
2385 | vq->dev = dev; | |
17c56d6d | 2386 | vq->name = name; |
93153077 RR |
2387 | |
2388 | /* | |
2389 | * This is the routine the service thread will run, and its Process ID | |
2390 | * once it's running. | |
2391 | */ | |
2392 | vq->service = service; | |
2393 | vq->thread = (pid_t)-1; | |
2394 | ||
2395 | /* Initialize the configuration. */ | |
d2dbdac3 | 2396 | reset_vq_pci_config(vq); |
93153077 RR |
2397 | vq->pci_config.queue_notify_off = 0; |
2398 | ||
2399 | /* Add one to the number of queues */ | |
2400 | vq->dev->mmio->cfg.num_queues++; | |
2401 | ||
93153077 RR |
2402 | /* |
2403 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
2404 | * second. | |
2405 | */ | |
2406 | for (i = &dev->vq; *i; i = &(*i)->next); | |
2407 | *i = vq; | |
2408 | } | |
2409 | ||
d9028eda | 2410 | /* The Guest accesses the feature bits via the PCI common config MMIO region */ |
93153077 RR |
2411 | static void add_pci_feature(struct device *dev, unsigned bit) |
2412 | { | |
2413 | dev->features |= (1ULL << bit); | |
2414 | } | |
2415 | ||
93153077 RR |
2416 | /* For devices with no config. */ |
2417 | static void no_device_config(struct device *dev) | |
2418 | { | |
2419 | dev->mmio_addr = get_mmio_region(dev->mmio_size); | |
2420 | ||
2421 | dev->config.bar[0] = dev->mmio_addr; | |
2422 | /* Bottom 4 bits must be zero */ | |
2423 | assert(~(dev->config.bar[0] & 0xF)); | |
2424 | } | |
2425 | ||
2426 | /* This puts the device config into BAR0 */ | |
2427 | static void set_device_config(struct device *dev, const void *conf, size_t len) | |
2428 | { | |
2429 | /* Set up BAR 0 */ | |
2430 | dev->mmio_size += len; | |
2431 | dev->mmio = realloc(dev->mmio, dev->mmio_size); | |
2432 | memcpy(dev->mmio + 1, conf, len); | |
2433 | ||
8dc425ff RR |
2434 | /* |
2435 | * 4.1.4.6: | |
2436 | * | |
2437 | * The device MUST present at least one VIRTIO_PCI_CAP_DEVICE_CFG | |
2438 | * capability for any device type which has a device-specific | |
2439 | * configuration. | |
2440 | */ | |
93153077 RR |
2441 | /* Hook up device cfg */ |
2442 | dev->config.cfg_access.cap.cap_next | |
2443 | = offsetof(struct pci_config, device); | |
2444 | ||
8dc425ff RR |
2445 | /* |
2446 | * 4.1.4.6.1: | |
2447 | * | |
2448 | * The offset for the device-specific configuration MUST be 4-byte | |
2449 | * aligned. | |
2450 | */ | |
2451 | assert(dev->config.cfg_access.cap.cap_next % 4 == 0); | |
2452 | ||
93153077 RR |
2453 | /* Fix up device cfg field length. */ |
2454 | dev->config.device.length = len; | |
2455 | ||
2456 | /* The rest is the same as the no-config case */ | |
2457 | no_device_config(dev); | |
2458 | } | |
2459 | ||
2460 | static void init_cap(struct virtio_pci_cap *cap, size_t caplen, int type, | |
2461 | size_t bar_offset, size_t bar_bytes, u8 next) | |
2462 | { | |
2463 | cap->cap_vndr = PCI_CAP_ID_VNDR; | |
2464 | cap->cap_next = next; | |
2465 | cap->cap_len = caplen; | |
2466 | cap->cfg_type = type; | |
2467 | cap->bar = 0; | |
2468 | memset(cap->padding, 0, sizeof(cap->padding)); | |
2469 | cap->offset = bar_offset; | |
2470 | cap->length = bar_bytes; | |
2471 | } | |
2472 | ||
2473 | /* | |
2474 | * This sets up the pci_config structure, as defined in the virtio 1.0 | |
2475 | * standard (and PCI standard). | |
2476 | */ | |
2477 | static void init_pci_config(struct pci_config *pci, u16 type, | |
2478 | u8 class, u8 subclass) | |
2479 | { | |
2480 | size_t bar_offset, bar_len; | |
2481 | ||
8dc425ff RR |
2482 | /* |
2483 | * 4.1.4.4.1: | |
2484 | * | |
2485 | * The device MUST either present notify_off_multiplier as an even | |
2486 | * power of 2, or present notify_off_multiplier as 0. | |
d39a6785 RR |
2487 | * |
2488 | * 2.1.2: | |
2489 | * | |
2490 | * The device MUST initialize device status to 0 upon reset. | |
8dc425ff | 2491 | */ |
93153077 RR |
2492 | memset(pci, 0, sizeof(*pci)); |
2493 | ||
2494 | /* 4.1.2.1: Devices MUST have the PCI Vendor ID 0x1AF4 */ | |
2495 | pci->vendor_id = 0x1AF4; | |
2496 | /* 4.1.2.1: ... PCI Device ID calculated by adding 0x1040 ... */ | |
2497 | pci->device_id = 0x1040 + type; | |
2498 | ||
2499 | /* | |
2500 | * PCI have specific codes for different types of devices. | |
2501 | * Linux doesn't care, but it's a good clue for people looking | |
2502 | * at the device. | |
93153077 RR |
2503 | */ |
2504 | pci->class = class; | |
2505 | pci->subclass = subclass; | |
2506 | ||
2507 | /* | |
8dc425ff RR |
2508 | * 4.1.2.1: |
2509 | * | |
2510 | * Non-transitional devices SHOULD have a PCI Revision ID of 1 or | |
2511 | * higher | |
93153077 RR |
2512 | */ |
2513 | pci->revid = 1; | |
2514 | ||
2515 | /* | |
8dc425ff RR |
2516 | * 4.1.2.1: |
2517 | * | |
2518 | * Non-transitional devices SHOULD have a PCI Subsystem Device ID of | |
2519 | * 0x40 or higher. | |
93153077 RR |
2520 | */ |
2521 | pci->subsystem_device_id = 0x40; | |
2522 | ||
2523 | /* We use our dummy interrupt controller, and irq_line is the irq */ | |
2524 | pci->irq_line = devices.next_irq++; | |
2525 | pci->irq_pin = 0; | |
2526 | ||
2527 | /* Support for extended capabilities. */ | |
2528 | pci->status = (1 << 4); | |
2529 | ||
2530 | /* Link them in. */ | |
8dc425ff RR |
2531 | /* |
2532 | * 4.1.4.3.1: | |
2533 | * | |
2534 | * The device MUST present at least one common configuration | |
2535 | * capability. | |
2536 | */ | |
93153077 RR |
2537 | pci->capabilities = offsetof(struct pci_config, common); |
2538 | ||
8dc425ff RR |
2539 | /* 4.1.4.3.1 ... offset MUST be 4-byte aligned. */ |
2540 | assert(pci->capabilities % 4 == 0); | |
2541 | ||
93153077 RR |
2542 | bar_offset = offsetof(struct virtio_pci_mmio, cfg); |
2543 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->cfg); | |
2544 | init_cap(&pci->common, sizeof(pci->common), VIRTIO_PCI_CAP_COMMON_CFG, | |
2545 | bar_offset, bar_len, | |
2546 | offsetof(struct pci_config, notify)); | |
2547 | ||
8dc425ff RR |
2548 | /* |
2549 | * 4.1.4.4.1: | |
2550 | * | |
2551 | * The device MUST present at least one notification capability. | |
2552 | */ | |
93153077 RR |
2553 | bar_offset += bar_len; |
2554 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->notify); | |
8dc425ff RR |
2555 | |
2556 | /* | |
2557 | * 4.1.4.4.1: | |
2558 | * | |
2559 | * The cap.offset MUST be 2-byte aligned. | |
2560 | */ | |
2561 | assert(pci->common.cap_next % 2 == 0); | |
2562 | ||
93153077 | 2563 | /* FIXME: Use a non-zero notify_off, for per-queue notification? */ |
8dc425ff RR |
2564 | /* |
2565 | * 4.1.4.4.1: | |
2566 | * | |
2567 | * The value cap.length presented by the device MUST be at least 2 and | |
2568 | * MUST be large enough to support queue notification offsets for all | |
2569 | * supported queues in all possible configurations. | |
2570 | */ | |
2571 | assert(bar_len >= 2); | |
2572 | ||
93153077 RR |
2573 | init_cap(&pci->notify.cap, sizeof(pci->notify), |
2574 | VIRTIO_PCI_CAP_NOTIFY_CFG, | |
2575 | bar_offset, bar_len, | |
2576 | offsetof(struct pci_config, isr)); | |
2577 | ||
2578 | bar_offset += bar_len; | |
2579 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->isr); | |
8dc425ff RR |
2580 | /* |
2581 | * 4.1.4.5.1: | |
2582 | * | |
2583 | * The device MUST present at least one VIRTIO_PCI_CAP_ISR_CFG | |
2584 | * capability. | |
2585 | */ | |
93153077 RR |
2586 | init_cap(&pci->isr, sizeof(pci->isr), |
2587 | VIRTIO_PCI_CAP_ISR_CFG, | |
2588 | bar_offset, bar_len, | |
2589 | offsetof(struct pci_config, cfg_access)); | |
2590 | ||
8dc425ff RR |
2591 | /* |
2592 | * 4.1.4.7.1: | |
2593 | * | |
2594 | * The device MUST present at least one VIRTIO_PCI_CAP_PCI_CFG | |
2595 | * capability. | |
2596 | */ | |
93153077 RR |
2597 | /* This doesn't have any presence in the BAR */ |
2598 | init_cap(&pci->cfg_access.cap, sizeof(pci->cfg_access), | |
2599 | VIRTIO_PCI_CAP_PCI_CFG, | |
2600 | 0, 0, 0); | |
2601 | ||
2602 | bar_offset += bar_len + sizeof(((struct virtio_pci_mmio *)0)->padding); | |
2603 | assert(bar_offset == sizeof(struct virtio_pci_mmio)); | |
2604 | ||
2605 | /* | |
2606 | * This gets sewn in and length set in set_device_config(). | |
2607 | * Some devices don't have a device configuration interface, so | |
2608 | * we never expose this if we don't call set_device_config(). | |
2609 | */ | |
2610 | init_cap(&pci->device, sizeof(pci->device), VIRTIO_PCI_CAP_DEVICE_CFG, | |
2611 | bar_offset, 0, 0); | |
2612 | } | |
2613 | ||
2e04ef76 | 2614 | /* |
d9028eda RR |
2615 | * This routine does all the creation and setup of a new device, but we don't |
2616 | * actually place the MMIO region until we know the size (if any) of the | |
2617 | * device-specific config. And we don't actually start the service threads | |
2618 | * until later. | |
a6bd8e13 | 2619 | * |
2e04ef76 RR |
2620 | * See what I mean about userspace being boring? |
2621 | */ | |
93153077 RR |
2622 | static struct device *new_pci_device(const char *name, u16 type, |
2623 | u8 class, u8 subclass) | |
2624 | { | |
2625 | struct device *dev = malloc(sizeof(*dev)); | |
2626 | ||
2627 | /* Now we populate the fields one at a time. */ | |
93153077 RR |
2628 | dev->name = name; |
2629 | dev->vq = NULL; | |
93153077 | 2630 | dev->running = false; |
d39a6785 | 2631 | dev->wrote_features_ok = false; |
93153077 RR |
2632 | dev->mmio_size = sizeof(struct virtio_pci_mmio); |
2633 | dev->mmio = calloc(1, dev->mmio_size); | |
2634 | dev->features = (u64)1 << VIRTIO_F_VERSION_1; | |
2635 | dev->features_accepted = 0; | |
2636 | ||
d9028eda | 2637 | if (devices.device_num + 1 >= MAX_PCI_DEVICES) |
93153077 RR |
2638 | errx(1, "Can only handle 31 PCI devices"); |
2639 | ||
2640 | init_pci_config(&dev->config, type, class, subclass); | |
2641 | assert(!devices.pci[devices.device_num+1]); | |
2642 | devices.pci[++devices.device_num] = dev; | |
2643 | ||
2644 | return dev; | |
2645 | } | |
2646 | ||
2e04ef76 RR |
2647 | /* |
2648 | * Our first setup routine is the console. It's a fairly simple device, but | |
2649 | * UNIX tty handling makes it uglier than it could be. | |
2650 | */ | |
17cbca2b | 2651 | static void setup_console(void) |
8ca47e00 RR |
2652 | { |
2653 | struct device *dev; | |
e8330d9b | 2654 | struct virtio_console_config conf; |
8ca47e00 | 2655 | |
dde79789 | 2656 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
2657 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
2658 | struct termios term = orig_term; | |
2e04ef76 RR |
2659 | /* |
2660 | * Then we turn off echo, line buffering and ^C etc: We want a | |
2661 | * raw input stream to the Guest. | |
2662 | */ | |
8ca47e00 RR |
2663 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
2664 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
2665 | } |
2666 | ||
ebff0113 | 2667 | dev = new_pci_device("console", VIRTIO_ID_CONSOLE, 0x07, 0x00); |
659a0e66 | 2668 | |
dde79789 | 2669 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
2670 | dev->priv = malloc(sizeof(struct console_abort)); |
2671 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 2672 | |
2e04ef76 RR |
2673 | /* |
2674 | * The console needs two virtqueues: the input then the output. When | |
56ae43df RR |
2675 | * they put something the input queue, we make sure we're listening to |
2676 | * stdin. When they put something in the output queue, we write it to | |
2e04ef76 RR |
2677 | * stdout. |
2678 | */ | |
17c56d6d RR |
2679 | add_pci_virtqueue(dev, console_input, "input"); |
2680 | add_pci_virtqueue(dev, console_output, "output"); | |
ebff0113 | 2681 | |
e8330d9b RR |
2682 | /* We need a configuration area for the emerg_wr early writes. */ |
2683 | add_pci_feature(dev, VIRTIO_CONSOLE_F_EMERG_WRITE); | |
2684 | set_device_config(dev, &conf, sizeof(conf)); | |
17cbca2b | 2685 | |
ebff0113 | 2686 | verbose("device %u: console\n", devices.device_num); |
8ca47e00 | 2687 | } |
17cbca2b | 2688 | /*:*/ |
8ca47e00 | 2689 | |
2e04ef76 RR |
2690 | /*M:010 |
2691 | * Inter-guest networking is an interesting area. Simplest is to have a | |
17cbca2b RR |
2692 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
2693 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 2694 | * |
9f54288d | 2695 | * More sophisticated is to use one of the tools developed for project like UML |
17cbca2b | 2696 | * to do networking. |
dde79789 | 2697 | * |
17cbca2b RR |
2698 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
2699 | * completely generic ("here's my vring, attach to your vring") and would work | |
2700 | * for any traffic. Of course, namespace and permissions issues need to be | |
2701 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
2702 | * multiple inter-guest channels behind one interface, although it would | |
2703 | * require some manner of hotplugging new virtio channels. | |
2704 | * | |
9f54288d | 2705 | * Finally, we could use a virtio network switch in the kernel, ie. vhost. |
2e04ef76 | 2706 | :*/ |
8ca47e00 RR |
2707 | |
2708 | static u32 str2ip(const char *ipaddr) | |
2709 | { | |
dec6a2be | 2710 | unsigned int b[4]; |
8ca47e00 | 2711 | |
dec6a2be MM |
2712 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
2713 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
2714 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
2715 | } | |
2716 | ||
2717 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
2718 | { | |
2719 | unsigned int m[6]; | |
2720 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
2721 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
2722 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
2723 | mac[0] = m[0]; | |
2724 | mac[1] = m[1]; | |
2725 | mac[2] = m[2]; | |
2726 | mac[3] = m[3]; | |
2727 | mac[4] = m[4]; | |
2728 | mac[5] = m[5]; | |
8ca47e00 RR |
2729 | } |
2730 | ||
2e04ef76 RR |
2731 | /* |
2732 | * This code is "adapted" from libbridge: it attaches the Host end of the | |
dde79789 RR |
2733 | * network device to the bridge device specified by the command line. |
2734 | * | |
2735 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
2e04ef76 RR |
2736 | * dislike bridging), and I just try not to break it. |
2737 | */ | |
8ca47e00 RR |
2738 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
2739 | { | |
2740 | int ifidx; | |
2741 | struct ifreq ifr; | |
2742 | ||
2743 | if (!*br_name) | |
2744 | errx(1, "must specify bridge name"); | |
2745 | ||
2746 | ifidx = if_nametoindex(if_name); | |
2747 | if (!ifidx) | |
2748 | errx(1, "interface %s does not exist!", if_name); | |
2749 | ||
2750 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 2751 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
2752 | ifr.ifr_ifindex = ifidx; |
2753 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
2754 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
2755 | } | |
2756 | ||
2e04ef76 RR |
2757 | /* |
2758 | * This sets up the Host end of the network device with an IP address, brings | |
dde79789 | 2759 | * it up so packets will flow, the copies the MAC address into the hwaddr |
2e04ef76 RR |
2760 | * pointer. |
2761 | */ | |
dec6a2be | 2762 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
2763 | { |
2764 | struct ifreq ifr; | |
f846619e | 2765 | struct sockaddr_in sin; |
8ca47e00 RR |
2766 | |
2767 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
2768 | strcpy(ifr.ifr_name, tapif); |
2769 | ||
2770 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
f846619e RR |
2771 | sin.sin_family = AF_INET; |
2772 | sin.sin_addr.s_addr = htonl(ipaddr); | |
2773 | memcpy(&ifr.ifr_addr, &sin, sizeof(sin)); | |
8ca47e00 | 2774 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
dec6a2be | 2775 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
2776 | ifr.ifr_flags = IFF_UP; |
2777 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
2778 | err(1, "Bringing interface %s up", tapif); |
2779 | } | |
2780 | ||
dec6a2be | 2781 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 2782 | { |
8ca47e00 | 2783 | struct ifreq ifr; |
bf6d4034 | 2784 | int vnet_hdr_sz; |
dec6a2be MM |
2785 | int netfd; |
2786 | ||
2787 | /* Start with this zeroed. Messy but sure. */ | |
2788 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 2789 | |
2e04ef76 RR |
2790 | /* |
2791 | * We open the /dev/net/tun device and tell it we want a tap device. A | |
dde79789 RR |
2792 | * tap device is like a tun device, only somehow different. To tell |
2793 | * the truth, I completely blundered my way through this code, but it | |
2e04ef76 RR |
2794 | * works now! |
2795 | */ | |
8ca47e00 | 2796 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 2797 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
2798 | strcpy(ifr.ifr_name, "tap%d"); |
2799 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
2800 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 2801 | |
398f187d RR |
2802 | if (ioctl(netfd, TUNSETOFFLOAD, |
2803 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
2804 | err(1, "Could not set features for tun device"); | |
2805 | ||
2e04ef76 RR |
2806 | /* |
2807 | * We don't need checksums calculated for packets coming in this | |
2808 | * device: trust us! | |
2809 | */ | |
8ca47e00 RR |
2810 | ioctl(netfd, TUNSETNOCSUM, 1); |
2811 | ||
bf6d4034 RR |
2812 | /* |
2813 | * In virtio before 1.0 (aka legacy virtio), we added a 16-bit | |
2814 | * field at the end of the network header iff | |
2815 | * VIRTIO_NET_F_MRG_RXBUF was negotiated. For virtio 1.0, | |
2816 | * that became the norm, but we need to tell the tun device | |
2817 | * about our expanded header (which is called | |
2818 | * virtio_net_hdr_mrg_rxbuf in the legacy system). | |
2819 | */ | |
206ad06b | 2820 | vnet_hdr_sz = sizeof(struct virtio_net_hdr_v1); |
bf6d4034 RR |
2821 | if (ioctl(netfd, TUNSETVNETHDRSZ, &vnet_hdr_sz) != 0) |
2822 | err(1, "Setting tun header size to %u", vnet_hdr_sz); | |
2823 | ||
dec6a2be MM |
2824 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
2825 | return netfd; | |
2826 | } | |
2827 | ||
2e04ef76 RR |
2828 | /*L:195 |
2829 | * Our network is a Host<->Guest network. This can either use bridging or | |
dec6a2be MM |
2830 | * routing, but the principle is the same: it uses the "tun" device to inject |
2831 | * packets into the Host as if they came in from a normal network card. We | |
2e04ef76 RR |
2832 | * just shunt packets between the Guest and the tun device. |
2833 | */ | |
dec6a2be MM |
2834 | static void setup_tun_net(char *arg) |
2835 | { | |
2836 | struct device *dev; | |
659a0e66 RR |
2837 | struct net_info *net_info = malloc(sizeof(*net_info)); |
2838 | int ipfd; | |
dec6a2be MM |
2839 | u32 ip = INADDR_ANY; |
2840 | bool bridging = false; | |
2841 | char tapif[IFNAMSIZ], *p; | |
2842 | struct virtio_net_config conf; | |
2843 | ||
659a0e66 | 2844 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 2845 | |
17cbca2b | 2846 | /* First we create a new network device. */ |
bf6d4034 | 2847 | dev = new_pci_device("net", VIRTIO_ID_NET, 0x02, 0x00); |
659a0e66 | 2848 | dev->priv = net_info; |
dde79789 | 2849 | |
2e04ef76 | 2850 | /* Network devices need a recv and a send queue, just like console. */ |
17c56d6d RR |
2851 | add_pci_virtqueue(dev, net_input, "rx"); |
2852 | add_pci_virtqueue(dev, net_output, "tx"); | |
8ca47e00 | 2853 | |
2e04ef76 RR |
2854 | /* |
2855 | * We need a socket to perform the magic network ioctls to bring up the | |
2856 | * tap interface, connect to the bridge etc. Any socket will do! | |
2857 | */ | |
8ca47e00 RR |
2858 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
2859 | if (ipfd < 0) | |
2860 | err(1, "opening IP socket"); | |
2861 | ||
dde79789 | 2862 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 2863 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
2864 | arg += strlen(BRIDGE_PFX); |
2865 | bridging = true; | |
2866 | } | |
2867 | ||
2868 | /* A mac address may follow the bridge name or IP address */ | |
2869 | p = strchr(arg, ':'); | |
2870 | if (p) { | |
2871 | str2mac(p+1, conf.mac); | |
bf6d4034 | 2872 | add_pci_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 2873 | *p = '\0'; |
dec6a2be MM |
2874 | } |
2875 | ||
2876 | /* arg is now either an IP address or a bridge name */ | |
2877 | if (bridging) | |
2878 | add_to_bridge(ipfd, tapif, arg); | |
2879 | else | |
8ca47e00 RR |
2880 | ip = str2ip(arg); |
2881 | ||
dec6a2be MM |
2882 | /* Set up the tun device. */ |
2883 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 2884 | |
398f187d | 2885 | /* Expect Guest to handle everything except UFO */ |
bf6d4034 RR |
2886 | add_pci_feature(dev, VIRTIO_NET_F_CSUM); |
2887 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
2888 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO4); | |
2889 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
2890 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
2891 | add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
2892 | add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
2893 | add_pci_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e | 2894 | /* We handle indirect ring entries */ |
bf6d4034 RR |
2895 | add_pci_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); |
2896 | set_device_config(dev, &conf, sizeof(conf)); | |
8ca47e00 | 2897 | |
a586d4f6 | 2898 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
2899 | close(ipfd); |
2900 | ||
dec6a2be MM |
2901 | if (bridging) |
2902 | verbose("device %u: tun %s attached to bridge: %s\n", | |
2903 | devices.device_num, tapif, arg); | |
2904 | else | |
2905 | verbose("device %u: tun %s: %s\n", | |
2906 | devices.device_num, tapif, arg); | |
8ca47e00 | 2907 | } |
a91d74a3 | 2908 | /*:*/ |
17cbca2b | 2909 | |
e1e72965 | 2910 | /* This hangs off device->priv. */ |
1842f23c | 2911 | struct vblk_info { |
17cbca2b RR |
2912 | /* The size of the file. */ |
2913 | off64_t len; | |
2914 | ||
2915 | /* The file descriptor for the file. */ | |
2916 | int fd; | |
2917 | ||
17cbca2b RR |
2918 | }; |
2919 | ||
e1e72965 RR |
2920 | /*L:210 |
2921 | * The Disk | |
2922 | * | |
a91d74a3 RR |
2923 | * The disk only has one virtqueue, so it only has one thread. It is really |
2924 | * simple: the Guest asks for a block number and we read or write that position | |
2925 | * in the file. | |
2926 | * | |
2927 | * Before we serviced each virtqueue in a separate thread, that was unacceptably | |
2928 | * slow: the Guest waits until the read is finished before running anything | |
2929 | * else, even if it could have been doing useful work. | |
2930 | * | |
2931 | * We could have used async I/O, except it's reputed to suck so hard that | |
2932 | * characters actually go missing from your code when you try to use it. | |
e1e72965 | 2933 | */ |
659a0e66 | 2934 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 2935 | { |
659a0e66 | 2936 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b | 2937 | unsigned int head, out_num, in_num, wlen; |
c0316a94 | 2938 | int ret, i; |
cb38fa23 | 2939 | u8 *in; |
c0316a94 | 2940 | struct virtio_blk_outhdr out; |
659a0e66 | 2941 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
2942 | off64_t off; |
2943 | ||
a91d74a3 RR |
2944 | /* |
2945 | * Get the next request, where we normally wait. It triggers the | |
2946 | * interrupt to acknowledge previously serviced requests (if any). | |
2947 | */ | |
659a0e66 | 2948 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
17cbca2b | 2949 | |
c0316a94 | 2950 | /* Copy the output header from the front of the iov (adjusts iov) */ |
1e1c17a7 | 2951 | iov_consume(vq->dev, iov, out_num, &out, sizeof(out)); |
c0316a94 RR |
2952 | |
2953 | /* Find and trim end of iov input array, for our status byte. */ | |
2954 | in = NULL; | |
2955 | for (i = out_num + in_num - 1; i >= out_num; i--) { | |
2956 | if (iov[i].iov_len > 0) { | |
2957 | in = iov[i].iov_base + iov[i].iov_len - 1; | |
2958 | iov[i].iov_len--; | |
2959 | break; | |
2960 | } | |
2961 | } | |
2962 | if (!in) | |
1e1c17a7 | 2963 | bad_driver_vq(vq, "Bad virtblk cmd with no room for status"); |
17cbca2b | 2964 | |
a91d74a3 RR |
2965 | /* |
2966 | * For historical reasons, block operations are expressed in 512 byte | |
2967 | * "sectors". | |
2968 | */ | |
c0316a94 | 2969 | off = out.sector * 512; |
17cbca2b | 2970 | |
50516547 | 2971 | if (out.type & VIRTIO_BLK_T_OUT) { |
2e04ef76 RR |
2972 | /* |
2973 | * Write | |
2974 | * | |
2975 | * Move to the right location in the block file. This can fail | |
2976 | * if they try to write past end. | |
2977 | */ | |
17cbca2b | 2978 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 2979 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 2980 | |
c0316a94 RR |
2981 | ret = writev(vblk->fd, iov, out_num); |
2982 | verbose("WRITE to sector %llu: %i\n", out.sector, ret); | |
17cbca2b | 2983 | |
2e04ef76 RR |
2984 | /* |
2985 | * Grr... Now we know how long the descriptor they sent was, we | |
17cbca2b | 2986 | * make sure they didn't try to write over the end of the block |
2e04ef76 RR |
2987 | * file (possibly extending it). |
2988 | */ | |
17cbca2b RR |
2989 | if (ret > 0 && off + ret > vblk->len) { |
2990 | /* Trim it back to the correct length */ | |
2991 | ftruncate64(vblk->fd, vblk->len); | |
2992 | /* Die, bad Guest, die. */ | |
1e1c17a7 | 2993 | bad_driver_vq(vq, "Write past end %llu+%u", off, ret); |
17cbca2b | 2994 | } |
7bc9fdda TH |
2995 | |
2996 | wlen = sizeof(*in); | |
2997 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
c0316a94 | 2998 | } else if (out.type & VIRTIO_BLK_T_FLUSH) { |
7bc9fdda TH |
2999 | /* Flush */ |
3000 | ret = fdatasync(vblk->fd); | |
3001 | verbose("FLUSH fdatasync: %i\n", ret); | |
1200e646 | 3002 | wlen = sizeof(*in); |
cb38fa23 | 3003 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b | 3004 | } else { |
2e04ef76 RR |
3005 | /* |
3006 | * Read | |
3007 | * | |
3008 | * Move to the right location in the block file. This can fail | |
3009 | * if they try to read past end. | |
3010 | */ | |
17cbca2b | 3011 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 3012 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 3013 | |
c0316a94 | 3014 | ret = readv(vblk->fd, iov + out_num, in_num); |
17cbca2b | 3015 | if (ret >= 0) { |
1200e646 | 3016 | wlen = sizeof(*in) + ret; |
cb38fa23 | 3017 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 3018 | } else { |
1200e646 | 3019 | wlen = sizeof(*in); |
cb38fa23 | 3020 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
3021 | } |
3022 | } | |
3023 | ||
a91d74a3 | 3024 | /* Finished that request. */ |
38bc2b8c | 3025 | add_used(vq, head, wlen); |
17cbca2b RR |
3026 | } |
3027 | ||
e1e72965 | 3028 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
3029 | static void setup_block_file(const char *filename) |
3030 | { | |
17cbca2b RR |
3031 | struct device *dev; |
3032 | struct vblk_info *vblk; | |
a586d4f6 | 3033 | struct virtio_blk_config conf; |
17cbca2b | 3034 | |
50516547 RR |
3035 | /* Create the device. */ |
3036 | dev = new_pci_device("block", VIRTIO_ID_BLOCK, 0x01, 0x80); | |
17cbca2b | 3037 | |
e1e72965 | 3038 | /* The device has one virtqueue, where the Guest places requests. */ |
17c56d6d | 3039 | add_pci_virtqueue(dev, blk_request, "request"); |
17cbca2b RR |
3040 | |
3041 | /* Allocate the room for our own bookkeeping */ | |
3042 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
3043 | ||
3044 | /* First we open the file and store the length. */ | |
3045 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
3046 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
3047 | ||
3048 | /* Tell Guest how many sectors this device has. */ | |
a586d4f6 | 3049 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b | 3050 | |
2e04ef76 RR |
3051 | /* |
3052 | * Tell Guest not to put in too many descriptors at once: two are used | |
3053 | * for the in and out elements. | |
3054 | */ | |
50516547 | 3055 | add_pci_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
a586d4f6 RR |
3056 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); |
3057 | ||
50516547 | 3058 | set_device_config(dev, &conf, sizeof(struct virtio_blk_config)); |
17cbca2b | 3059 | |
17cbca2b | 3060 | verbose("device %u: virtblock %llu sectors\n", |
50516547 | 3061 | devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 3062 | } |
28fd6d7f | 3063 | |
2e04ef76 | 3064 | /*L:211 |
a454bb36 | 3065 | * Our random number generator device reads from /dev/urandom into the Guest's |
28fd6d7f | 3066 | * input buffers. The usual case is that the Guest doesn't want random numbers |
a454bb36 | 3067 | * and so has no buffers although /dev/urandom is still readable, whereas |
28fd6d7f RR |
3068 | * console is the reverse. |
3069 | * | |
2e04ef76 RR |
3070 | * The same logic applies, however. |
3071 | */ | |
3072 | struct rng_info { | |
3073 | int rfd; | |
3074 | }; | |
3075 | ||
659a0e66 | 3076 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
3077 | { |
3078 | int len; | |
3079 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
3080 | struct rng_info *rng_info = vq->dev->priv; |
3081 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
3082 | |
3083 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 3084 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f | 3085 | if (out_num) |
1e1c17a7 | 3086 | bad_driver_vq(vq, "Output buffers in rng?"); |
28fd6d7f | 3087 | |
2e04ef76 | 3088 | /* |
a91d74a3 RR |
3089 | * Just like the console write, we loop to cover the whole iovec. |
3090 | * In this case, short reads actually happen quite a bit. | |
2e04ef76 | 3091 | */ |
28fd6d7f | 3092 | while (!iov_empty(iov, in_num)) { |
659a0e66 | 3093 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f | 3094 | if (len <= 0) |
a454bb36 | 3095 | err(1, "Read from /dev/urandom gave %i", len); |
1e1c17a7 | 3096 | iov_consume(vq->dev, iov, in_num, NULL, len); |
28fd6d7f RR |
3097 | totlen += len; |
3098 | } | |
3099 | ||
3100 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 3101 | add_used(vq, head, totlen); |
28fd6d7f RR |
3102 | } |
3103 | ||
2e04ef76 RR |
3104 | /*L:199 |
3105 | * This creates a "hardware" random number device for the Guest. | |
3106 | */ | |
28fd6d7f RR |
3107 | static void setup_rng(void) |
3108 | { | |
3109 | struct device *dev; | |
659a0e66 | 3110 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 3111 | |
a454bb36 RR |
3112 | /* Our device's private info simply contains the /dev/urandom fd. */ |
3113 | rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY); | |
28fd6d7f | 3114 | |
2e04ef76 | 3115 | /* Create the new device. */ |
0d5b5d39 | 3116 | dev = new_pci_device("rng", VIRTIO_ID_RNG, 0xff, 0); |
659a0e66 | 3117 | dev->priv = rng_info; |
28fd6d7f RR |
3118 | |
3119 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
17c56d6d | 3120 | add_pci_virtqueue(dev, rng_input, "input"); |
28fd6d7f | 3121 | |
0d5b5d39 RR |
3122 | /* We don't have any configuration space */ |
3123 | no_device_config(dev); | |
3124 | ||
3125 | verbose("device %u: rng\n", devices.device_num); | |
28fd6d7f | 3126 | } |
a6bd8e13 | 3127 | /* That's the end of device setup. */ |
ec04b13f | 3128 | |
a6bd8e13 | 3129 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
3130 | static void __attribute__((noreturn)) restart_guest(void) |
3131 | { | |
3132 | unsigned int i; | |
3133 | ||
2e04ef76 RR |
3134 | /* |
3135 | * Since we don't track all open fds, we simply close everything beyond | |
3136 | * stderr. | |
3137 | */ | |
ec04b13f BR |
3138 | for (i = 3; i < FD_SETSIZE; i++) |
3139 | close(i); | |
8c79873d | 3140 | |
659a0e66 RR |
3141 | /* Reset all the devices (kills all threads). */ |
3142 | cleanup_devices(); | |
3143 | ||
ec04b13f BR |
3144 | execv(main_args[0], main_args); |
3145 | err(1, "Could not exec %s", main_args[0]); | |
3146 | } | |
8ca47e00 | 3147 | |
2e04ef76 RR |
3148 | /*L:220 |
3149 | * Finally we reach the core of the Launcher which runs the Guest, serves | |
3150 | * its input and output, and finally, lays it to rest. | |
3151 | */ | |
56739c80 | 3152 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
3153 | { |
3154 | for (;;) { | |
69a09dc1 | 3155 | struct lguest_pending notify; |
8ca47e00 RR |
3156 | int readval; |
3157 | ||
3158 | /* We read from the /dev/lguest device to run the Guest. */ | |
69a09dc1 | 3159 | readval = pread(lguest_fd, ¬ify, sizeof(notify), cpu_id); |
69a09dc1 | 3160 | if (readval == sizeof(notify)) { |
00f8d546 | 3161 | if (notify.trap == 13) { |
c565650b RR |
3162 | verbose("Emulating instruction at %#x\n", |
3163 | getreg(eip)); | |
3164 | emulate_insn(notify.insn); | |
6a54f9ab RR |
3165 | } else if (notify.trap == 14) { |
3166 | verbose("Emulating MMIO at %#x\n", | |
3167 | getreg(eip)); | |
3168 | emulate_mmio(notify.addr, notify.insn); | |
69a09dc1 RR |
3169 | } else |
3170 | errx(1, "Unknown trap %i addr %#08x\n", | |
3171 | notify.trap, notify.addr); | |
dde79789 | 3172 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
3173 | } else if (errno == ENOENT) { |
3174 | char reason[1024] = { 0 }; | |
e3283fa0 | 3175 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 3176 | errx(1, "%s", reason); |
ec04b13f BR |
3177 | /* ERESTART means that we need to reboot the guest */ |
3178 | } else if (errno == ERESTART) { | |
3179 | restart_guest(); | |
659a0e66 RR |
3180 | /* Anything else means a bug or incompatible change. */ |
3181 | } else | |
8ca47e00 | 3182 | err(1, "Running guest failed"); |
8ca47e00 RR |
3183 | } |
3184 | } | |
a6bd8e13 | 3185 | /*L:240 |
e1e72965 RR |
3186 | * This is the end of the Launcher. The good news: we are over halfway |
3187 | * through! The bad news: the most fiendish part of the code still lies ahead | |
3188 | * of us. | |
dde79789 | 3189 | * |
e1e72965 RR |
3190 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
3191 | * "make Host". | |
2e04ef76 | 3192 | :*/ |
8ca47e00 RR |
3193 | |
3194 | static struct option opts[] = { | |
3195 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
3196 | { "tunnet", 1, NULL, 't' }, |
3197 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 3198 | { "rng", 0, NULL, 'r' }, |
8ca47e00 | 3199 | { "initrd", 1, NULL, 'i' }, |
8aeb36e8 PS |
3200 | { "username", 1, NULL, 'u' }, |
3201 | { "chroot", 1, NULL, 'c' }, | |
8ca47e00 RR |
3202 | { NULL }, |
3203 | }; | |
3204 | static void usage(void) | |
3205 | { | |
3206 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 3207 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
3208 | "|--block=<filename>|--initrd=<filename>]...\n" |
3209 | "<mem-in-mb> vmlinux [args...]"); | |
3210 | } | |
3211 | ||
3c6b5bfa | 3212 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
3213 | int main(int argc, char *argv[]) |
3214 | { | |
2e04ef76 | 3215 | /* Memory, code startpoint and size of the (optional) initrd. */ |
58a24566 | 3216 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
3217 | /* Two temporaries. */ |
3218 | int i, c; | |
3c6b5bfa | 3219 | /* The boot information for the Guest. */ |
43d33b21 | 3220 | struct boot_params *boot; |
dde79789 | 3221 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
3222 | const char *initrd_name = NULL; |
3223 | ||
8aeb36e8 PS |
3224 | /* Password structure for initgroups/setres[gu]id */ |
3225 | struct passwd *user_details = NULL; | |
3226 | ||
3227 | /* Directory to chroot to */ | |
3228 | char *chroot_path = NULL; | |
3229 | ||
ec04b13f BR |
3230 | /* Save the args: we "reboot" by execing ourselves again. */ |
3231 | main_args = argv; | |
ec04b13f | 3232 | |
2e04ef76 | 3233 | /* |
d9028eda RR |
3234 | * First we initialize the device list. We remember next interrupt |
3235 | * number to use for devices (1: remember that 0 is used by the timer). | |
2e04ef76 | 3236 | */ |
17cbca2b | 3237 | devices.next_irq = 1; |
8ca47e00 | 3238 | |
a91d74a3 | 3239 | /* We're CPU 0. In fact, that's the only CPU possible right now. */ |
e3283fa0 | 3240 | cpu_id = 0; |
a91d74a3 | 3241 | |
2e04ef76 RR |
3242 | /* |
3243 | * We need to know how much memory so we can set up the device | |
dde79789 RR |
3244 | * descriptor and memory pages for the devices as we parse the command |
3245 | * line. So we quickly look through the arguments to find the amount | |
2e04ef76 RR |
3246 | * of memory now. |
3247 | */ | |
6570c459 RR |
3248 | for (i = 1; i < argc; i++) { |
3249 | if (argv[i][0] != '-') { | |
3c6b5bfa | 3250 | mem = atoi(argv[i]) * 1024 * 1024; |
2e04ef76 RR |
3251 | /* |
3252 | * We start by mapping anonymous pages over all of | |
3c6b5bfa RR |
3253 | * guest-physical memory range. This fills it with 0, |
3254 | * and ensures that the Guest won't be killed when it | |
2e04ef76 RR |
3255 | * tries to access it. |
3256 | */ | |
3c6b5bfa RR |
3257 | guest_base = map_zeroed_pages(mem / getpagesize() |
3258 | + DEVICE_PAGES); | |
3259 | guest_limit = mem; | |
0a6bcc18 | 3260 | guest_max = guest_mmio = mem + DEVICE_PAGES*getpagesize(); |
6570c459 RR |
3261 | break; |
3262 | } | |
3263 | } | |
dde79789 | 3264 | |
713e3f72 RR |
3265 | /* We always have a console device, and it's always device 1. */ |
3266 | setup_console(); | |
3267 | ||
dde79789 | 3268 | /* The options are fairly straight-forward */ |
8ca47e00 RR |
3269 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
3270 | switch (c) { | |
3271 | case 'v': | |
3272 | verbose = true; | |
3273 | break; | |
8ca47e00 | 3274 | case 't': |
17cbca2b | 3275 | setup_tun_net(optarg); |
8ca47e00 RR |
3276 | break; |
3277 | case 'b': | |
17cbca2b | 3278 | setup_block_file(optarg); |
8ca47e00 | 3279 | break; |
28fd6d7f RR |
3280 | case 'r': |
3281 | setup_rng(); | |
3282 | break; | |
8ca47e00 RR |
3283 | case 'i': |
3284 | initrd_name = optarg; | |
3285 | break; | |
8aeb36e8 PS |
3286 | case 'u': |
3287 | user_details = getpwnam(optarg); | |
3288 | if (!user_details) | |
3289 | err(1, "getpwnam failed, incorrect username?"); | |
3290 | break; | |
3291 | case 'c': | |
3292 | chroot_path = optarg; | |
3293 | break; | |
8ca47e00 RR |
3294 | default: |
3295 | warnx("Unknown argument %s", argv[optind]); | |
3296 | usage(); | |
3297 | } | |
3298 | } | |
2e04ef76 RR |
3299 | /* |
3300 | * After the other arguments we expect memory and kernel image name, | |
3301 | * followed by command line arguments for the kernel. | |
3302 | */ | |
8ca47e00 RR |
3303 | if (optind + 2 > argc) |
3304 | usage(); | |
3305 | ||
3c6b5bfa RR |
3306 | verbose("Guest base is at %p\n", guest_base); |
3307 | ||
8e709469 RR |
3308 | /* Initialize the (fake) PCI host bridge device. */ |
3309 | init_pci_host_bridge(); | |
3310 | ||
8ca47e00 | 3311 | /* Now we load the kernel */ |
47436aa4 | 3312 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 3313 | |
3c6b5bfa RR |
3314 | /* Boot information is stashed at physical address 0 */ |
3315 | boot = from_guest_phys(0); | |
3316 | ||
dde79789 | 3317 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
3318 | if (initrd_name) { |
3319 | initrd_size = load_initrd(initrd_name, mem); | |
2e04ef76 RR |
3320 | /* |
3321 | * These are the location in the Linux boot header where the | |
3322 | * start and size of the initrd are expected to be found. | |
3323 | */ | |
43d33b21 RR |
3324 | boot->hdr.ramdisk_image = mem - initrd_size; |
3325 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 3326 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 3327 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
3328 | } |
3329 | ||
2e04ef76 RR |
3330 | /* |
3331 | * The Linux boot header contains an "E820" memory map: ours is a | |
3332 | * simple, single region. | |
3333 | */ | |
43d33b21 RR |
3334 | boot->e820_entries = 1; |
3335 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
2e04ef76 RR |
3336 | /* |
3337 | * The boot header contains a command line pointer: we put the command | |
3338 | * line after the boot header. | |
3339 | */ | |
43d33b21 | 3340 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
e1e72965 | 3341 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 3342 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 3343 | |
e22a5398 RR |
3344 | /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */ |
3345 | boot->hdr.kernel_alignment = 0x1000000; | |
3346 | ||
814a0e5c | 3347 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 3348 | boot->hdr.version = 0x207; |
814a0e5c RR |
3349 | |
3350 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 3351 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 3352 | |
43d33b21 RR |
3353 | /* Tell the entry path not to try to reload segment registers. */ |
3354 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 3355 | |
9f54288d | 3356 | /* We tell the kernel to initialize the Guest. */ |
56739c80 | 3357 | tell_kernel(start); |
dde79789 | 3358 | |
a91d74a3 | 3359 | /* Ensure that we terminate if a device-servicing child dies. */ |
659a0e66 RR |
3360 | signal(SIGCHLD, kill_launcher); |
3361 | ||
3362 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
3363 | atexit(cleanup_devices); | |
8ca47e00 | 3364 | |
8aeb36e8 PS |
3365 | /* If requested, chroot to a directory */ |
3366 | if (chroot_path) { | |
3367 | if (chroot(chroot_path) != 0) | |
3368 | err(1, "chroot(\"%s\") failed", chroot_path); | |
3369 | ||
3370 | if (chdir("/") != 0) | |
3371 | err(1, "chdir(\"/\") failed"); | |
3372 | ||
3373 | verbose("chroot done\n"); | |
3374 | } | |
3375 | ||
3376 | /* If requested, drop privileges */ | |
3377 | if (user_details) { | |
3378 | uid_t u; | |
3379 | gid_t g; | |
3380 | ||
3381 | u = user_details->pw_uid; | |
3382 | g = user_details->pw_gid; | |
3383 | ||
3384 | if (initgroups(user_details->pw_name, g) != 0) | |
3385 | err(1, "initgroups failed"); | |
3386 | ||
3387 | if (setresgid(g, g, g) != 0) | |
3388 | err(1, "setresgid failed"); | |
3389 | ||
3390 | if (setresuid(u, u, u) != 0) | |
3391 | err(1, "setresuid failed"); | |
3392 | ||
3393 | verbose("Dropping privileges completed\n"); | |
3394 | } | |
3395 | ||
dde79789 | 3396 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 3397 | run_guest(); |
8ca47e00 | 3398 | } |
f56a384e RR |
3399 | /*:*/ |
3400 | ||
3401 | /*M:999 | |
3402 | * Mastery is done: you now know everything I do. | |
3403 | * | |
3404 | * But surely you have seen code, features and bugs in your wanderings which | |
3405 | * you now yearn to attack? That is the real game, and I look forward to you | |
3406 | * patching and forking lguest into the Your-Name-Here-visor. | |
3407 | * | |
3408 | * Farewell, and good coding! | |
3409 | * Rusty Russell. | |
3410 | */ |