20060303

1 files changed, 1531 insertions, 0 deletions

diff --git a/os/port/devlogfs.c b/os/port/devlogfs.c
new file mode 100755
index 00000000..630d3eba
--- /dev/null
+++ b/os/port/devlogfs.c
@@ -0,0 +1,1531 @@
+#ifndef EMU
+#include "u.h"
+#include "../port/lib.h"
+#include "../port/error.h"
+#else
+#include	"error.h"
+#endif
+#include	"dat.h"
+#include	"fns.h"
+#include	"kernel.h"
+#include	"logfs.h"
+#include	"nandfs.h"
+
+#ifndef EMU
+#define Sleep sleep
+#define Wakeup wakeup
+#endif
+
+#ifndef offsetof
+#define offsetof(T,X) ((ulong)&(((T*)0)->X))
+#endif
+
+typedef struct Devlogfs Devlogfs;
+typedef struct DevlogfsSession DevlogfsSession;
+
+//#define CALLTRACE
+
+enum {
+	DEVLOGFSDEBUG = 0,
+	DEVLOGFSIODEBUG = 0,
+	DEVLOGFSBAD = 1,
+};
+
+enum {
+	Qdir,
+	Qctl,
+	Qusers,
+	Qdump,
+	Qfs,
+	Qfsboot,
+	Qend,
+};
+
+typedef enum DevlogfsServerState { Closed, BootOpen, NeedVersion, NeedAttach, Attached, Hungup } DevlogfsServerState;
+
+struct Devlogfs {
+	QLock qlock;
+	QLock	rlock;
+	QLock	wlock;
+	Ref ref;
+	int instance;
+	int trace;	/* (debugging) trace of read/write actions */
+	int nand;
+	char *name;
+	char *device;
+	char *filename[Qend - Qfs];
+	LogfsLowLevel *ll;
+	Chan *flash, *flashctl;
+	QLock bootqlock;
+	int logfstrace;
+	LogfsBoot *lb;
+	/* stuff for server */
+	ulong openflags;
+	Fcall in;
+	Fcall out;
+	int reading;
+	DevlogfsServerState state;
+	Rendez readrendez;
+	Rendez writerendez;
+	uint readcount;
+	ulong readbufsize;
+	uchar *readbuf;
+	uchar *readp;
+	LogfsServer *server;
+	Devlogfs *next;
+};
+
+#define MAXMSIZE 8192
+
+static struct {
+	RWlock rwlock;		/* rlock when walking, wlock when changing */
+	QLock configqlock;		/* serialises addition of new configurations */
+	Devlogfs *head;
+	char *defname;
+} devlogfslist;
+
+static LogfsIdentityStore *is;
+
+#ifndef EMU
+char Eunknown[] = "unknown user or group id";
+#endif
+
+static	void	devlogfsfree(Devlogfs*);
+
+#define SPLITPATH(path, qtype, instance, qid, qt) { instance = path >> 4; qid = path & 0xf; qt = qtype & QTDIR; }
+#define DATAQID(q, qt) (!(qt) && (q) >= Qfs && (q) < Qend)
+#define MKPATH(instance, qid) ((instance << 4) | qid)
+
+#define PREFIX "logfs"
+
+static char *devlogfsprefix = PREFIX;
+static char *devlogfsctlname = PREFIX "ctl";
+static char *devlogfsusersname = PREFIX "users";
+static char *devlogfsdumpname = PREFIX "dump";
+static char *devlogfsbootsuffix = "boot";
+static char *devlogfs9pversion = "9P2000";
+
+static void
+errorany(char *errmsg)
+{
+	if (errmsg)
+		error(errmsg);
+}
+
+static void *
+emalloc(ulong size)
+{
+	void *p;
+	p = logfsrealloc(nil, size);
+	if (p == nil)
+		error(Enomem);
+	return p;
+}
+
+static char *
+estrdup(char *q)
+{
+	void *p;
+	if (q == nil)
+		return nil;
+	p = logfsrealloc(nil, strlen(q) + 1);
+	if (p == nil)
+		error(Enomem);
+	return strcpy(p, q);
+}
+
+static char *
+estrconcat(char *a, ...)
+{
+	va_list l;
+	char *p, *r;
+	int t;
+
+	t = strlen(a);
+	va_start(l, a);
+	while ((p = va_arg(l, char *)) != nil)
+		t += strlen(p);
+
+	r = logfsrealloc(nil, t + 1);
+	if (r == nil)
+		error(Enomem);
+
+	strcpy(r, a);
+	va_start(l, a);
+	while ((p = va_arg(l, char *)) != nil)
+		strcat(r, p);
+
+	va_end(l);
+
+	return r;
+}
+
+static int
+gen(Chan *c, int i, Dir *dp, int lockit)
+{
+	Devlogfs *l;
+	long size;
+	Qid qid;
+	qid.vers = 0;
+	qid.type = 0;
+
+	if (i + Qctl < Qfs) {
+		switch (i + Qctl) {
+		case Qctl:
+			qid.path = Qctl;
+			devdir(c, qid, devlogfsctlname, 0, eve, 0666, dp);
+			return 1;
+		case Qusers:
+			qid.path = Qusers;
+			devdir(c, qid, devlogfsusersname, 0, eve, 0444, dp);
+			return 1;
+		case Qdump:
+			qid.path = Qdump;
+			devdir(c, qid, devlogfsdumpname, 0, eve, 0444, dp);
+			return 1;
+		}
+	}
+
+	i -= Qfs - Qctl;
+
+	if (lockit)
+		rlock(&devlogfslist.rwlock);
+
+	if (waserror()) {
+		if (lockit)
+			runlock(&devlogfslist.rwlock);
+		nexterror();
+	}
+
+	for (l = devlogfslist.head; l; l = l->next) {
+		if (i < Qend - Qfs)
+			break;
+		i -= Qend - Qfs;
+	}
+
+	if (l == nil) {
+		poperror();
+		if (lockit)
+			runlock(&devlogfslist.rwlock);
+		return -1;
+	}
+
+	switch (Qfs + i) {
+	case Qfsboot:
+		size = l->lb ? logfsbootgetsize(l->lb) : 0;
+		break;
+	default:
+		size = 0;
+		break;
+	}
+	/* perhaps the user id should come from the underlying file */
+	qid.path = MKPATH(l->instance, Qfs + i);
+	devdir(c, qid, l->filename[i], size, eve, 0666, dp);
+
+	poperror();
+	if (lockit)
+		runlock(&devlogfslist.rwlock);
+
+	return 1;
+}
+
+static int
+devlogfsgen(Chan *c, char *n, Dirtab *tab, int ntab, int i, Dir *dp)
+{
+	USED(n);
+	USED(tab);
+	USED(ntab);
+	return gen(c, i, dp, 1);
+}
+
+static int
+devlogfsgennolock(Chan *c, char *n, Dirtab *tab, int ntab, int i, Dir *dp)
+{
+	USED(n);
+	USED(tab);
+	USED(ntab);
+	return gen(c, i, dp, 0);
+}
+
+/* called under lock */
+static Devlogfs *
+devlogfsfind(int instance)
+{
+	Devlogfs *l;
+
+	for (l = devlogfslist.head; l; l = l->next)
+		if (l->instance == instance)
+			break;
+	return l;
+}
+
+static Devlogfs *
+devlogfsget(int instance)
+{
+	Devlogfs *l;
+	rlock(&devlogfslist.rwlock);
+	for (l = devlogfslist.head; l; l = l->next)
+		if (l->instance == instance)
+			break;
+	if (l)
+		incref(&l->ref);
+	runlock(&devlogfslist.rwlock);
+	return l;
+}
+
+static Devlogfs *
+devlogfsfindbyname(char *name)
+{
+	Devlogfs *l;
+
+	rlock(&devlogfslist.rwlock);
+	for (l = devlogfslist.head; l; l = l->next)
+		if (strcmp(l->name, name) == 0)
+			break;
+	runlock(&devlogfslist.rwlock);
+	return l;
+}
+
+static Devlogfs *
+devlogfssetdefname(char *name)
+{
+	Devlogfs *l;
+	char *searchname;
+	wlock(&devlogfslist.rwlock);
+	if (waserror()) {
+		wunlock(&devlogfslist.rwlock);
+		nexterror();
+	}
+	if (name == nil)
+		searchname = devlogfslist.defname;
+	else
+		searchname = name;
+	for (l = devlogfslist.head; l; l = l->next)
+		if (strcmp(l->name, searchname) == 0)
+			break;
+	if (l == nil) {
+		logfsfreemem(devlogfslist.defname);
+		devlogfslist.defname = nil;
+	}
+	else if (name) {
+		if (devlogfslist.defname) {
+			logfsfreemem(devlogfslist.defname);
+			devlogfslist.defname = nil;
+		}
+		devlogfslist.defname = estrdup(name);
+	}
+	poperror();
+	wunlock(&devlogfslist.rwlock);
+	return l;
+}
+
+static Chan *
+devlogfskopen(char *name, char *suffix, int mode)
+{
+	Chan *c;
+	char *fn;
+	int fd;
+
+	fn = estrconcat(name, suffix, 0);
+	fd = kopen(fn, mode);
+	logfsfreemem(fn);
+	if (fd < 0)
+		error(up->env->errstr);
+	c = fdtochan(up->env->fgrp, fd, mode, 0, 1);
+	kclose(fd);
+	return c;
+}
+
+static char *
+xread(void *a, void *buf, long nbytes, ulong offset)
+{
+	Devlogfs *l = a;
+	long rv;
+
+	if (DEVLOGFSIODEBUG || l->trace)
+		print("devlogfs: %s: read(0x%lux, %ld)\n", l->device, offset, nbytes);
+	l->flash->offset = offset;
+	rv = kchanio(l->flash, buf, nbytes, OREAD);
+	if (rv < 0) {
+		print("devlogfs: %s: flash read error: %s\n", l->device, up->env->errstr);
+		return up->env->errstr;
+	}
+	if (rv != nbytes) {
+		print("devlogfs: %s: short flash read: offset %lud, %ld not %ld\n", l->device, offset, rv, nbytes);
+		return "short read";
+	}
+	return nil;
+}
+
+static char *
+xwrite(void *a, void *buf, long nbytes, ulong offset)
+{
+	Devlogfs *l = a;
+	long rv;
+
+	if (DEVLOGFSIODEBUG || l->trace)
+		print("devlogfs: %s: write(0x%lux, %ld)\n", l->device, offset, nbytes);
+	l->flash->offset = offset;
+	rv = kchanio(l->flash, buf, nbytes, OWRITE);
+	if (rv < 0) {
+		print("devlogfs: %s: flash write error: %s\n", l->device, up->env->errstr);
+		return up->env->errstr;
+	}
+	if (rv != nbytes) {
+		print("devlogfs: %s: short flash write: offset %lud, %ld not %ld\n", l->device, offset, rv, nbytes);
+		return "short write";
+	}
+	return nil;
+}
+
+static char *
+xerase(void *a, long address)
+{
+	Devlogfs *l = a;
+	char cmd[40];
+
+	if (DEVLOGFSIODEBUG || l->trace)
+		print("devlogfs: %s: erase(0x%lux)\n", l->device, address);
+	snprint(cmd, sizeof(cmd), "erase 0x%8.8lux", address);
+	if (kchanio(l->flashctl, cmd, strlen(cmd), OWRITE) <= 0) {
+		print("devlogfs: %s: flash erase error: %s\n", l->device, up->env->errstr);
+		return up->env->errstr;
+	}
+	return nil;
+}
+
+static char *
+xsync(void *a)
+{
+	Devlogfs *l = a;
+
+	if (DEVLOGFSIODEBUG || l->trace)
+		print("devlogfs: %s: sync()\n", l->device);
+	if (kchanio(l->flashctl, "sync", 4, OWRITE) <= 0){
+		print("devlogfs: %s: flash sync error: %s\n", l->device, up->env->errstr);
+		return up->env->errstr;
+	}
+	return nil;
+}
+
+//#define LEAKHUNT
+#ifdef LEAKHUNT
+#define MAXLIVE 2000
+typedef struct Live {
+	void *p;
+	int freed;
+	ulong callerpc;
+} Live;
+
+static Live livemem[MAXLIVE];
+
+static void
+leakalloc(void *p, ulong callerpc)
+{
+	int x;
+	int use = -1;
+	for (x = 0; x < MAXLIVE; x++) {
+		if (livemem[x].p == p) {
+			if (!livemem[x].freed)
+				print("leakalloc: unexpected realloc of 0x%.8lux from 0x%.8lux\n", p, callerpc);
+//			else
+//				print("leakalloc: reusing address 0x%.8lux from 0x%.8lux\n", p, callerpc);
+			livemem[x].freed = 0;
+			livemem[x].callerpc = callerpc;
+			return;
+		}
+		else if (use < 0 && livemem[x].p == 0)
+			use = x;
+	}
+	if (use < 0)
+		panic("leakalloc: too many live entries");
+	livemem[use].p = p;
+	livemem[use].freed = 0;
+	livemem[use].callerpc = callerpc;
+}
+
+static void
+leakaudit(void)
+{
+	int x;
+	for (x = 0; x < MAXLIVE; x++) {
+		if (livemem[x].p && !livemem[x].freed)
+			print("leakaudit: 0x%.8lux from 0x%.8lux\n", livemem[x].p, livemem[x].callerpc);
+	}
+}
+
+static void
+leakfree(void *p, ulong callerpc)
+{
+	int x;
+	if (p == nil)
+		return;
+	for (x = 0; x < MAXLIVE; x++) {
+		if (livemem[x].p == p) {
+			if (livemem[x].freed)
+				print("leakfree: double free of 0x%.8lux from 0x%.8lux, originally by 0x%.8lux\n",
+					p, callerpc, livemem[x].callerpc);
+			livemem[x].freed = 1;
+			livemem[x].callerpc = callerpc;
+			return;
+		}
+	}
+	print("leakfree: free of unalloced address 0x%.8lux from 0x%.8lux\n", p, callerpc);
+	leakaudit();
+}
+
+static void
+leakrealloc(void *newp, void *oldp, ulong callerpc)
+{
+	leakfree(oldp, callerpc);
+	leakalloc(newp, callerpc);
+}
+#endif
+
+
+#ifdef LEAKHUNT
+static void *_realloc(void *p, ulong size, ulong callerpc)
+#else
+void *
+logfsrealloc(void *p, ulong size)
+#endif
+{
+	void *q;
+	ulong osize;
+	if (waserror()) {
+		print("wobbly thrown in memory allocator: %s\n", up->env->errstr);
+		nexterror();
+	}
+	if (p == nil) {
+		q = smalloc(size);
+		poperror();
+#ifdef LEAKHUNT
+		leakrealloc(q, nil, callerpc);
+#endif
+		return q;
+	}
+	q = realloc(p, size);
+	if (q) {
+		poperror();
+#ifdef LEAKHUNT
+		leakrealloc(q, p, callerpc);
+#endif
+		return q;
+	}
+	q = smalloc(size);
+	osize = msize(p);
+	if (osize > size)
+		osize = size;
+	memmove(q, p, osize);
+	free(p);
+	poperror();
+#ifdef LEAKHUNT
+	leakrealloc(q, p, callerpc);
+#endif
+	return q;
+}
+
+#ifdef LEAKHUNT
+void *
+logfsrealloc(void *p, ulong size)
+{
+	return _realloc(p, size, getcallerpc(&p));
+}
+
+void *
+nandfsrealloc(void *p, ulong size)
+{
+	return _realloc(p, size, getcallerpc(&p));
+}
+#else
+void *
+nandfsrealloc(void *p, ulong size)
+{
+	return logfsrealloc(p, size);
+}
+#endif
+
+void
+logfsfreemem(void *p)
+{
+#ifdef LEAKHUNT
+	leakfree(p, getcallerpc(&p));
+#endif
+	free(p);
+}
+
+void
+nandfsfreemem(void *p)
+{
+#ifdef LEAKHUNT
+	leakfree(p, getcallerpc(&p));
+#endif
+	free(p);
+}
+
+static Devlogfs *
+devlogfsconfig(char *name, char *device)
+{
+	Devlogfs *newl, *l;
+	int i;
+	int n;
+	char buf[100], *fields[12];
+	long rawblocksize, rawsize;
+
+	newl = nil;
+
+	qlock(&devlogfslist.configqlock);
+
+	if (waserror()) {
+		qunlock(&devlogfslist.configqlock);
+		devlogfsfree(newl);
+		nexterror();
+	}
+
+	rlock(&devlogfslist.rwlock);
+	for (l = devlogfslist.head; l; l = l->next)
+		if (strcmp(l->name, name) == 0) {
+			runlock(&devlogfslist.rwlock);
+			error(Einuse);
+		}
+
+	/* horrid n^2 solution to finding a unique instance number */
+
+	for (i = 0;; i++) {
+		for (l = devlogfslist.head; l; l = l->next)
+			if (l->instance == i)
+				break;
+		if (l == nil)
+			break;
+	}
+	runlock(&devlogfslist.rwlock);
+
+	newl = emalloc(sizeof(Devlogfs));
+	newl->instance = i;
+	newl->name = estrdup(name);
+	newl->device = estrdup(device);
+	newl->filename[Qfs - Qfs] = estrconcat(devlogfsprefix, name, nil);
+	newl->filename[Qfsboot - Qfs] = estrconcat(devlogfsprefix, name, devlogfsbootsuffix, nil);
+	newl->flash = devlogfskopen(device, nil, ORDWR);
+	newl->flashctl = devlogfskopen(device, "ctl", ORDWR);
+	newl->flashctl->offset = 0;
+	if ((n = kchanio(newl->flashctl, buf, sizeof(buf), OREAD)) <= 0) {
+		print("devlogfsconfig: read ctl failed: %s\n", up->env->errstr);
+		error(up->env->errstr);
+	}
+
+	if (n >= sizeof(buf))
+		n = sizeof(buf) - 1;
+	buf[n] = 0;
+	n = tokenize(buf, fields, nelem(fields));
+	if(n < 7)
+		error("unexpected flashctl format");
+	newl->nand = strcmp(fields[3], "nand") == 0;
+	rawblocksize = strtol(fields[6], nil, 0);
+	rawsize = strtol(fields[5], nil, 0)-strtol(fields[4], nil, 0);
+	if(newl->nand == 0)
+		error("only NAND supported at the moment");
+	errorany(nandfsinit(newl, rawsize, rawblocksize, xread, xwrite, xerase, xsync, &newl->ll));
+	wlock(&devlogfslist.rwlock);
+	newl->next = devlogfslist.head;
+	devlogfslist.head = newl;
+	logfsfreemem(devlogfslist.defname);
+	devlogfslist.defname = nil;
+	if (!waserror()){
+		devlogfslist.defname = estrdup(name);
+		poperror();
+	}
+	wunlock(&devlogfslist.rwlock);
+	poperror();
+	qunlock(&devlogfslist.configqlock);
+	return newl;
+}
+
+static void
+devlogfsunconfig(Devlogfs *devlogfs)
+{
+	Devlogfs **lp;
+
+	qlock(&devlogfslist.configqlock);
+
+	if (waserror()) {
+		qunlock(&devlogfslist.configqlock);
+		nexterror();
+	}
+
+	wlock(&devlogfslist.rwlock);
+
+	if (waserror()) {
+		wunlock(&devlogfslist.rwlock);
+		nexterror();
+	}
+
+	for (lp = &devlogfslist.head; *lp && (*lp) != devlogfs; lp = &(*lp)->next)
+		;
+	if (*lp == nil) {
+		if (DEVLOGFSBAD)
+			print("devlogfsunconfig: not in list\n");
+	}
+	else
+		*lp = devlogfs->next;
+
+	poperror();
+	wunlock(&devlogfslist.rwlock);
+
+	/* now invisible to the naked eye */
+	devlogfsfree(devlogfs);
+	poperror();
+	qunlock(&devlogfslist.configqlock);
+}
+
+static void
+devlogfsllopen(Devlogfs *l)
+{
+	qlock(&l->qlock);
+	if (waserror()) {
+		qunlock(&l->qlock);
+		nexterror();
+	}
+	if (l->lb == nil)
+		errorany(logfsbootopen(l->ll, 0, 0, l->logfstrace, 1, &l->lb));
+	l->state = BootOpen;
+	poperror();
+	qunlock(&l->qlock);
+}
+
+static void
+devlogfsllformat(Devlogfs *l, long bootsize)
+{
+	qlock(&l->qlock);
+	if (waserror()) {
+		qunlock(&l->qlock);
+		nexterror();
+	}
+	if (l->lb == nil)
+		errorany(logfsformat(l->ll, 0, 0, bootsize, l->logfstrace));
+	poperror();
+	qunlock(&l->qlock);
+}
+
+static Chan *
+devlogfsattach(char *spec)
+{
+	Chan *c;
+#ifdef CALLTRACE
+	print("devlogfsattach(spec = %s) - start\n", spec);
+#endif
+	/* create the identity store on first attach */
+	if (is == nil)
+		errorany(logfsisnew(&is));
+	c =  devattach(0x29f, spec);
+//	c =  devattach(L'ʟ', spec);
+#ifdef CALLTRACE
+	print("devlogfsattach(spec = %s) - return %.8lux\n", spec, (ulong)c);
+#endif
+	return c;
+}
+
+static Walkqid*
+devlogfswalk(Chan *c, Chan *nc, char **name, int nname)
+{
+	int instance, qid, qt, clone;
+	Walkqid *wq;
+
+#ifdef CALLTRACE
+	print("devlogfswalk(c = 0x%.8lux, nc = 0x%.8lux, name = 0x%.8lux, nname = %d) - start\n",
+		(ulong)c, (ulong)nc, (ulong)name, nname);
+#endif
+	clone = 0;
+	if(nc == nil){
+		nc = devclone(c);
+		nc->type = 0;
+		SPLITPATH(c->qid.path, c->qid.type, instance, qid, qt);
+		if(DATAQID(qid, qt))
+			nc->aux = devlogfsget(instance);
+		clone = 1;
+	}
+	wq = devwalk(c, nc, name, nname, 0, 0, devlogfsgen);
+	if (wq == nil || wq->nqid < nname) {
+		if(clone)
+			cclose(nc);
+	}
+	else if (clone) {
+		wq->clone = nc;
+		nc->type = c->type;
+	}
+#ifdef CALLTRACE
+	print("devlogfswalk(c = 0x%.8lux, nc = 0x%.8lux, name = 0x%.8lux, nname = %d) - return\n",
+		(ulong)c, (ulong)nc, (ulong)name, nname);
+#endif
+	return wq;
+}
+
+static int
+devlogfsstat(Chan *c, uchar *dp, int n)
+{
+#ifdef CALLTRACE
+	print("devlogfsstat(c = 0x%.8lux, dp = 0x%.8lux n= %d)\n",
+		(ulong)c, (ulong)dp, n);
+#endif
+	return devstat(c, dp, n, 0, 0, devlogfsgen);
+}
+
+static Chan*
+devlogfsopen(Chan *c, int omode)
+{
+	int instance, qid, qt;
+
+	omode = openmode(omode);
+	SPLITPATH(c->qid.path, c->qid.type, instance, qid, qt);
+#ifdef CALLTRACE
+	print("devlogfsopen(c = 0x%.8lux, omode = %o, instance = %d, qid = %d, qt = %d)\n",
+		(ulong)c, omode, instance, qid, qt);
+#endif
+
+
+	rlock(&devlogfslist.rwlock);
+	if (waserror()) {
+		runlock(&devlogfslist.rwlock);
+#ifdef CALLTRACE
+		print("devlogfsopen(c = 0x%.8lux, omode = %o) - error %s\n", (ulong)c, omode, up->env->errstr);
+#endif
+		nexterror();
+	}
+
+	if (DATAQID(qid, qt)) {
+		Devlogfs *d;
+		d = devlogfsfind(instance);
+		if (d == nil)
+			error(Enodev);
+		if (strcmp(up->env->user, eve) != 0)
+			error(Eperm);
+		if (qid == Qfs && d->state != BootOpen)
+			error(Eperm);
+		if (d->server == nil) {
+			errorany(logfsservernew(d->lb, d->ll, is, d->openflags, d->logfstrace, &d->server));
+			d->state = NeedVersion;
+		}
+		c = devopen(c, omode, 0, 0, devlogfsgennolock);
+		incref(&d->ref);
+		c->aux = d;
+	}
+	else if (qid == Qctl || qid == Qusers) {
+		if (strcmp(up->env->user, eve) != 0)
+			error(Eperm);
+		c = devopen(c, omode, 0, 0, devlogfsgennolock);
+	}
+	else
+		c = devopen(c, omode, 0, 0, devlogfsgennolock);
+	poperror();
+	runlock(&devlogfslist.rwlock);
+#ifdef CALLTRACE
+	print("devlogfsopen(c = 0x%.8lux, omode = %o) - return\n", (ulong)c, omode);
+#endif
+	return c;
+}
+
+static void
+devlogfsclose(Chan *c)
+{
+	int instance, qid, qt;
+#ifdef CALLTRACE
+	print("devlogfsclose(c = 0x%.8lux)\n", (ulong)c);
+#endif
+	SPLITPATH(c->qid.path, c->qid.type, instance, qid, qt);
+	USED(instance);
+	if(DATAQID(qid, qt) && (c->flag & COPEN) != 0) {
+		Devlogfs *d;
+		d = c->aux;
+		qlock(&d->qlock);
+		if (qid == Qfs && d->state == Attached) {
+			logfsserverflush(d->server);
+			logfsserverfree(&d->server);
+			d->state = BootOpen;
+		}
+		qunlock(&d->qlock);
+		decref(&d->ref);
+	}
+#ifdef CALLTRACE
+	print("devlogfsclose(c = 0x%.8lux) - return\n", (ulong)c);
+#endif
+}
+
+typedef char *(SMARTIOFN)(void *magic, void *buf, long n, ulong offset, int write);
+
+static void
+smartio(SMARTIOFN *io, void *magic, void *buf, long n, ulong offset, long blocksize, int write)
+{
+	void *tmp = nil;
+	ulong blocks, toread;
+
+	if (waserror()) {
+		logfsfreemem(tmp);
+		nexterror();
+	}
+	if (offset % blocksize) {
+		ulong aoffset;
+		int tmpoffset;
+		int tocopy;
+
+		if (tmp == nil)
+			tmp = emalloc(blocksize);
+		aoffset = offset / blocksize;
+		aoffset *= blocksize;
+		errorany((*io)(magic, tmp, blocksize, aoffset, 0));
+		tmpoffset = offset - aoffset;
+		tocopy = blocksize - tmpoffset;
+		if (tocopy > n)
+			tocopy = n;
+		if (write) {
+			memmove((uchar *)tmp + tmpoffset, buf, tocopy);
+			errorany((*io)(magic, tmp, blocksize, aoffset, 1));
+		}
+		else
+			memmove(buf, (uchar *)tmp + tmpoffset, tocopy);
+		buf = (uchar *)buf + tocopy;
+		n -= tocopy;
+		offset = aoffset + blocksize;
+	}
+	blocks = n / blocksize;
+	toread = blocks * blocksize;
+	errorany((*io)(magic, buf, toread, offset, write));
+	buf = (uchar *)buf + toread;
+	n -= toread;
+	offset += toread;
+	if (n) {
+		if (tmp == nil)
+			tmp = emalloc(blocksize);
+		errorany((*io)(magic, tmp, blocksize, offset, 0));
+		if (write) {
+			memmove(tmp, buf, n);
+			errorany((*io)(magic, tmp, blocksize, offset, 1));
+		}
+		memmove(buf, tmp, n);
+	}
+	poperror();
+	logfsfreemem(tmp);
+}
+
+static int
+readok(void *a)
+{
+	Devlogfs *d = a;
+	return d->reading;
+}
+
+static int
+writeok(void *a)
+{
+	Devlogfs *d = a;
+	return !d->reading;
+}
+
+static long
+lfsrvread(Devlogfs *d, void *buf, long n)
+{
+	qlock(&d->rlock);
+	if(waserror()){
+		qunlock(&d->rlock);
+		nexterror();
+	}
+	if (d->state == Hungup)
+		error(Ehungup);
+	Sleep(&d->readrendez, readok, d);
+	if (n > d->readcount)
+		n = d->readcount;
+	memmove(buf, d->readp, n);
+	d->readp += n;
+	d->readcount -= n;
+	if (d->readcount == 0) {
+		d->reading = 0;
+		Wakeup(&d->writerendez);
+	}
+	poperror();
+	qunlock(&d->rlock);
+	return n;
+}
+
+static void
+reply(Devlogfs *d)
+{
+	d->readp = d->readbuf;
+	d->readcount = convS2M(&d->out, d->readp, d->readbufsize);
+//print("reply is %d bytes\n", d->readcount);
+	if (d->readcount == 0)
+		panic("logfs: reply: did not fit\n");
+	d->reading = 1;
+	Wakeup(&d->readrendez);
+}
+
+static void
+rerror(Devlogfs *d, char *ename)
+{
+	d->out.type = Rerror;
+	d->out.ename = ename;
+	reply(d);
+}
+
+static struct {
+	QLock qlock;
+	int (*read)(void *magic, Devlogfs *d, int line, char *buf, int buflen);
+	void *magic;
+	Devlogfs *d;
+	int line;
+} dump;
+
+static void *
+extentdumpinit(Devlogfs *d, int argc, char **argv)
+{
+	int *p;
+	ulong path;
+	u32int flashaddr, length;
+	long block;
+	int page, offset;
+
+	if (argc != 1)
+		error(Ebadarg);
+	path = strtoul(argv[0], 0, 0);
+	errorany(logfsserverreadpathextent(d->server, path, 0, &flashaddr, &length, &block, &page, &offset));
+	p = emalloc(sizeof(ulong));
+	*p = path;
+	return p;
+}
+
+static int
+extentdumpread(void *magic, Devlogfs *d, int line, char *buf, int buflen)
+{
+	ulong *p = magic;
+	u32int flashaddr, length;
+	long block;
+	int page, offset;
+	USED(d);
+	errorany(logfsserverreadpathextent(d->server, *p, line, &flashaddr, &length, &block, &page, &offset));
+	if (length == 0)
+		return 0;
+	return snprint(buf, buflen, "%.8ux %ud %ld %d %d\n", flashaddr, length, block, page, offset);
+}
+
+static void
+devlogfsdumpinit(Devlogfs *d,
+	void *(*init)(Devlogfs *d, int argc, char **argv),
+	int (*read)(void *magic, Devlogfs *d, int line, char *buf, int buflen), int argc, char **argv)
+{
+	qlock(&dump.qlock);
+	if (waserror()) {
+		qunlock(&dump.qlock);
+		nexterror();
+	}
+	if (d) {
+		if (d->state < NeedVersion)
+			error("not mounted");
+		qlock(&d->qlock);
+		if (waserror()) {
+			qunlock(&d->qlock);
+			nexterror();
+		}
+	}
+	if (dump.magic) {
+		logfsfreemem(dump.magic);
+		dump.magic = nil;
+	}
+	dump.d = d;
+	dump.magic = (*init)(d, argc, argv);
+	dump.read = read;
+	dump.line = 0;
+	if (d) {
+		poperror();
+		qunlock(&d->qlock);
+	}
+	poperror();
+	qunlock(&dump.qlock);
+}
+
+static long
+devlogfsdumpread(char *buf, int buflen)
+{
+	char *tmp = nil;
+	long n;
+	qlock(&dump.qlock);
+	if (waserror()) {
+		logfsfreemem(tmp);
+		qunlock(&dump.qlock);
+		nexterror();
+	}
+	if (dump.magic == nil)
+		error(Eio);
+	tmp = emalloc(READSTR);
+	if (dump.d) {
+		if (dump.d->state < NeedVersion)
+			error("not mounted");
+		qlock(&dump.d->qlock);
+		if (waserror()) {
+			qunlock(&dump.d->qlock);
+			nexterror();
+		}
+	}
+	n = (*dump.read)(dump.magic, dump.d, dump.line, tmp, READSTR);
+	if (n) {
+		dump.line++;
+		n = readstr(0, buf, buflen, tmp);
+	}
+	if (dump.d) {
+		poperror();
+		qunlock(&dump.d->qlock);
+	}
+	logfsfreemem(tmp);
+	poperror();
+	qunlock(&dump.qlock);
+	return n;
+}
+
+static void
+devlogfsserverlogsweep(Devlogfs *d, int justone)
+{
+	int didsomething;
+	if (d->state < NeedVersion)
+		error("not mounted");
+	qlock(&d->qlock);
+	if (waserror()) {
+		qunlock(&d->qlock);
+		nexterror();
+	}
+	errorany(logfsserverlogsweep(d->server, justone, &didsomething));
+	poperror();
+	qunlock(&d->qlock);
+}
+
+static void
+devlogfsserversync(Devlogfs *d)
+{
+	if (d->state < NeedVersion)
+		return;
+	qlock(&d->qlock);
+	if (waserror()) {
+		qunlock(&d->qlock);
+		nexterror();
+	}
+	errorany(logfsserverflush(d->server));
+	poperror();
+	qunlock(&d->qlock);
+}
+
+static void
+lfssrvwrite(Devlogfs *d, void *buf, long n)
+{
+	volatile int locked = 0;
+
+	qlock(&d->wlock);
+	if(waserror()){
+		qunlock(&d->wlock);
+		nexterror();
+	}
+	if (d->state == Hungup)
+		error(Ehungup);
+	Sleep(&d->writerendez, writeok, d);
+	if (convM2S(buf, n, &d->in) != n) {
+		/*
+		 * someone is writing drivel; have nothing to do with them anymore
+		 * most common cause; trying to mount authenticated
+		 */
+		d->state = Hungup;
+		error(Ehungup);
+	}
+	d->out.tag = d->in.tag;
+	d->out.fid = d->in.fid;
+	d->out.type = d->in.type + 1;
+	if (waserror()) {
+		if (locked)
+			qunlock(&d->qlock);
+		rerror(d, up->env->errstr);
+		goto Replied;
+	}
+	if (d->in.type != Tversion && d->in.type != Tattach) {
+		if (d->state != Attached)
+			error("must be attached");
+		qlock(&d->qlock);
+		locked = 1;
+	}
+	switch (d->in.type) {
+	case Tauth:
+		error("no authentication needed");
+	case Tversion: {
+		char *rversion;
+		if (d->state != NeedVersion)
+			error("unexpected Tversion");
+		 if (d->in.tag != NOTAG)
+			error("protocol botch");
+		/*
+		 * check the version string
+		 */
+		if (strcmp(d->in.version, devlogfs9pversion) != 0)
+			rversion = "unknown";
+		else
+			rversion = devlogfs9pversion;
+		/*
+		 * allocate the reply buffer
+		 */
+		d->readbufsize = d->in.msize;
+		if (d->readbufsize > MAXMSIZE)
+			d->readbufsize = MAXMSIZE;
+		d->readbuf = emalloc(d->readbufsize);
+		/*
+		 * compose the Rversion
+		 */
+		d->out.msize = d->readbufsize;
+		d->out.version = rversion;
+		d->state = NeedAttach;
+		break;
+	}
+	case Tattach:
+		if (d->state != NeedAttach)
+			error("unexpected attach");
+		if (d->in.afid != NOFID)
+			error("unexpected afid");
+		errorany(logfsserverattach(d->server, d->in.fid, d->in.uname, &d->out.qid));
+		d->state = Attached;
+		break;
+	case Tclunk:
+		errorany(logfsserverclunk(d->server, d->in.fid));
+		break;
+	case Tcreate:
+		errorany(logfsservercreate(d->server, d->in.fid, d->in.name, d->in.perm, d->in.mode, &d->out.qid));
+		d->out.iounit = d->readbufsize - 11;
+		break;
+	case Tflush:
+		break;
+	case Topen:
+		errorany(logfsserveropen(d->server, d->in.fid, d->in.mode, &d->out.qid));
+		d->out.iounit = d->readbufsize - 11;
+		break;
+	case Tread:
+		d->out.data = (char *)d->readbuf + 11;
+		/* TODO - avoid memmove */
+		errorany(logfsserverread(d->server, d->in.fid, d->in.offset, d->in.count, (uchar *)d->out.data,
+			d->readbufsize - 11, &d->out.count));
+		break;
+	case Tremove:
+		errorany(logfsserverremove(d->server, d->in.fid));
+		break;
+	case Tstat:
+		d->out.stat = d->readbuf + 9;
+		/* TODO - avoid memmove */
+		errorany(logfsserverstat(d->server, d->in.fid, d->out.stat, d->readbufsize - 9, &d->out.nstat));
+//		print("nstat %d\n", d->out.nstat);
+		break;
+	case Twalk:
+		errorany(logfsserverwalk(d->server, d->in.fid, d->in.newfid,
+			d->in.nwname, d->in.wname, &d->out.nwqid, d->out.wqid));
+		break;
+	case Twrite:
+		errorany(logfsserverwrite(d->server, d->in.fid, d->in.offset, d->in.count, (uchar *)d->in.data,
+			&d->out.count));
+		break;
+	case Twstat:
+		errorany(logfsserverwstat(d->server, d->in.fid, d->in.stat, d->in.nstat));
+		break;
+	default:
+		print("lfssrvwrite: msg %d unimplemented\n", d->in.type);
+		error("unimplemented");
+	}
+	poperror();
+	if (locked)
+		qunlock(&d->qlock);
+	reply(d);
+Replied:
+	poperror();
+	qunlock(&d->wlock);
+}
+
+static long
+devlogfsread(Chan *c, void *buf, long n, vlong off)
+{
+	int instance, qid, qt;
+
+	SPLITPATH(c->qid.path, c->qid.type, instance, qid, qt);
+	USED(instance);
+#ifdef CALLTRACE
+	print("devlogfsread(c = 0x%.8lux, buf = 0x%.8lux, n = %ld, instance = %d, qid = %d, qt = %d) - start\n",
+		(ulong)c, (ulong)buf, n, instance, qid, qt);
+#endif
+	if(qt & QTDIR) {
+#ifdef CALLTRACE
+		print("devlogfsread(c = 0x%.8lux, buf = 0x%.8lux, n = %ld, instance = %d, qid = %d, qt = %d) - calling devdirread\n",
+			(ulong)c, (ulong)buf, n, instance, qid, qt);
+#endif
+		return devdirread(c, buf, n, 0, 0, devlogfsgen);
+	}
+
+	if(DATAQID(qid, qt)) {
+		if (qid == Qfsboot) {
+			Devlogfs *l = c->aux;
+			qlock(&l->bootqlock);
+			if (waserror()) {
+				qunlock(&l->bootqlock);
+				nexterror();
+			}
+			smartio((SMARTIOFN *)logfsbootio, l->lb, buf, n, off, logfsbootgetiosize(l->lb), 0);
+			poperror();
+			qunlock(&l->bootqlock);
+			return n;
+		}
+		else if (qid == Qfs) {
+			Devlogfs *d = c->aux;
+			return lfsrvread(d, buf, n);
+		}
+		error(Eio);
+	}
+
+	if (qid == Qusers) {
+		long nr;
+		errorany(logfsisusersread(is, buf, n, (ulong)off, &nr));
+		return nr;
+	}
+	else if (qid == Qdump)
+		return devlogfsdumpread(buf, n);
+
+	if (qid != Qctl)
+		error(Egreg);
+
+	return 0;
+}
+
+enum {
+	CMconfig,
+	CMformat,
+	CMopen,
+	CMsweep,
+	CMtrace,
+	CMunconfig,
+	CMextent,
+	CMsweepone,
+	CMtest,
+	CMleakaudit,
+	CMsync
+};
+
+static Cmdtab fscmds[] = {
+	{CMconfig, "config", 2},
+	{CMformat, "format", 2},
+	{CMopen, "open", 0},
+	{CMsweep, "sweep", 1},
+	{CMsweepone, "sweepone", 1},
+	{CMtrace, "trace", 0},
+	{CMunconfig, "unconfig", 1},
+	{CMextent, "extent", 0},
+	{CMtest, "test", 0},
+	{CMleakaudit, "leakaudit", 1},
+	{CMsync, "sync", 1},
+};
+
+static long
+devlogfswrite(Chan *c, void *buf, long n, vlong off)
+{
+	int instance, qid, qt, i;
+	Cmdbuf *cmd;
+	Cmdtab *ct;
+
+	if(n <= 0)
+		return 0;
+	SPLITPATH(c->qid.path, c->qid.type, instance, qid, qt);
+#ifdef CALLTRACE
+	print("devlogfswrite(c = 0x%.8lux, buf = 0x%.8lux, n = %ld, instance = %d, qid = %d, qt = %d) - start\n",
+		(ulong)c, (ulong)buf, n, instance, qid, qt);
+#endif
+	USED(instance);
+	if(DATAQID(qid, qt)){
+		if (qid == Qfsboot) {
+			Devlogfs *l = c->aux;
+			qlock(&l->bootqlock);
+			if (waserror()) {
+				qunlock(&l->bootqlock);
+				nexterror();
+			}
+			smartio((SMARTIOFN *)logfsbootio, l->lb, buf, n, off, logfsbootgetiosize(l->lb), 1);
+			poperror();
+			qunlock(&l->bootqlock);
+			return n;
+		}
+		else if (qid == Qfs) {
+			Devlogfs *d = c->aux;
+			lfssrvwrite(d, buf, n);
+			return n;
+		}
+		error(Eio);
+	}
+	else if (qid == Qctl) {
+		Devlogfs *l = nil;
+		int a;
+
+		cmd = parsecmd(buf, n);
+		if(waserror()){
+			free(cmd);
+			nexterror();
+		}
+		i = cmd->nf;
+		if(0){print("i=%d", i); for(i=0; i<cmd->nf; i++)print(" %q", cmd->f[i]); print("\n");}
+		if (i <= 0)
+			error(Ebadarg);
+		if (i == 3 && strcmp(cmd->f[0], "uname") == 0) {
+			switch (cmd->f[2][0]) {
+			default:
+				errorany(logfsisgroupcreate(is, cmd->f[1], cmd->f[2]));
+				break;
+			case ':':
+				errorany(logfsisgroupcreate(is, cmd->f[1], cmd->f[2] + 1));
+				break;
+			case '%':
+				errorany(logfsisgrouprename(is, cmd->f[1], cmd->f[2] + 1));
+				break;
+			case '=':
+				errorany(logfsisgroupsetleader(is, cmd->f[1], cmd->f[2] + 1));
+				break;
+			case '+':
+				errorany(logfsisgroupaddmember(is, cmd->f[1], cmd->f[2] + 1));
+				break;
+			case '-':
+				errorany(logfsisgroupremovemember(is, cmd->f[1], cmd->f[2] + 1));
+				break;
+			}
+			i = 0;
+		}
+		if (i == 4 && strcmp(cmd->f[0], "fsys") == 0 && strcmp(cmd->f[2], "config") == 0) {
+			l = devlogfsconfig(cmd->f[1], cmd->f[3]);
+			i = 0;
+		}
+		else if (i >= 2 && strcmp(cmd->f[0], "fsys") == 0) {
+			l = devlogfssetdefname(cmd->f[1]);
+			if (l == nil)
+				errorf("file system %q not configured", cmd->f[1]);
+			i -= 2;
+			cmd->f += 2;
+			cmd->nf = i;
+		}
+		if (i != 0) {
+			ct = lookupcmd(cmd, fscmds, nelem(fscmds));
+			if (l == nil)
+				l = devlogfssetdefname(nil);
+			if(l == nil && ct->index != CMleakaudit)
+				error("file system not configured");
+			switch(ct->index){
+			case CMopen:
+				for (a = 1; a < i; a++)
+					if (cmd->f[a][0] == '-')
+						switch (cmd->f[a][1]) {
+						case 'P':
+							l->openflags |= LogfsOpenFlagNoPerm;
+							break;
+						case 'W':
+							l->openflags |= LogfsOpenFlagWstatAllow;
+							break;
+						default:
+							error(Ebadarg);
+						}
+				devlogfsllopen(l);
+				break;
+			case CMformat:
+				devlogfsllformat(l, strtol(cmd->f[1], nil, 0));
+				break;
+			case CMsweep:
+				devlogfsserverlogsweep(l, 0);
+				break;
+			case CMsweepone:
+				devlogfsserverlogsweep(l, 1);
+				break;
+			case CMtrace:
+				l->logfstrace = i > 1 ? strtol(cmd->f[1], nil, 0) : 0;
+				if (l->server)
+					logfsservertrace(l->server, l->logfstrace);
+				if (l->lb)
+					logfsboottrace(l->lb, l->logfstrace);
+				break;
+			case CMunconfig:
+				if (l->ref.ref > 0)
+					error(Einuse);
+				devlogfsunconfig(l);
+				break;
+			case CMextent:
+				if(i < 2)
+					error(Ebadarg);
+				devlogfsdumpinit(l, extentdumpinit, extentdumpread, i - 1, cmd->f + 1);
+				break;
+			case CMtest:
+				if(i < 2)
+					error(Ebadarg);
+				errorany(logfsservertestcmd(l->server, i - 1, cmd->f + 1));
+				break;
+			case CMleakaudit:
+#ifdef LEAKHUNT
+				leakaudit();
+#endif
+				break;
+			case CMsync:
+				devlogfsserversync(l);
+				break;
+			default:
+				error(Ebadarg);
+			}
+		}
+		poperror();
+		free(cmd);
+		return n;
+	}
+	error(Egreg);
+	return 0;		/* not reached */
+}
+
+static void
+devlogfsfree(Devlogfs *devlogfs)
+{
+	if (devlogfs != nil) {
+		int i;
+		logfsfreemem(devlogfs->device);
+		logfsfreemem(devlogfs->name);
+		for (i = 0; i < Qend - Qfs; i++)
+			logfsfreemem(devlogfs->filename[i]);
+		cclose(devlogfs->flash);
+		cclose(devlogfs->flashctl);
+		qlock(&devlogfs->qlock);
+		logfsserverfree(&devlogfs->server);
+		logfsbootfree(devlogfs->lb);
+		if (devlogfs->ll)
+			(*devlogfs->ll->free)(devlogfs->ll);
+		logfsfreemem(devlogfs->readbuf);
+		qunlock(&devlogfs->qlock);
+		logfsfreemem(devlogfs);
+	}
+}
+
+#ifdef EMU
+ulong
+logfsnow(void)
+{
+	extern vlong timeoffset;
+	return (timeoffset + osusectime()) / 1000000;
+}
+#endif
+
+Dev logfsdevtab = {
+	0x29f,
+//	L'ʟ',
+	"logfs",
+
+#ifndef EMU
+	devreset,
+#endif
+	devinit,
+#ifndef EMU
+	devshutdown,
+#endif
+	devlogfsattach,
+	devlogfswalk,
+	devlogfsstat,
+	devlogfsopen,
+	devcreate,
+	devlogfsclose,
+	devlogfsread,
+	devbread,
+	devlogfswrite,
+	devbwrite,
+	devremove,
+	devwstat,
+};