Don't open same disk twice on OpenFirmware.
PATCH 1/1 Don't open same disk twice on OpenFirmware.
Hello, all. According to David Miller sparc's openboot doesn't support opening same disk twice. So I implemented handle reusage logic. Tested on imac g3
Signature · 2 lines
Regards Vladimir 'φ-coder/phcoder' Serbinenko
=== modified file 'disk/ieee1275/ofdisk.c'
ieee1275/ofdisk.c +41 -11
@@ -1,7 +1,7 @@/* ofdisk.c - Open Firmware disk access. */ /* * GRUB -- GRand Unified Bootloader -* Copyright (C) 2004,2006,2007,2008 Free Software Foundation, Inc. +* Copyright (C) 2004,2006,2007,2008,2009 Free Software Foundation, Inc. * * GRUB is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by
@@ -26,6 +26,8 @@struct ofdisk_hash_ent {char *devpath; +int refs; +grub_ieee1275_ihandle_t dev_ihandle;struct ofdisk_hash_ent *next; };
@@ -65,6 +67,8 @@ ofdisk_hash_add (char *devpath) { p->devpath = devpath; p->next = *head; + p->refs =0 ; + p->dev_ihandle =0 ; *head = p; }return p;
@@ -170,6 +174,23 @@ grub_ofdisk_open (const char *name, grubif (!op)return grub_errno; +if (op->dev_ihandle) + { + op->refs++; + +/* XXX: There is no property to read the number of blocks. There +should be a property `#blocks', but it is not there. Perhaps it +is possible to use seek for this. */ + disk->total_sectors =0xFFFFFFFFUL ; + + disk->id = (unsigned long ) op; + +/* XXX: Read this, somehow. */ + disk->has_partitions =1 ; + disk->data = op; +return 0 ; + } +grub_dprintf ("disk" ,"Opening `%s'.\n" , op->devpath);grub_ieee1275_open (op->devpath, &dev_ihandle);
@@ -179,8 +200,8 @@ grub_ofdisk_open (const char *name, grub goto fail; } -grub_dprintf ("disk" ,"Opened `%s' as handle %p.\n" , op->devpath, - (void *) (unsigned long ) dev_ihandle); +grub_dprintf ("disk" ,"Opened `%s' as handle 0x%lx.\n" , op->devpath, + (unsigned long ) dev_ihandle);if (grub_ieee1275_finddevice (op->devpath, &dev)) {
@@ -201,6 +222,9 @@ grub_ofdisk_open (const char *name, grub goto fail; } + op->dev_ihandle = dev_ihandle; + op->refs++; +/* XXX: There is no property to read the number of blocks. There should be a property `#blocks', but it is not there. Perhaps it is possible to use seek for this. */
@@ -210,7 +234,7 @@ grub_ofdisk_open (const char *name, grub/* XXX: Read this, somehow. */ disk->has_partitions =1 ; - disk->data = (void *) (unsigned long ) dev_ihandle; + disk->data = op;return 0 ; fail:
@@ -222,9 +246,15 @@ grub_ofdisk_open (const char *name, grubstatic void grub_ofdisk_close (grub_disk_t disk) { -grub_dprintf ("disk" ,"Closing handle %p.\n" , - (void *) disk->data); -grub_ieee1275_close ((grub_ieee1275_ihandle_t ) (unsigned long ) disk->data); +struct ofdisk_hash_ent *data = disk->data; + + data->refs--; +if (data->refs) +return ; + +grub_dprintf ("disk" ,"Closing handle %p.\n" , data); +grub_ieee1275_close (data->dev_ihandle); + data->dev_ihandle =0 ; } static grub_err_t
@@ -233,21 +263,21 @@ grub_ofdisk_read (grub_disk_t disk, grub {grub_ssize_t status, actual;unsigned long long pos; +struct ofdisk_hash_ent *data = disk->data;grub_dprintf ("disk" ,"Reading handle %p: sector 0x%llx, size 0x%lx, buf %p.\n" , - (void *) disk->data, (long long ) sector, (long ) size, buf); + data, (long long ) sector, (long ) size, buf); pos = sector *512UL ; -grub_ieee1275_seek ((grub_ieee1275_ihandle_t ) (unsigned long ) disk->data, +grub_ieee1275_seek (data->dev_ihandle, (int ) (pos >>32 ), (int ) pos &0xFFFFFFFFUL , &status);if (status <0 )return grub_error (GRUB_ERR_READ_ERROR ,"Seek error, can't seek block %llu" , (long long ) sector); -grub_ieee1275_read ((grub_ieee1275_ihandle_t ) (unsigned long ) disk->data, - buf, size *512UL , &actual); +grub_ieee1275_read (data->dev_ihandle, buf, size *512UL , &actual);if (actual != actual)return grub_error (GRUB_ERR_READ_ERROR ,"Read error on block: %llu" , (long long ) sector);
From: Vladimir 'φ-coder/phcoder' Serbinenko <[email protected]>
Date: Fri, 18 Dec 2009 17:17:13 +0100
Quoted 3 lines
Hello, all. According to David Miller sparc's openboot doesn't support opening same disk twice. So I implemented handle reusage logic. Tested on imac g3
At a minimum you have to seek to the beginning of the device as that is one of the many side effects of openning the device.
I really don't think this is a good idea, really.
We sould just enforce the restriction, and put a debugging check there to make sure the invariant is never violated.
From: David Miller <[email protected]>
Date: Fri, 18 Dec 2009 10:33:13 -0800 (PST)
Quoted 10 lines
From: Vladimir 'φ-coder/phcoder' Serbinenko <[email protected]>
Date: Fri, 18 Dec 2009 17:17:13 +0100Quoted 3 lines
Hello, all. According to David Miller sparc's openboot doesn't support opening same disk twice. So I implemented handle reusage logic. Tested on imac g3
At a minimum you have to seek to the beginning of the device as that is one of the many side effects of openning the device.
In fact this proves that your idea can't ever work. Each openned instance will expect the file offset pointer to be at different locations.
You can't share open instances, therefore.
We really just have to make sure GRUB never violates this restriction.
Quoted 17 lines · David Miller
From: David Miller <[email protected]>
Date: Fri, 18 Dec 2009 10:33:13 -0800 (PST)Quoted 4 lines
From: Vladimir 'φ-coder/phcoder' Serbinenko <[email protected]>
Date: Fri, 18 Dec 2009 17:17:13 +0100Quoted 8 lines
Quoted 4 lines
Hello, all. According to David Miller sparc's openboot doesn't support opening same disk twice. So I implemented handle reusage logic. Tested on imac g3
At a minimum you have to seek to the beginning of the device as that is one of the many side effects of openning the device.
Quoted 9 lines
In fact this proves that your idea can't ever work. Each openned instance will expect the file offset pointer to be at different locations.
You can't share open instances, therefore.
We really just have to make sure GRUB never violates this restriction.
Actually we don't open files, only disks and we explicitely seek to desired position. It was so even before my patch.
Signature · 2 lines
Regards Vladimir 'φ-coder/phcoder' Serbinenko