2 * Copyright (c) 2012 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
41 #include <sys/disklabel.h>
42 #include <sys/disklabel32.h>
43 #include <sys/disklabel64.h>
44 #include <sys/diskslice.h>
45 #include <sys/diskmbr.h>
47 #include <sys/malloc.h>
48 #include <sys/device.h>
49 #include <sys/devfs.h>
50 #include <sys/thread.h>
51 #include <sys/queue.h>
56 #include <vfs/hammer2/hammer2_disk.h>
60 #include <sys/mplock2.h>
61 #include <sys/msgport2.h>
62 #include <sys/thread2.h>
74 static MALLOC_DEFINE(M_DMSG_DISK, "dmsg_disk", "disk dmsg");
76 static int disk_iocom_reconnect(struct disk *dp, struct file *fp);
77 static int disk_rcvdmsg(kdmsg_msg_t *msg);
79 static void disk_blk_open(struct disk *dp, kdmsg_msg_t *msg);
80 static void disk_blk_read(struct disk *dp, kdmsg_msg_t *msg);
81 static void disk_blk_write(struct disk *dp, kdmsg_msg_t *msg);
82 static void disk_blk_flush(struct disk *dp, kdmsg_msg_t *msg);
83 static void disk_blk_freeblks(struct disk *dp, kdmsg_msg_t *msg);
84 static void diskiodone(struct bio *bio);
87 disk_iocom_init(struct disk *dp)
89 kdmsg_iocom_init(&dp->d_iocom, dp,
90 KDMSG_IOCOMF_AUTOCONN |
91 KDMSG_IOCOMF_AUTOSPAN |
92 KDMSG_IOCOMF_AUTOCIRC,
93 M_DMSG_DISK, disk_rcvdmsg);
97 disk_iocom_update(struct disk *dp)
102 disk_iocom_uninit(struct disk *dp)
104 kdmsg_iocom_uninit(&dp->d_iocom);
108 disk_iocom_ioctl(struct disk *dp, int cmd, void *data)
111 struct disk_ioc_recluster *recl;
117 fp = holdfp(curproc->p_fd, recl->fd, -1);
119 error = disk_iocom_reconnect(dp, fp);
133 disk_iocom_reconnect(struct disk *dp, struct file *fp)
137 ksnprintf(devname, sizeof(devname), "%s%d",
138 dev_dname(dp->d_rawdev), dkunit(dp->d_rawdev));
140 kdmsg_iocom_reconnect(&dp->d_iocom, fp, devname);
142 dp->d_iocom.auto_lnk_conn.pfs_type = DMSG_PFSTYPE_SERVER;
143 dp->d_iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
144 dp->d_iocom.auto_lnk_conn.peer_type = DMSG_PEER_BLOCK;
145 dp->d_iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_BLOCK;
146 dp->d_iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
147 ksnprintf(dp->d_iocom.auto_lnk_conn.cl_label,
148 sizeof(dp->d_iocom.auto_lnk_conn.cl_label),
149 "%s/%s", hostname, devname);
150 if (dp->d_info.d_serialno) {
151 ksnprintf(dp->d_iocom.auto_lnk_conn.fs_label,
152 sizeof(dp->d_iocom.auto_lnk_conn.fs_label),
153 "%s", dp->d_info.d_serialno);
156 dp->d_iocom.auto_lnk_span.pfs_type = DMSG_PFSTYPE_SERVER;
157 dp->d_iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
158 dp->d_iocom.auto_lnk_span.peer_type = DMSG_PEER_BLOCK;
159 dp->d_iocom.auto_lnk_span.media.block.bytes =
160 dp->d_info.d_media_size;
161 dp->d_iocom.auto_lnk_span.media.block.blksize =
162 dp->d_info.d_media_blksize;
163 ksnprintf(dp->d_iocom.auto_lnk_span.cl_label,
164 sizeof(dp->d_iocom.auto_lnk_span.cl_label),
165 "%s/%s", hostname, devname);
166 if (dp->d_info.d_serialno) {
167 ksnprintf(dp->d_iocom.auto_lnk_span.fs_label,
168 sizeof(dp->d_iocom.auto_lnk_span.fs_label),
169 "%s", dp->d_info.d_serialno);
172 kdmsg_iocom_autoinitiate(&dp->d_iocom, NULL);
178 disk_rcvdmsg(kdmsg_msg_t *msg)
180 struct disk *dp = msg->iocom->handle;
183 * Handle debug messages (these might not be in transactions)
185 switch(msg->any.head.cmd & DMSGF_CMDSWMASK) {
188 * Execute shell command (not supported atm)
190 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
192 case DMSG_DBG_SHELL | DMSGF_REPLY:
194 msg->aux_data[msg->aux_size - 1] = 0;
195 kprintf("diskiocom: DEBUGMSG: %s\n", msg->aux_data);
201 * All remaining messages must be in a transaction
203 * NOTE! We are switching on the first message's command. The
204 * actual message command within the transaction may be
205 * different (if streaming within a transaction).
207 if (msg->state == NULL) {
208 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
212 switch(msg->state->rxcmd & DMSGF_CMDSWMASK) {
215 disk_blk_open(dp, msg);
218 disk_blk_read(dp, msg);
221 disk_blk_write(dp, msg);
224 disk_blk_flush(dp, msg);
226 case DMSG_BLK_FREEBLKS:
227 disk_blk_freeblks(dp, msg);
230 if ((msg->any.head.cmd & DMSGF_REPLY) == 0) {
231 if (msg->any.head.cmd & DMSGF_DELETE)
232 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
234 kdmsg_msg_result(msg, DMSG_ERR_NOSUPP);
243 disk_blk_open(struct disk *dp, kdmsg_msg_t *msg)
245 struct dios_open *openst;
246 int error = DMSG_ERR_NOSUPP;
249 openst = msg->state->any.any;
250 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_OPEN) {
251 if (openst == NULL) {
252 openst = kmalloc(sizeof(*openst), M_DEVBUF,
254 msg->state->any.any = openst;
257 if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
259 if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
261 error = dev_dopen(dp->d_rawdev, fflags, S_IFCHR, proc0.p_ucred, NULL);
265 if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
267 if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
271 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_CLOSE &&
274 if ((msg->any.blk_open.modes & DMSG_BLKOPEN_RD) &&
278 if ((msg->any.blk_open.modes & DMSG_BLKOPEN_WR) &&
282 error = dev_dclose(dp->d_rawdev, fflags, S_IFCHR, NULL);
286 if (msg->any.blk_open.modes & DMSG_BLKOPEN_RD)
288 if (msg->any.blk_open.modes & DMSG_BLKOPEN_WR)
292 if (msg->any.head.cmd & DMSGF_DELETE) {
294 while (openst->openrd && openst->openwr) {
297 dev_dclose(dp->d_rawdev, FREAD|FWRITE, S_IFCHR, NULL);
299 while (openst->openrd) {
301 dev_dclose(dp->d_rawdev, FREAD, S_IFCHR, NULL);
303 while (openst->openwr) {
305 dev_dclose(dp->d_rawdev, FWRITE, S_IFCHR, NULL);
307 kfree(openst, M_DEVBUF);
308 msg->state->any.any = NULL;
310 kdmsg_msg_reply(msg, error);
312 kdmsg_msg_result(msg, error);
318 disk_blk_read(struct disk *dp, kdmsg_msg_t *msg)
320 struct dios_io *iost;
323 int error = DMSG_ERR_NOSUPP;
327 * Only DMSG_BLK_READ commands imply read ops.
329 iost = msg->state->any.any;
330 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_READ) {
331 if (msg->any.blk_read.bytes < DEV_BSIZE ||
332 msg->any.blk_read.bytes > MAXPHYS) {
333 error = DMSG_ERR_PARAM;
337 iost = kmalloc(sizeof(*iost), M_DEVBUF,
339 msg->state->any.any = iost;
342 bp = geteblk(msg->any.blk_read.bytes);
344 bp->b_cmd = BUF_CMD_READ;
345 bp->b_bcount = msg->any.blk_read.bytes;
346 bp->b_resid = bp->b_bcount;
347 bio->bio_offset = msg->any.blk_read.offset;
348 bio->bio_caller_info1.ptr = msg->state;
349 bio->bio_done = diskiodone;
350 /* kdmsg_state_hold(msg->state); */
352 atomic_add_int(&iost->count, 1);
353 if (msg->any.head.cmd & DMSGF_DELETE)
356 dev_dstrategy(dp->d_rawdev, bio);
360 if (msg->any.head.cmd & DMSGF_DELETE) {
361 if (iost && iost->count == 0) {
362 kfree(iost, M_DEVBUF);
363 msg->state->any.any = NULL;
365 kdmsg_msg_reply(msg, error);
367 kdmsg_msg_result(msg, error);
374 disk_blk_write(struct disk *dp, kdmsg_msg_t *msg)
376 struct dios_io *iost;
379 int error = DMSG_ERR_NOSUPP;
383 * Only DMSG_BLK_WRITE commands imply read ops.
385 iost = msg->state->any.any;
386 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_WRITE) {
387 if (msg->any.blk_write.bytes < DEV_BSIZE ||
388 msg->any.blk_write.bytes > MAXPHYS) {
389 error = DMSG_ERR_PARAM;
393 iost = kmalloc(sizeof(*iost), M_DEVBUF,
395 msg->state->any.any = iost;
399 * Issue WRITE. Short data implies zeros. Try to optimize
400 * the buffer cache buffer for the case where we can just
401 * use the message's data pointer.
404 if (msg->aux_size >= msg->any.blk_write.bytes)
407 bp = geteblk(msg->any.blk_write.bytes);
409 bp->b_cmd = BUF_CMD_WRITE;
410 bp->b_bcount = msg->any.blk_write.bytes;
411 bp->b_resid = bp->b_bcount;
412 if (msg->aux_size >= msg->any.blk_write.bytes) {
413 bp->b_data = msg->aux_data;
415 bcopy(msg->aux_data, bp->b_data, msg->aux_size);
416 bzero(bp->b_data + msg->aux_size,
417 msg->any.blk_write.bytes - msg->aux_size);
419 bio->bio_offset = msg->any.blk_write.offset;
420 bio->bio_caller_info1.ptr = msg->state;
421 bio->bio_done = diskiodone;
422 /* kdmsg_state_hold(msg->state); */
424 atomic_add_int(&iost->count, 1);
425 if (msg->any.head.cmd & DMSGF_DELETE)
428 dev_dstrategy(dp->d_rawdev, bio);
432 if (msg->any.head.cmd & DMSGF_DELETE) {
433 if (iost && iost->count == 0) {
434 kfree(iost, M_DEVBUF);
435 msg->state->any.any = NULL;
437 kdmsg_msg_reply(msg, error);
439 kdmsg_msg_result(msg, error);
446 disk_blk_flush(struct disk *dp, kdmsg_msg_t *msg)
448 struct dios_io *iost;
451 int error = DMSG_ERR_NOSUPP;
455 * Only DMSG_BLK_FLUSH commands imply read ops.
457 iost = msg->state->any.any;
458 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_FLUSH) {
460 iost = kmalloc(sizeof(*iost), M_DEVBUF,
462 msg->state->any.any = iost;
467 bp->b_cmd = BUF_CMD_FLUSH;
468 bp->b_bcount = msg->any.blk_flush.bytes;
470 bio->bio_offset = msg->any.blk_flush.offset;
471 bio->bio_caller_info1.ptr = msg->state;
472 bio->bio_done = diskiodone;
473 /* kdmsg_state_hold(msg->state); */
475 atomic_add_int(&iost->count, 1);
476 if (msg->any.head.cmd & DMSGF_DELETE)
479 dev_dstrategy(dp->d_rawdev, bio);
482 if (msg->any.head.cmd & DMSGF_DELETE) {
483 if (iost && iost->count == 0) {
484 kfree(iost, M_DEVBUF);
485 msg->state->any.any = NULL;
487 kdmsg_msg_reply(msg, error);
489 kdmsg_msg_result(msg, error);
496 disk_blk_freeblks(struct disk *dp, kdmsg_msg_t *msg)
498 struct dios_io *iost;
501 int error = DMSG_ERR_NOSUPP;
505 * Only DMSG_BLK_FREEBLKS commands imply read ops.
507 iost = msg->state->any.any;
508 if ((msg->any.head.cmd & DMSGF_CMDSWMASK) == DMSG_BLK_FREEBLKS) {
510 iost = kmalloc(sizeof(*iost), M_DEVBUF,
512 msg->state->any.any = iost;
517 bp->b_cmd = BUF_CMD_FREEBLKS;
518 bp->b_bcount = msg->any.blk_freeblks.bytes;
520 bio->bio_offset = msg->any.blk_freeblks.offset;
521 bio->bio_caller_info1.ptr = msg->state;
522 bio->bio_done = diskiodone;
523 /* kdmsg_state_hold(msg->state); */
525 atomic_add_int(&iost->count, 1);
526 if (msg->any.head.cmd & DMSGF_DELETE)
529 dev_dstrategy(dp->d_rawdev, bio);
532 if (msg->any.head.cmd & DMSGF_DELETE) {
533 if (iost && iost->count == 0) {
534 kfree(iost, M_DEVBUF);
535 msg->state->any.any = NULL;
537 kdmsg_msg_reply(msg, error);
539 kdmsg_msg_result(msg, error);
546 diskiodone(struct bio *bio)
548 struct buf *bp = bio->bio_buf;
549 kdmsg_state_t *state = bio->bio_caller_info1.ptr;
551 struct dios_io *iost = state->any.any;
558 cmd = DMSG_LNK_ERROR;
564 cmd = DMSG_LNK_ERROR;
566 bytes = bp->b_bcount;
569 if (bp->b_flags & B_ERROR) {
577 case BUF_CMD_FREEBLKS:
578 if (bp->b_flags & B_ERROR)
584 panic("diskiodone: Unknown bio cmd = %d\n",
585 bio->bio_buf->b_cmd);
586 error = 0; /* avoid compiler warning */
587 break; /* NOT REACHED */
591 * Convert error to DMSG_ERR_* code.
597 * Convert LNK_ERROR or BLK_ERROR if non-zero resid. READS will
598 * have already converted cmd to BLK_ERROR and set up data to return.
600 if (resid && cmd == DMSG_LNK_ERROR)
601 cmd = DMSG_BLK_ERROR;
602 /* XXX txcmd is delayed so this won't work for streaming */
603 if ((state->txcmd & DMSGF_CREATE) == 0) /* assume serialized */
606 if (atomic_fetchadd_int(&iost->count, -1) == 1)
609 atomic_add_int(&iost->count, -1);
614 * Allocate a basic or extended reply. Be careful not to populate
615 * extended header fields unless we allocated an extended reply.
617 rmsg = kdmsg_msg_alloc_state(state, cmd, NULL, 0);
619 rmsg->aux_data = kmalloc(bytes, state->iocom->mmsg, M_INTWAIT);
620 rmsg->aux_size = bytes;
621 rmsg->flags |= KDMSG_FLAG_AUXALLOC;
622 bcopy(data, rmsg->aux_data, bytes);
624 rmsg->any.blk_error.head.error = error;
625 if ((cmd & DMSGF_BASECMDMASK) == DMSG_BLK_ERROR)
626 rmsg->any.blk_error.resid = resid;
627 bio->bio_caller_info1.ptr = NULL;
628 /* kdmsg_state_drop(state); */
629 kdmsg_msg_write(rmsg);
630 if (bp->b_flags & B_PAGING) {
631 relpbuf(bio->bio_buf, NULL);
633 bp->b_flags |= B_INVAL | B_AGE;