MIPS: Yosemite, Emma: Fix off-by-two in arcs_cmdline buffer size check
[linux-2.6/linux-mips.git] / drivers / block / xen-blkback / xenbus.c
blob5fd2010f7d2bd96e1dc7b4c290fa71299d480d14
1 /* Xenbus code for blkif backend
2 Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
3 Copyright (C) 2005 XenSource Ltd
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
17 #include <stdarg.h>
18 #include <linux/module.h>
19 #include <linux/kthread.h>
20 #include <xen/events.h>
21 #include <xen/grant_table.h>
22 #include "common.h"
24 struct backend_info {
25 struct xenbus_device *dev;
26 struct xen_blkif *blkif;
27 struct xenbus_watch backend_watch;
28 unsigned major;
29 unsigned minor;
30 char *mode;
33 static struct kmem_cache *xen_blkif_cachep;
34 static void connect(struct backend_info *);
35 static int connect_ring(struct backend_info *);
36 static void backend_changed(struct xenbus_watch *, const char **,
37 unsigned int);
39 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
41 return be->dev;
44 static int blkback_name(struct xen_blkif *blkif, char *buf)
46 char *devpath, *devname;
47 struct xenbus_device *dev = blkif->be->dev;
49 devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
50 if (IS_ERR(devpath))
51 return PTR_ERR(devpath);
53 devname = strstr(devpath, "/dev/");
54 if (devname != NULL)
55 devname += strlen("/dev/");
56 else
57 devname = devpath;
59 snprintf(buf, TASK_COMM_LEN, "blkback.%d.%s", blkif->domid, devname);
60 kfree(devpath);
62 return 0;
65 static void xen_update_blkif_status(struct xen_blkif *blkif)
67 int err;
68 char name[TASK_COMM_LEN];
70 /* Not ready to connect? */
71 if (!blkif->irq || !blkif->vbd.bdev)
72 return;
74 /* Already connected? */
75 if (blkif->be->dev->state == XenbusStateConnected)
76 return;
78 /* Attempt to connect: exit if we fail to. */
79 connect(blkif->be);
80 if (blkif->be->dev->state != XenbusStateConnected)
81 return;
83 err = blkback_name(blkif, name);
84 if (err) {
85 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
86 return;
89 err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
90 if (err) {
91 xenbus_dev_error(blkif->be->dev, err, "block flush");
92 return;
94 invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
96 blkif->xenblkd = kthread_run(xen_blkif_schedule, blkif, name);
97 if (IS_ERR(blkif->xenblkd)) {
98 err = PTR_ERR(blkif->xenblkd);
99 blkif->xenblkd = NULL;
100 xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
104 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
106 struct xen_blkif *blkif;
108 blkif = kmem_cache_alloc(xen_blkif_cachep, GFP_KERNEL);
109 if (!blkif)
110 return ERR_PTR(-ENOMEM);
112 memset(blkif, 0, sizeof(*blkif));
113 blkif->domid = domid;
114 spin_lock_init(&blkif->blk_ring_lock);
115 atomic_set(&blkif->refcnt, 1);
116 init_waitqueue_head(&blkif->wq);
117 blkif->st_print = jiffies;
118 init_waitqueue_head(&blkif->waiting_to_free);
120 return blkif;
123 static int map_frontend_page(struct xen_blkif *blkif, unsigned long shared_page)
125 struct gnttab_map_grant_ref op;
127 gnttab_set_map_op(&op, (unsigned long)blkif->blk_ring_area->addr,
128 GNTMAP_host_map, shared_page, blkif->domid);
130 if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &op, 1))
131 BUG();
133 if (op.status) {
134 DPRINTK("Grant table operation failure !\n");
135 return op.status;
138 blkif->shmem_ref = shared_page;
139 blkif->shmem_handle = op.handle;
141 return 0;
144 static void unmap_frontend_page(struct xen_blkif *blkif)
146 struct gnttab_unmap_grant_ref op;
148 gnttab_set_unmap_op(&op, (unsigned long)blkif->blk_ring_area->addr,
149 GNTMAP_host_map, blkif->shmem_handle);
151 if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
152 BUG();
155 static int xen_blkif_map(struct xen_blkif *blkif, unsigned long shared_page,
156 unsigned int evtchn)
158 int err;
160 /* Already connected through? */
161 if (blkif->irq)
162 return 0;
164 blkif->blk_ring_area = alloc_vm_area(PAGE_SIZE);
165 if (!blkif->blk_ring_area)
166 return -ENOMEM;
168 err = map_frontend_page(blkif, shared_page);
169 if (err) {
170 free_vm_area(blkif->blk_ring_area);
171 return err;
174 switch (blkif->blk_protocol) {
175 case BLKIF_PROTOCOL_NATIVE:
177 struct blkif_sring *sring;
178 sring = (struct blkif_sring *)blkif->blk_ring_area->addr;
179 BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
180 break;
182 case BLKIF_PROTOCOL_X86_32:
184 struct blkif_x86_32_sring *sring_x86_32;
185 sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring_area->addr;
186 BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
187 break;
189 case BLKIF_PROTOCOL_X86_64:
191 struct blkif_x86_64_sring *sring_x86_64;
192 sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring_area->addr;
193 BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
194 break;
196 default:
197 BUG();
200 err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
201 xen_blkif_be_int, 0,
202 "blkif-backend", blkif);
203 if (err < 0) {
204 unmap_frontend_page(blkif);
205 free_vm_area(blkif->blk_ring_area);
206 blkif->blk_rings.common.sring = NULL;
207 return err;
209 blkif->irq = err;
211 return 0;
214 static void xen_blkif_disconnect(struct xen_blkif *blkif)
216 if (blkif->xenblkd) {
217 kthread_stop(blkif->xenblkd);
218 blkif->xenblkd = NULL;
221 atomic_dec(&blkif->refcnt);
222 wait_event(blkif->waiting_to_free, atomic_read(&blkif->refcnt) == 0);
223 atomic_inc(&blkif->refcnt);
225 if (blkif->irq) {
226 unbind_from_irqhandler(blkif->irq, blkif);
227 blkif->irq = 0;
230 if (blkif->blk_rings.common.sring) {
231 unmap_frontend_page(blkif);
232 free_vm_area(blkif->blk_ring_area);
233 blkif->blk_rings.common.sring = NULL;
237 void xen_blkif_free(struct xen_blkif *blkif)
239 if (!atomic_dec_and_test(&blkif->refcnt))
240 BUG();
241 kmem_cache_free(xen_blkif_cachep, blkif);
244 int __init xen_blkif_interface_init(void)
246 xen_blkif_cachep = kmem_cache_create("blkif_cache",
247 sizeof(struct xen_blkif),
248 0, 0, NULL);
249 if (!xen_blkif_cachep)
250 return -ENOMEM;
252 return 0;
256 * sysfs interface for VBD I/O requests
259 #define VBD_SHOW(name, format, args...) \
260 static ssize_t show_##name(struct device *_dev, \
261 struct device_attribute *attr, \
262 char *buf) \
264 struct xenbus_device *dev = to_xenbus_device(_dev); \
265 struct backend_info *be = dev_get_drvdata(&dev->dev); \
267 return sprintf(buf, format, ##args); \
269 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
271 VBD_SHOW(oo_req, "%d\n", be->blkif->st_oo_req);
272 VBD_SHOW(rd_req, "%d\n", be->blkif->st_rd_req);
273 VBD_SHOW(wr_req, "%d\n", be->blkif->st_wr_req);
274 VBD_SHOW(f_req, "%d\n", be->blkif->st_f_req);
275 VBD_SHOW(rd_sect, "%d\n", be->blkif->st_rd_sect);
276 VBD_SHOW(wr_sect, "%d\n", be->blkif->st_wr_sect);
278 static struct attribute *xen_vbdstat_attrs[] = {
279 &dev_attr_oo_req.attr,
280 &dev_attr_rd_req.attr,
281 &dev_attr_wr_req.attr,
282 &dev_attr_f_req.attr,
283 &dev_attr_rd_sect.attr,
284 &dev_attr_wr_sect.attr,
285 NULL
288 static struct attribute_group xen_vbdstat_group = {
289 .name = "statistics",
290 .attrs = xen_vbdstat_attrs,
293 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
294 VBD_SHOW(mode, "%s\n", be->mode);
296 int xenvbd_sysfs_addif(struct xenbus_device *dev)
298 int error;
300 error = device_create_file(&dev->dev, &dev_attr_physical_device);
301 if (error)
302 goto fail1;
304 error = device_create_file(&dev->dev, &dev_attr_mode);
305 if (error)
306 goto fail2;
308 error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
309 if (error)
310 goto fail3;
312 return 0;
314 fail3: sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
315 fail2: device_remove_file(&dev->dev, &dev_attr_mode);
316 fail1: device_remove_file(&dev->dev, &dev_attr_physical_device);
317 return error;
320 void xenvbd_sysfs_delif(struct xenbus_device *dev)
322 sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
323 device_remove_file(&dev->dev, &dev_attr_mode);
324 device_remove_file(&dev->dev, &dev_attr_physical_device);
328 static void xen_vbd_free(struct xen_vbd *vbd)
330 if (vbd->bdev)
331 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
332 vbd->bdev = NULL;
335 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
336 unsigned major, unsigned minor, int readonly,
337 int cdrom)
339 struct xen_vbd *vbd;
340 struct block_device *bdev;
341 struct request_queue *q;
343 vbd = &blkif->vbd;
344 vbd->handle = handle;
345 vbd->readonly = readonly;
346 vbd->type = 0;
348 vbd->pdevice = MKDEV(major, minor);
350 bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
351 FMODE_READ : FMODE_WRITE, NULL);
353 if (IS_ERR(bdev)) {
354 DPRINTK("xen_vbd_create: device %08x could not be opened.\n",
355 vbd->pdevice);
356 return -ENOENT;
359 vbd->bdev = bdev;
360 if (vbd->bdev->bd_disk == NULL) {
361 DPRINTK("xen_vbd_create: device %08x doesn't exist.\n",
362 vbd->pdevice);
363 xen_vbd_free(vbd);
364 return -ENOENT;
366 vbd->size = vbd_sz(vbd);
368 if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
369 vbd->type |= VDISK_CDROM;
370 if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
371 vbd->type |= VDISK_REMOVABLE;
373 q = bdev_get_queue(bdev);
374 if (q && q->flush_flags)
375 vbd->flush_support = true;
377 DPRINTK("Successful creation of handle=%04x (dom=%u)\n",
378 handle, blkif->domid);
379 return 0;
381 static int xen_blkbk_remove(struct xenbus_device *dev)
383 struct backend_info *be = dev_get_drvdata(&dev->dev);
385 DPRINTK("");
387 if (be->major || be->minor)
388 xenvbd_sysfs_delif(dev);
390 if (be->backend_watch.node) {
391 unregister_xenbus_watch(&be->backend_watch);
392 kfree(be->backend_watch.node);
393 be->backend_watch.node = NULL;
396 if (be->blkif) {
397 xen_blkif_disconnect(be->blkif);
398 xen_vbd_free(&be->blkif->vbd);
399 xen_blkif_free(be->blkif);
400 be->blkif = NULL;
403 kfree(be);
404 dev_set_drvdata(&dev->dev, NULL);
405 return 0;
408 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
409 struct backend_info *be, int state)
411 struct xenbus_device *dev = be->dev;
412 int err;
414 err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
415 "%d", state);
416 if (err)
417 xenbus_dev_fatal(dev, err, "writing feature-flush-cache");
419 return err;
423 * Entry point to this code when a new device is created. Allocate the basic
424 * structures, and watch the store waiting for the hotplug scripts to tell us
425 * the device's physical major and minor numbers. Switch to InitWait.
427 static int xen_blkbk_probe(struct xenbus_device *dev,
428 const struct xenbus_device_id *id)
430 int err;
431 struct backend_info *be = kzalloc(sizeof(struct backend_info),
432 GFP_KERNEL);
433 if (!be) {
434 xenbus_dev_fatal(dev, -ENOMEM,
435 "allocating backend structure");
436 return -ENOMEM;
438 be->dev = dev;
439 dev_set_drvdata(&dev->dev, be);
441 be->blkif = xen_blkif_alloc(dev->otherend_id);
442 if (IS_ERR(be->blkif)) {
443 err = PTR_ERR(be->blkif);
444 be->blkif = NULL;
445 xenbus_dev_fatal(dev, err, "creating block interface");
446 goto fail;
449 /* setup back pointer */
450 be->blkif->be = be;
452 err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
453 "%s/%s", dev->nodename, "physical-device");
454 if (err)
455 goto fail;
457 err = xenbus_switch_state(dev, XenbusStateInitWait);
458 if (err)
459 goto fail;
461 return 0;
463 fail:
464 DPRINTK("failed");
465 xen_blkbk_remove(dev);
466 return err;
471 * Callback received when the hotplug scripts have placed the physical-device
472 * node. Read it and the mode node, and create a vbd. If the frontend is
473 * ready, connect.
475 static void backend_changed(struct xenbus_watch *watch,
476 const char **vec, unsigned int len)
478 int err;
479 unsigned major;
480 unsigned minor;
481 struct backend_info *be
482 = container_of(watch, struct backend_info, backend_watch);
483 struct xenbus_device *dev = be->dev;
484 int cdrom = 0;
485 char *device_type;
487 DPRINTK("");
489 err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
490 &major, &minor);
491 if (XENBUS_EXIST_ERR(err)) {
493 * Since this watch will fire once immediately after it is
494 * registered, we expect this. Ignore it, and wait for the
495 * hotplug scripts.
497 return;
499 if (err != 2) {
500 xenbus_dev_fatal(dev, err, "reading physical-device");
501 return;
504 if ((be->major || be->minor) &&
505 ((be->major != major) || (be->minor != minor))) {
506 pr_warn(DRV_PFX "changing physical device (from %x:%x to %x:%x) not supported.\n",
507 be->major, be->minor, major, minor);
508 return;
511 be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
512 if (IS_ERR(be->mode)) {
513 err = PTR_ERR(be->mode);
514 be->mode = NULL;
515 xenbus_dev_fatal(dev, err, "reading mode");
516 return;
519 device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
520 if (!IS_ERR(device_type)) {
521 cdrom = strcmp(device_type, "cdrom") == 0;
522 kfree(device_type);
525 if (be->major == 0 && be->minor == 0) {
526 /* Front end dir is a number, which is used as the handle. */
528 char *p = strrchr(dev->otherend, '/') + 1;
529 long handle;
530 err = strict_strtoul(p, 0, &handle);
531 if (err)
532 return;
534 be->major = major;
535 be->minor = minor;
537 err = xen_vbd_create(be->blkif, handle, major, minor,
538 (NULL == strchr(be->mode, 'w')), cdrom);
539 if (err) {
540 be->major = 0;
541 be->minor = 0;
542 xenbus_dev_fatal(dev, err, "creating vbd structure");
543 return;
546 err = xenvbd_sysfs_addif(dev);
547 if (err) {
548 xen_vbd_free(&be->blkif->vbd);
549 be->major = 0;
550 be->minor = 0;
551 xenbus_dev_fatal(dev, err, "creating sysfs entries");
552 return;
555 /* We're potentially connected now */
556 xen_update_blkif_status(be->blkif);
562 * Callback received when the frontend's state changes.
564 static void frontend_changed(struct xenbus_device *dev,
565 enum xenbus_state frontend_state)
567 struct backend_info *be = dev_get_drvdata(&dev->dev);
568 int err;
570 DPRINTK("%s", xenbus_strstate(frontend_state));
572 switch (frontend_state) {
573 case XenbusStateInitialising:
574 if (dev->state == XenbusStateClosed) {
575 pr_info(DRV_PFX "%s: prepare for reconnect\n",
576 dev->nodename);
577 xenbus_switch_state(dev, XenbusStateInitWait);
579 break;
581 case XenbusStateInitialised:
582 case XenbusStateConnected:
584 * Ensure we connect even when two watches fire in
585 * close successsion and we miss the intermediate value
586 * of frontend_state.
588 if (dev->state == XenbusStateConnected)
589 break;
592 * Enforce precondition before potential leak point.
593 * xen_blkif_disconnect() is idempotent.
595 xen_blkif_disconnect(be->blkif);
597 err = connect_ring(be);
598 if (err)
599 break;
600 xen_update_blkif_status(be->blkif);
601 break;
603 case XenbusStateClosing:
604 xenbus_switch_state(dev, XenbusStateClosing);
605 break;
607 case XenbusStateClosed:
608 xen_blkif_disconnect(be->blkif);
609 xenbus_switch_state(dev, XenbusStateClosed);
610 if (xenbus_dev_is_online(dev))
611 break;
612 /* fall through if not online */
613 case XenbusStateUnknown:
614 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
615 device_unregister(&dev->dev);
616 break;
618 default:
619 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
620 frontend_state);
621 break;
626 /* ** Connection ** */
630 * Write the physical details regarding the block device to the store, and
631 * switch to Connected state.
633 static void connect(struct backend_info *be)
635 struct xenbus_transaction xbt;
636 int err;
637 struct xenbus_device *dev = be->dev;
639 DPRINTK("%s", dev->otherend);
641 /* Supply the information about the device the frontend needs */
642 again:
643 err = xenbus_transaction_start(&xbt);
644 if (err) {
645 xenbus_dev_fatal(dev, err, "starting transaction");
646 return;
649 err = xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
650 if (err)
651 goto abort;
653 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
654 (unsigned long long)vbd_sz(&be->blkif->vbd));
655 if (err) {
656 xenbus_dev_fatal(dev, err, "writing %s/sectors",
657 dev->nodename);
658 goto abort;
661 /* FIXME: use a typename instead */
662 err = xenbus_printf(xbt, dev->nodename, "info", "%u",
663 be->blkif->vbd.type |
664 (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
665 if (err) {
666 xenbus_dev_fatal(dev, err, "writing %s/info",
667 dev->nodename);
668 goto abort;
670 err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
671 (unsigned long)
672 bdev_logical_block_size(be->blkif->vbd.bdev));
673 if (err) {
674 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
675 dev->nodename);
676 goto abort;
679 err = xenbus_transaction_end(xbt, 0);
680 if (err == -EAGAIN)
681 goto again;
682 if (err)
683 xenbus_dev_fatal(dev, err, "ending transaction");
685 err = xenbus_switch_state(dev, XenbusStateConnected);
686 if (err)
687 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
688 dev->nodename);
690 return;
691 abort:
692 xenbus_transaction_end(xbt, 1);
696 static int connect_ring(struct backend_info *be)
698 struct xenbus_device *dev = be->dev;
699 unsigned long ring_ref;
700 unsigned int evtchn;
701 char protocol[64] = "";
702 int err;
704 DPRINTK("%s", dev->otherend);
706 err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
707 &ring_ref, "event-channel", "%u", &evtchn, NULL);
708 if (err) {
709 xenbus_dev_fatal(dev, err,
710 "reading %s/ring-ref and event-channel",
711 dev->otherend);
712 return err;
715 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
716 err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
717 "%63s", protocol, NULL);
718 if (err)
719 strcpy(protocol, "unspecified, assuming native");
720 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
721 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
722 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
723 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
724 else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
725 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
726 else {
727 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
728 return -1;
730 pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s)\n",
731 ring_ref, evtchn, be->blkif->blk_protocol, protocol);
733 /* Map the shared frame, irq etc. */
734 err = xen_blkif_map(be->blkif, ring_ref, evtchn);
735 if (err) {
736 xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
737 ring_ref, evtchn);
738 return err;
741 return 0;
745 /* ** Driver Registration ** */
748 static const struct xenbus_device_id xen_blkbk_ids[] = {
749 { "vbd" },
750 { "" }
754 static struct xenbus_driver xen_blkbk = {
755 .name = "vbd",
756 .owner = THIS_MODULE,
757 .ids = xen_blkbk_ids,
758 .probe = xen_blkbk_probe,
759 .remove = xen_blkbk_remove,
760 .otherend_changed = frontend_changed
764 int xen_blkif_xenbus_init(void)
766 return xenbus_register_backend(&xen_blkbk);