* [PATCH v1 0/7] Add a mtd raid layer
@ 2015-12-01 10:52 Dongsheng Yang
2015-12-01 10:52 ` [PATCH v1 1/7] mtd: nandsim: remove the abuse of reference to globle variable Dongsheng Yang
` (8 more replies)
0 siblings, 9 replies; 10+ messages in thread
From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw)
To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang
Hi guys,
I am here to release v1 of the mtd raid layer. It's a layer
shown as below:
--------- -------------
| ubi | | mtdchar |
--------- -------------
| |
v v
----------------------------
| MTD RAID | <----- a new layer here (drivers/mtd/mtd_raid/*)
----------------------------
|
v
----------------
| mtd driver |
----------------
Of course, that's optional.
There are two ways to use mtd raid.
(1). For multi-chips driver:
/dev/mtd0
|--------------------------------|
| ---------------------------- |
| | MTD RAID | |
| ---------------------------- |
| | | |
| v v |
| ----------- ----------- |
| | mtd_info| | mtd_info| |
| ----------- ----------- |
|--------------------------------|
When a driver have multiple chips, we can use mtd raid framework to build
them into a raid array and expose a mtd device. There are some drivers
are using mtd_concat, we can do the same thing with mtd_single. At the same
time, mtd raid provides more choices to them, such as raid0 or raid1.
(2). For user:
/dev/mtd2
|--------------------------------|
| ---------------------------- |
| | MTD RAID | |
| ---------------------------- |
| | | |
| v v |
| ----------- ----------- |
| |/dev/mtd0| |/dev/mtd1| |
| ----------- ----------- |
|--------------------------------|
we have a user tool for mtd raid, and user can use it to build raid array.
Example:
# mtd_raid create --level=0 /dev/mtd0 /dev/mtd1 /dev/mtd2 /dev/mtd3
# mtdinfo /dev/mtd4
mtd4
Name: mtdraid0-1
Type: nand
Eraseblock size: 65536 bytes, 64.0 KiB
Amount of eraseblocks: 8192 (536870912 bytes, 512.0 MiB)
Minimum input/output unit size: 512 bytes
Sub-page size: 256 bytes
OOB size: 16 bytes
Character device major/minor: 90:8
Bad blocks are allowed: true
Device is writable: true
There are 4 mtd devices simulated by nandsim, then we create a raid0 with
them.
---------------------------------------------------------------------------
| device| size | throughput (dd if=/dev/mtdX of=/dev/null bs=1M count=10)|
|--------------------------------------------------------------------------
| mtd0 | 128M | 14.0 MB/s |
|--------------------------------------------------------------------------
| mtd1 | 128M | 14.0 MB/s |
|--------------------------------------------------------------------------
| mtd2 | 128M | 14.0 MB/s |
|--------------------------------------------------------------------------
| mtd3 | 128M | 14.0 MB/s |
|--------------------------------------------------------------------------
| mtd4 | 512M | 51.1 MB/s |
---------------------------------------------------------------------------
Also we can use raid1 to get a mirror for robustness.
TODO:
In this framework, we can implement more raid levels. The most useful
one I think is raid5/6, which would do a parity and fixup work in reading. That's
important for MLC or TLC. Raid-level is refered to:
https://en.wikipedia.org/wiki/Standard_RAID_levels
Code:
kernel: https://github.com/yangdongsheng/linux mtd_raid_v1
user : https://github.com/yangdongsheng/mtd-utils mtd_raid_v1
Thanx
Yang
Dongsheng Yang (7):
mtd: nandsim: remove the abuse of reference to globle variable.
mtd: nansim: add support of multi devices.
mtd: mtd_raid: Init a new layer of MTD RAID
mtd: mtd_raid: introduce raid single implementation
mtd: mtd_raid: introduce raid0 implementation
mtd: mtd_raid: introduce raid1 implementation
mtd: mtd_raid: implement ioctl for mtd raid
Documentation/ioctl/ioctl-number.txt | 1 +
drivers/mtd/Kconfig | 2 +
drivers/mtd/Makefile | 1 +
drivers/mtd/mtd_raid/Kconfig | 12 +
drivers/mtd/mtd_raid/Makefile | 3 +
drivers/mtd/mtd_raid/core.c | 1000 ++++++++++++++++++++++++++++++++++
drivers/mtd/mtd_raid/ioctl.c | 137 +++++
drivers/mtd/mtd_raid/mtd_raid.h | 264 +++++++++
drivers/mtd/mtd_raid/raid0.c | 136 +++++
drivers/mtd/mtd_raid/raid1.c | 156 ++++++
drivers/mtd/mtd_raid/raid_io.c | 449 +++++++++++++++
drivers/mtd/mtd_raid/raid_single.c | 156 ++++++
drivers/mtd/nand/nandsim.c | 312 ++++++-----
include/uapi/mtd/mtd-raid-user.h | 33 ++
14 files changed, 2518 insertions(+), 144 deletions(-)
create mode 100644 drivers/mtd/mtd_raid/Kconfig
create mode 100644 drivers/mtd/mtd_raid/Makefile
create mode 100644 drivers/mtd/mtd_raid/core.c
create mode 100644 drivers/mtd/mtd_raid/ioctl.c
create mode 100644 drivers/mtd/mtd_raid/mtd_raid.h
create mode 100644 drivers/mtd/mtd_raid/raid0.c
create mode 100644 drivers/mtd/mtd_raid/raid1.c
create mode 100644 drivers/mtd/mtd_raid/raid_io.c
create mode 100644 drivers/mtd/mtd_raid/raid_single.c
create mode 100644 include/uapi/mtd/mtd-raid-user.h
--
1.8.4.2
^ permalink raw reply [flat|nested] 10+ messages in thread* [PATCH v1 1/7] mtd: nandsim: remove the abuse of reference to globle variable. 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 2/7] mtd: nansim: add support of multi devices Dongsheng Yang ` (7 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang The function of do_bit_flips() is working on struct nandsim, where we can not see struct mtd_info which in a higher level of view. This patch remove reference to nsmtd in do_bit_flips(). If you want to get the same information of it, please print it in the mtd level. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/nand/nandsim.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/drivers/mtd/nand/nandsim.c b/drivers/mtd/nand/nandsim.c index 95d0cc4..d4a94c4 100644 --- a/drivers/mtd/nand/nandsim.c +++ b/drivers/mtd/nand/nandsim.c @@ -1456,9 +1456,8 @@ static void do_bit_flips(struct nandsim *ns, int num) int pos = prandom_u32() % (num * 8); ns->buf.byte[pos / 8] ^= (1 << (pos % 8)); NS_WARN("read_page: flipping bit %d in page %d " - "reading from %d ecc: corrected=%u failed=%u\n", - pos, ns->regs.row, ns->regs.column + ns->regs.off, - nsmtd->ecc_stats.corrected, nsmtd->ecc_stats.failed); + "reading from %d\n", + pos, ns->regs.row, ns->regs.column + ns->regs.off); } } } -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 2/7] mtd: nansim: add support of multi devices. 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 1/7] mtd: nandsim: remove the abuse of reference to globle variable Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 3/7] mtd: mtd_raid: Init a new layer of MTD RAID Dongsheng Yang ` (6 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang This patch introduces a new option names as dev_count to specify the count of dev. Then we can simulate multi nand flashes with memory. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/nand/nandsim.c | 307 ++++++++++++++++++++++++--------------------- 1 file changed, 166 insertions(+), 141 deletions(-) diff --git a/drivers/mtd/nand/nandsim.c b/drivers/mtd/nand/nandsim.c index d4a94c4..479de7d 100644 --- a/drivers/mtd/nand/nandsim.c +++ b/drivers/mtd/nand/nandsim.c @@ -107,6 +107,7 @@ static unsigned int overridesize = 0; static char *cache_file = NULL; static unsigned int bbt; static unsigned int bch; +static unsigned int dev_count = 1; static u_char id_bytes[8] = { [0] = CONFIG_NANDSIM_FIRST_ID_BYTE, [1] = CONFIG_NANDSIM_SECOND_ID_BYTE, @@ -139,6 +140,7 @@ module_param(overridesize, uint, 0400); module_param(cache_file, charp, 0400); module_param(bbt, uint, 0400); module_param(bch, uint, 0400); +module_param(dev_count, uint, 0400); MODULE_PARM_DESC(id_bytes, "The ID bytes returned by NAND Flash 'read ID' command"); MODULE_PARM_DESC(first_id_byte, "The first byte returned by NAND Flash 'read ID' command (manufacturer ID) (obsolete)"); @@ -447,8 +449,7 @@ static unsigned long *erase_block_wear = NULL; static unsigned int wear_eb_count = 0; static unsigned long total_wear = 0; -/* MTD structure for NAND controller */ -static struct mtd_info *nsmtd; +static struct mtd_info *nsmtds; static int nandsim_debugfs_show(struct seq_file *m, void *private) { @@ -2221,6 +2222,9 @@ static void ns_nand_read_buf(struct mtd_info *mtd, u_char *buf, int len) return; } +#define NSMTD_SIZE (sizeof(struct mtd_info) + sizeof(struct nand_chip) + \ + sizeof(struct nandsim)) + /* * Module initialization function */ @@ -2228,7 +2232,8 @@ static int __init ns_init_module(void) { struct nand_chip *chip; struct nandsim *nand; - int retval = -ENOMEM, i; + struct mtd_info *nsmtd = NULL; + int retval = -ENOMEM, i, mtd_id; if (bus_width != 8 && bus_width != 16) { NS_ERR("wrong bus width (%d), use only 8 or 16\n", bus_width); @@ -2236,153 +2241,156 @@ static int __init ns_init_module(void) } /* Allocate and initialize mtd_info, nand_chip and nandsim structures */ - nsmtd = kzalloc(sizeof(struct mtd_info) + sizeof(struct nand_chip) - + sizeof(struct nandsim), GFP_KERNEL); - if (!nsmtd) { + nsmtds = kzalloc(NSMTD_SIZE * dev_count, GFP_KERNEL); + if (!nsmtds) { NS_ERR("unable to allocate core structures.\n"); return -ENOMEM; } - chip = (struct nand_chip *)(nsmtd + 1); - nsmtd->priv = (void *)chip; - nand = (struct nandsim *)(chip + 1); - chip->priv = (void *)nand; - /* - * Register simulator's callbacks. - */ - chip->cmd_ctrl = ns_hwcontrol; - chip->read_byte = ns_nand_read_byte; - chip->dev_ready = ns_device_ready; - chip->write_buf = ns_nand_write_buf; - chip->read_buf = ns_nand_read_buf; - chip->read_word = ns_nand_read_word; - chip->ecc.mode = NAND_ECC_SOFT; - /* The NAND_SKIP_BBTSCAN option is necessary for 'overridesize' */ - /* and 'badblocks' parameters to work */ - chip->options |= NAND_SKIP_BBTSCAN; - - switch (bbt) { - case 2: - chip->bbt_options |= NAND_BBT_NO_OOB; - case 1: - chip->bbt_options |= NAND_BBT_USE_FLASH; - case 0: - break; - default: - NS_ERR("bbt has to be 0..2\n"); - retval = -EINVAL; - goto error; - } - /* - * Perform minimum nandsim structure initialization to handle - * the initial ID read command correctly - */ - if (id_bytes[6] != 0xFF || id_bytes[7] != 0xFF) - nand->geom.idbytes = 8; - else if (id_bytes[4] != 0xFF || id_bytes[5] != 0xFF) - nand->geom.idbytes = 6; - else if (id_bytes[2] != 0xFF || id_bytes[3] != 0xFF) - nand->geom.idbytes = 4; - else - nand->geom.idbytes = 2; - nand->regs.status = NS_STATUS_OK(nand); - nand->nxstate = STATE_UNKNOWN; - nand->options |= OPT_PAGE512; /* temporary value */ - memcpy(nand->ids, id_bytes, sizeof(nand->ids)); - if (bus_width == 16) { - nand->busw = 16; - chip->options |= NAND_BUSWIDTH_16; - } - - nsmtd->owner = THIS_MODULE; - - if ((retval = parse_weakblocks()) != 0) - goto error; - - if ((retval = parse_weakpages()) != 0) - goto error; - - if ((retval = parse_gravepages()) != 0) - goto error; - - retval = nand_scan_ident(nsmtd, 1, NULL); - if (retval) { - NS_ERR("cannot scan NAND Simulator device\n"); - if (retval > 0) - retval = -ENXIO; - goto error; - } - - if (bch) { - unsigned int eccsteps, eccbytes; - if (!mtd_nand_has_bch()) { - NS_ERR("BCH ECC support is disabled\n"); + for (mtd_id = 0; mtd_id < dev_count; mtd_id++) { + nsmtd = ((void *)nsmtds) + NSMTD_SIZE * mtd_id; + chip = (struct nand_chip *)(nsmtd + 1); + nsmtd->priv = (void *)chip; + nand = (struct nandsim *)(chip + 1); + chip->priv = (void *)nand; + + /* + * Register simulator's callbacks. + */ + chip->cmd_ctrl = ns_hwcontrol; + chip->read_byte = ns_nand_read_byte; + chip->dev_ready = ns_device_ready; + chip->write_buf = ns_nand_write_buf; + chip->read_buf = ns_nand_read_buf; + chip->read_word = ns_nand_read_word; + chip->ecc.mode = NAND_ECC_SOFT; + /* The NAND_SKIP_BBTSCAN option is necessary for 'overridesize' */ + /* and 'badblocks' parameters to work */ + chip->options |= NAND_SKIP_BBTSCAN; + + switch (bbt) { + case 2: + chip->bbt_options |= NAND_BBT_NO_OOB; + case 1: + chip->bbt_options |= NAND_BBT_USE_FLASH; + case 0: + break; + default: + NS_ERR("bbt has to be 0..2\n"); retval = -EINVAL; goto error; } - /* use 512-byte ecc blocks */ - eccsteps = nsmtd->writesize/512; - eccbytes = (bch*13+7)/8; - /* do not bother supporting small page devices */ - if ((nsmtd->oobsize < 64) || !eccsteps) { - NS_ERR("bch not available on small page devices\n"); - retval = -EINVAL; - goto error; + /* + * Perform minimum nandsim structure initialization to handle + * the initial ID read command correctly + */ + if (id_bytes[6] != 0xFF || id_bytes[7] != 0xFF) + nand->geom.idbytes = 8; + else if (id_bytes[4] != 0xFF || id_bytes[5] != 0xFF) + nand->geom.idbytes = 6; + else if (id_bytes[2] != 0xFF || id_bytes[3] != 0xFF) + nand->geom.idbytes = 4; + else + nand->geom.idbytes = 2; + nand->regs.status = NS_STATUS_OK(nand); + nand->nxstate = STATE_UNKNOWN; + nand->options |= OPT_PAGE512; /* temporary value */ + memcpy(nand->ids, id_bytes, sizeof(nand->ids)); + if (bus_width == 16) { + nand->busw = 16; + chip->options |= NAND_BUSWIDTH_16; } - if ((eccbytes*eccsteps+2) > nsmtd->oobsize) { - NS_ERR("invalid bch value %u\n", bch); - retval = -EINVAL; + + nsmtd->owner = THIS_MODULE; + + if ((retval = parse_weakblocks()) != 0) + goto error; + + if ((retval = parse_weakpages()) != 0) + goto error; + + if ((retval = parse_gravepages()) != 0) + goto error; + + retval = nand_scan_ident(nsmtd, 1, NULL); + if (retval) { + NS_ERR("cannot scan NAND Simulator device\n"); + if (retval > 0) + retval = -ENXIO; goto error; } - chip->ecc.mode = NAND_ECC_SOFT_BCH; - chip->ecc.size = 512; - chip->ecc.strength = bch; - chip->ecc.bytes = eccbytes; - NS_INFO("using %u-bit/%u bytes BCH ECC\n", bch, chip->ecc.size); - } - retval = nand_scan_tail(nsmtd); - if (retval) { - NS_ERR("can't register NAND Simulator\n"); - if (retval > 0) - retval = -ENXIO; - goto error; - } + if (bch) { + unsigned int eccsteps, eccbytes; + if (!mtd_nand_has_bch()) { + NS_ERR("BCH ECC support is disabled\n"); + retval = -EINVAL; + goto error; + } + /* use 512-byte ecc blocks */ + eccsteps = nsmtd->writesize/512; + eccbytes = (bch*13+7)/8; + /* do not bother supporting small page devices */ + if ((nsmtd->oobsize < 64) || !eccsteps) { + NS_ERR("bch not available on small page devices\n"); + retval = -EINVAL; + goto error; + } + if ((eccbytes*eccsteps+2) > nsmtd->oobsize) { + NS_ERR("invalid bch value %u\n", bch); + retval = -EINVAL; + goto error; + } + chip->ecc.mode = NAND_ECC_SOFT_BCH; + chip->ecc.size = 512; + chip->ecc.strength = bch; + chip->ecc.bytes = eccbytes; + NS_INFO("using %u-bit/%u bytes BCH ECC\n", bch, chip->ecc.size); + } + + retval = nand_scan_tail(nsmtd); + if (retval) { + NS_ERR("can't register NAND Simulator\n"); + if (retval > 0) + retval = -ENXIO; + goto error; + } - if (overridesize) { - uint64_t new_size = (uint64_t)nsmtd->erasesize << overridesize; - if (new_size >> overridesize != nsmtd->erasesize) { - NS_ERR("overridesize is too big\n"); - retval = -EINVAL; - goto err_exit; + if (overridesize) { + uint64_t new_size = (uint64_t)nsmtd->erasesize << overridesize; + if (new_size >> overridesize != nsmtd->erasesize) { + NS_ERR("overridesize is too big\n"); + retval = -EINVAL; + goto err_exit; + } + /* N.B. This relies on nand_scan not doing anything with the size before we change it */ + nsmtd->size = new_size; + chip->chipsize = new_size; + chip->chip_shift = ffs(nsmtd->erasesize) + overridesize - 1; + chip->pagemask = (chip->chipsize >> chip->page_shift) - 1; } - /* N.B. This relies on nand_scan not doing anything with the size before we change it */ - nsmtd->size = new_size; - chip->chipsize = new_size; - chip->chip_shift = ffs(nsmtd->erasesize) + overridesize - 1; - chip->pagemask = (chip->chipsize >> chip->page_shift) - 1; - } - if ((retval = setup_wear_reporting(nsmtd)) != 0) - goto err_exit; + if ((retval = setup_wear_reporting(nsmtd)) != 0) + goto err_exit; - if ((retval = nandsim_debugfs_create(nand)) != 0) - goto err_exit; + if ((retval = nandsim_debugfs_create(nand)) != 0) + goto err_exit; - if ((retval = init_nandsim(nsmtd)) != 0) - goto err_exit; + if ((retval = init_nandsim(nsmtd)) != 0) + goto err_exit; - if ((retval = chip->scan_bbt(nsmtd)) != 0) - goto err_exit; + if ((retval = chip->scan_bbt(nsmtd)) != 0) + goto err_exit; - if ((retval = parse_badblocks(nand, nsmtd)) != 0) - goto err_exit; + if ((retval = parse_badblocks(nand, nsmtd)) != 0) + goto err_exit; - /* Register NAND partitions */ - retval = mtd_device_register(nsmtd, &nand->partitions[0], - nand->nbparts); - if (retval != 0) - goto err_exit; + /* Register NAND partitions */ + retval = mtd_device_register(nsmtd, &nand->partitions[0], + nand->nbparts); + if (retval != 0) + goto err_exit; + } return 0; @@ -2392,7 +2400,19 @@ err_exit: for (i = 0;i < ARRAY_SIZE(nand->partitions); ++i) kfree(nand->partitions[i].name); error: - kfree(nsmtd); + /* Cleanup for already */ + while (mtd_id) { + nsmtd = ((void *)nsmtds) + NSMTD_SIZE * mtd_id; + nand = ((struct nand_chip *)nsmtd->priv)->priv; + + free_nandsim(nand); + nand_release(nsmtd); + for (i = 0;i < ARRAY_SIZE(nand->partitions); ++i) + kfree(nand->partitions[i].name); + mtd_id--; + } + + kfree(nsmtds); free_lists(); return retval; @@ -2405,15 +2425,20 @@ module_init(ns_init_module); */ static void __exit ns_cleanup_module(void) { - struct nandsim *ns = ((struct nand_chip *)nsmtd->priv)->priv; - int i; - - nandsim_debugfs_remove(ns); - free_nandsim(ns); /* Free nandsim private resources */ - nand_release(nsmtd); /* Unregister driver */ - for (i = 0;i < ARRAY_SIZE(ns->partitions); ++i) - kfree(ns->partitions[i].name); - kfree(nsmtd); /* Free other structures */ + struct nandsim *ns; + struct mtd_info *nsmtd; + int i, mtd_id; + + for (mtd_id = 0; mtd_id < dev_count; mtd_id++) { + nsmtd = ((void *)nsmtds) + NSMTD_SIZE * mtd_id; + ns = ((struct nand_chip *)nsmtd->priv)->priv; + nandsim_debugfs_remove(ns); + free_nandsim(ns); /* Free nandsim private resources */ + nand_release(nsmtd); /* Unregister driver */ + for (i = 0;i < ARRAY_SIZE(ns->partitions); ++i) + kfree(ns->partitions[i].name); + } + kfree(nsmtds); /* Free other structures */ free_lists(); } -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 3/7] mtd: mtd_raid: Init a new layer of MTD RAID 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 1/7] mtd: nandsim: remove the abuse of reference to globle variable Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 2/7] mtd: nansim: add support of multi devices Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 4/7] mtd: mtd_raid: introduce raid single implementation Dongsheng Yang ` (5 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang MTD RAID layer is a generic raid layer in mtd subsystem. we can implement different raid level in this framework. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- Documentation/ioctl/ioctl-number.txt | 1 + drivers/mtd/Kconfig | 2 + drivers/mtd/Makefile | 1 + drivers/mtd/mtd_raid/Kconfig | 12 + drivers/mtd/mtd_raid/Makefile | 3 + drivers/mtd/mtd_raid/core.c | 1000 ++++++++++++++++++++++++++++++++++ drivers/mtd/mtd_raid/mtd_raid.h | 264 +++++++++ drivers/mtd/mtd_raid/raid_io.c | 449 +++++++++++++++ include/uapi/mtd/mtd-raid-user.h | 33 ++ 9 files changed, 1765 insertions(+) create mode 100644 drivers/mtd/mtd_raid/Kconfig create mode 100644 drivers/mtd/mtd_raid/Makefile create mode 100644 drivers/mtd/mtd_raid/core.c create mode 100644 drivers/mtd/mtd_raid/mtd_raid.h create mode 100644 drivers/mtd/mtd_raid/raid_io.c create mode 100644 include/uapi/mtd/mtd-raid-user.h diff --git a/Documentation/ioctl/ioctl-number.txt b/Documentation/ioctl/ioctl-number.txt index df1b25e..e13d60f 100644 --- a/Documentation/ioctl/ioctl-number.txt +++ b/Documentation/ioctl/ioctl-number.txt @@ -334,3 +334,4 @@ Code Seq#(hex) Include File Comments 0xF6 all LTTng Linux Trace Toolkit Next Generation <mailto:mathieu.desnoyers@efficios.com> 0xFD all linux/dm-ioctl.h +0xFE all mtd/mtd-raid-user.h diff --git a/drivers/mtd/Kconfig b/drivers/mtd/Kconfig index a03ad29..63fcdbe 100644 --- a/drivers/mtd/Kconfig +++ b/drivers/mtd/Kconfig @@ -338,4 +338,6 @@ source "drivers/mtd/spi-nor/Kconfig" source "drivers/mtd/ubi/Kconfig" +source "drivers/mtd/mtd_raid/Kconfig" + endif # MTD diff --git a/drivers/mtd/Makefile b/drivers/mtd/Makefile index 99bb9a1..1de2202 100644 --- a/drivers/mtd/Makefile +++ b/drivers/mtd/Makefile @@ -34,3 +34,4 @@ obj-y += chips/ lpddr/ maps/ devices/ nand/ onenand/ tests/ obj-$(CONFIG_MTD_SPI_NOR) += spi-nor/ obj-$(CONFIG_MTD_UBI) += ubi/ +obj-$(CONFIG_MTD_RAID) += mtd_raid/ diff --git a/drivers/mtd/mtd_raid/Kconfig b/drivers/mtd/mtd_raid/Kconfig new file mode 100644 index 0000000..b70a68f --- /dev/null +++ b/drivers/mtd/mtd_raid/Kconfig @@ -0,0 +1,12 @@ +menuconfig MTD_RAID + tristate "MTD RAID Support" + depends on MTD + help + This is a module for MTD raid. There are two ways to use + it. One for multi-chips flash drivers, driver can call + mtd_raid_create() to create raid device in requested level. + The other for user, user can use problem of mtd_raid to + create raid device from flashes they have in system. + + More about raid: + <https://en.wikipedia.org/wiki/Standard_RAID_levels>. diff --git a/drivers/mtd/mtd_raid/Makefile b/drivers/mtd/mtd_raid/Makefile new file mode 100644 index 0000000..517149b4 --- /dev/null +++ b/drivers/mtd/mtd_raid/Makefile @@ -0,0 +1,3 @@ +obj-$(CONFIG_MTD_RAID) += mtd_raid.o + +mtd_raid-y += ioctl.o raid_single.o core.o raid0.o raid1.o raid_io.o diff --git a/drivers/mtd/mtd_raid/core.c b/drivers/mtd/mtd_raid/core.c new file mode 100644 index 0000000..69854b2 --- /dev/null +++ b/drivers/mtd/mtd_raid/core.c @@ -0,0 +1,1000 @@ +/* + * Part of MTD RAID + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +/* + * TODO + * - merge requests + * - To support writev + * - raid10 + * - raid5/6 + * + * This is the core part of mtd raid layer. We implemented the generic mtd_func + * in this part. each raid devices share the same generic mtd interfaces. There + * is only interfaces for each mtd operations, the real io with each flashes are + * handled by raid_io.c. + * + * There is a global list for all struct mtd_raid. It's a 2-D list, we give each + * raid level a list to manage all mtd_raids in this level. When a mtd raid device + * created, we need to register it into related list. When a mtd raid device to + * be destroyed, we need to unregister it from related list. + **/ + +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/fs.h> +#include <linux/major.h> +#include <linux/miscdevice.h> +#include <linux/sched.h> +#include <linux/kthread.h> +#include <linux/slab.h> +#include <mtd/mtd-raid-user.h> + +#include "mtd_raid.h" + +/* We have a raid_list for each raid level. */ +struct raid_list { + int count; + struct list_head head; +}; + +/* To protect mtd_raid_list. */ +static spinlock_t mtd_raid_list_lock; +static struct raid_list mtd_raid_list[MTD_RAID_LEVEL_MAX]; + +int mtd_raid_list_init(void) +{ + int i = 0; + + spin_lock_init(&mtd_raid_list_lock); + for (i = 0; i < MTD_RAID_LEVEL_MAX; i++) + INIT_LIST_HEAD(&mtd_raid_list[i].head); + + return 0; +} + +void mtd_raid_list_destroy(void) +{ + struct mtd_raid *mtd_raid, *next; + int i = 0; + + for (i = 0; i < MTD_RAID_LEVEL_MAX; i++) { + list_for_each_entry_safe(mtd_raid, next, &mtd_raid_list[i].head, node) { + mtd_raid_destroy(mtd_raid); + } + } +} + +int mtd_raid_list_register(enum mtd_raid_level raid_level, struct mtd_raid *mtd_raid) +{ + int raid_id = 0; + + spin_lock(&mtd_raid_list_lock); + list_add_tail(&mtd_raid->node, &mtd_raid_list[raid_level].head); + raid_id = ++mtd_raid_list[raid_level].count; + spin_unlock(&mtd_raid_list_lock); + + return raid_id; +} + +struct mtd_raid *mtd_raid_list_get(int mtd_num) +{ + struct mtd_raid *raid; + int i = 0; + + spin_lock(&mtd_raid_list_lock); + for (i = 0; i < MTD_RAID_LEVEL_MAX; i++) { + list_for_each_entry(raid, &mtd_raid_list[i].head, node) { + if (raid->mtd.index == mtd_num) { + spin_unlock(&mtd_raid_list_lock); + return raid; + } + } + } + spin_unlock(&mtd_raid_list_lock); + + return NULL; +} + +void mtd_raid_list_unregister(struct mtd_raid *mtd_raid) +{ + spin_lock(&mtd_raid_list_lock); + list_del(&mtd_raid->node); + spin_unlock(&mtd_raid_list_lock); +} + +/* MTD interfaces */ +/* Check the range of addr:len to see is that out of address space */ +int check_offs(struct mtd_info *mtd, loff_t addr, size_t len) +{ + /* Do not allow out of address space of device */ + if (addr < 0) { + pr_err("%s: From a negative address.\n", + __func__); + return -EINVAL; + } + + if (addr > mtd->size) { + pr_err("%s: From an address out of size.\n", + __func__); + return -EINVAL; + } + + if (len > mtd->size - addr) { + pr_err("%s: Read past size of mtd device.\n", + __func__); + return -EINVAL; + } + + return 0; +} + +/* It's more strict than check_offs(). It also check the ofs and len + * to see are they aligned with aligned_size. + **/ +int check_offs_aligned(struct mtd_info *mtd, loff_t ofs, uint64_t len, + size_t aligned_size) +{ + int ret = 0; + loff_t tmp_ofs = ofs; + loff_t tmp_len = len; + + /* Start address must align on given length */ + if (do_div(tmp_ofs, aligned_size)) { + pr_err("%s: Unaligned address\n", __func__); + return -EINVAL; + } + + /* Length must align on given length */ + if (do_div(tmp_len, aligned_size)) { + pr_err("%s: Length not aligned\n", + __func__); + return -EINVAL; + } + + /* Do not allow out of address space of device */ + ret = check_offs(mtd, ofs, len); + + return ret; +} + +/* Interface for mtd->_read() */ +int mtd_raid_read(struct mtd_info *mtd, loff_t from, size_t len, + size_t *retlen, u_char *buf) +{ + loff_t subdev_off; + int ret = 0, err = 0; + int devid, i_copy = 0; + size_t retsize, size; + struct mtd_raid_ctx *ctx; + struct mtd_raid_dev *subdev; + struct mtd_raid *raid = MTD_RAID(mtd); + struct mtd_raid_read_request *read_req; + struct mtd_raid_request *request; + + if (check_offs(mtd, from, len)) + return -EINVAL; + + ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); + if (!ctx) + return -ENOMEM; + + ret = mtd_raid_ctx_init(ctx); + if (ret) + goto free; + + while (len) { + err = raid->ops->logical_to_physical(raid, from, len, i_copy, + &devid, &subdev_off, &size); + if (err) { + ret = err; + goto out; + } + + subdev = &raid->devs[devid]; + err = mtd_raid_dev_read(ctx, subdev, subdev_off, size, &retsize, buf); + if (unlikely(err)) { + ret = err; + goto out; + } + + buf += retsize; + from += retsize; + len -= retsize; + } +wait: + ret = mtd_raid_ctx_wait(ctx); + if (ret) { + if (!list_empty(&ctx->error_list)) { + request = list_first_entry(&ctx->error_list, + struct mtd_raid_request, node); + read_req = READ_REQUEST(request); + ret = read_req->retval; + goto out; + } else if (!list_empty(&ctx->failed_list)) { + ret = -EBADMSG; + } else { + if (!ret) + ret = -EUCLEAN; + } + } else { + goto out; + } + + if (++i_copy >= raid->ncopies) + goto out; + + ret = mtd_raid_ctx_retry(ctx, i_copy); + if (ret) + goto out; + goto wait; +out: + mtd->ecc_stats.failed += ctx->failed; + mtd->ecc_stats.corrected += ctx->corrected; + /* Fill retlen */ + *retlen = 0; + list_for_each_entry(request, &ctx->all_list, node_all) { + read_req = READ_REQUEST(request); + if (read_req->retval && !mtd_is_bitflip_or_eccerr(read_req->retval)) + break; + *retlen += read_req->retlen; + } + mtd_raid_ctx_destroy(ctx); +free: + kfree(ctx); + + return ret; +} + +/* Interface for mtd->_write() */ +int mtd_raid_write(struct mtd_info *mtd, loff_t to, size_t len, + size_t * retlen, const u_char * buf) +{ + int err = 0; + int i = 0; + int devid; + loff_t subdev_off; + size_t retsize, size; + struct mtd_info *subdev = NULL; + struct mtd_raid *raid = MTD_RAID(mtd); + + if (!(mtd->flags & MTD_WRITEABLE)) + return -EROFS; + + if (check_offs(mtd, to, len)) + return -EINVAL; + + while (len) { + for (i = 0; i < raid->ncopies; i++) { + err = raid->ops->logical_to_physical(raid, to, len, i, + &devid, &subdev_off, &size); + if (err) + goto out; + + subdev = raid->devs[devid].mtd; + err = mtd_write(subdev, subdev_off, size, &retsize, buf); + if (unlikely(err)) + goto out; + } + + *retlen += retsize; + len -= retsize; + buf += retsize; + to += retsize; + } + +out: + return err; +} + +int mtd_raid_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops) +{ + size_t size, readlen = ops->len; + loff_t subdev_off; + struct mtd_oob_ops devops = *ops; + uint64_t oobsize; + struct mtd_raid_ctx *ctx; + struct mtd_raid_dev *subdev; + struct mtd_raid *raid = MTD_RAID(mtd); + struct mtd_raid_read_oob_request *read_oob_req; + struct mtd_raid_request *request; + int i_copy = 0, devid = 0, i = 0, ret = 0; + + /* Check parameters for reading oob */ + if (ops->datbuf && check_offs(mtd, from, ops->len)) + return -EINVAL; + + /* Get oobsize depending on mode */ + if (ops->mode == MTD_OPS_AUTO_OOB) + oobsize = mtd->oobavail; + else + oobsize = mtd->oobsize; + + /* Check ooboffs */ + if (devops.ooboffs >= oobsize) + return -EINVAL; + + /* Check len and from */ + oobsize -= devops.ooboffs; + if (devops.datbuf) { + readlen = devops.len; + } else { + readlen = DIV_ROUND_UP(devops.ooblen, oobsize); + readlen *= mtd->writesize; + } + + if (readlen > mtd->size - from) + return -EINVAL; + + ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); + if (!ctx) + return -ENOMEM; + + ret = mtd_raid_ctx_init(ctx); + if (ret) + goto free; + + while (readlen > 0) { + ret = raid->ops->logical_to_physical(raid, from, readlen, i, + &devid, &subdev_off, &size); + if (unlikely(ret)) + goto out; + + if (devops.datbuf) { + devops.len = size; + } else { + if (devops.ooblen > (size / mtd->writesize * oobsize)) + devops.ooblen = size / mtd->writesize * oobsize; + } + + /* Read data from subdev */ + subdev = &raid->devs[devid]; + ret = mtd_raid_dev_read_oob(ctx, subdev, subdev_off, size, &devops); + if (unlikely(ret)) + goto out; + + readlen -= size; + from += size; + + if (devops.datbuf) + devops.datbuf += devops.len; + + if (devops.oobbuf) + devops.oobbuf += devops.ooblen; + } + +wait: + ret = mtd_raid_ctx_wait(ctx); + if (ret) { + /* Not all request succeeded */ + if (!list_empty(&ctx->error_list)) { + request = list_first_entry(&ctx->error_list, + struct mtd_raid_request, node); + read_oob_req = READ_OOB_REQUEST(request); + ret = read_oob_req->retval; + goto out; + } else if (!list_empty(&ctx->failed_list)) { + ret = -EBADMSG; + } else { + if (!ret) + ret = -EUCLEAN; + } + } else { + goto out; + } + + if (++i_copy >= raid->ncopies) + goto out; + + ret = mtd_raid_ctx_retry(ctx, i_copy); + if (ret) + goto out; + goto wait; +out: + mtd->ecc_stats.failed += ctx->failed; + mtd->ecc_stats.corrected += ctx->corrected; + /* Fill retlen */ + ops->retlen = ops->oobretlen = 0; + list_for_each_entry(request, &ctx->all_list, node_all) { + read_oob_req = READ_OOB_REQUEST(request); + if (read_oob_req->retval && !mtd_is_bitflip_or_eccerr(read_oob_req->retval)) + break; + ops->retlen += read_oob_req->ops.retlen; + ops->oobretlen += read_oob_req->ops.oobretlen; + } + mtd_raid_ctx_destroy(ctx); +free: + kfree(ctx); + + return ret; +} + +int mtd_raid_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + struct mtd_oob_ops devops = *ops; + uint64_t oobsize; + int devid = 0, i = 0, ret = 0; + loff_t subdev_off; + size_t size, writelen = ops->len; + struct mtd_info *subdev = NULL; + + if (!(mtd->flags & MTD_WRITEABLE)) + return -EROFS; + + if (ops->datbuf && check_offs(mtd, to, ops->len)) + return -EINVAL; + + ops->retlen = ops->oobretlen = 0; + + if (ops->mode == MTD_OPS_AUTO_OOB) + oobsize = mtd->oobavail; + else + oobsize = mtd->oobsize; + + if (to < 0 || to > mtd->size) + return -EINVAL; + + if (devops.ooboffs >= oobsize) + return -EINVAL; + + oobsize -= devops.ooboffs; + + if (devops.datbuf) { + writelen = devops.len; + } else { + writelen = DIV_ROUND_UP(devops.ooblen, oobsize); + writelen *= mtd->writesize; + } + + if (writelen > mtd->size - to) + return -EINVAL; + + while (writelen > 0) { + for (i = 0; i < raid->ncopies; i++) { + ret = raid->ops->logical_to_physical(raid, to, writelen, i, + &devid, &subdev_off, &size); + if (unlikely(ret)) + goto out; + + if (devops.datbuf) { + devops.len = size; + } else { + if (devops.ooblen > (size / mtd->writesize * oobsize)) + devops.ooblen = size / mtd->writesize * oobsize; + } + + subdev = raid->devs[devid].mtd; + ret = mtd_write_oob(subdev, subdev_off, &devops); + if (ret) + goto out; + + writelen -= size; + to += size; + + if (devops.datbuf) { + devops.datbuf += devops.len; + ops->retlen += devops.len; + } + + if (devops.oobbuf) { + devops.oobbuf += devops.ooblen; + ops->oobretlen += devops.ooblen; + } + } + } +out: + return ret; +} + +//TODO make it async and paralleled +/* lblock means logical block */ +static int raid_erase_lblock(struct mtd_raid *raid, struct erase_info *instr) +{ + int devid; + size_t size; + loff_t subdev_off; + int i, icopy = 0; + int err = 0, ret = 0; + uint64_t addr, len; + struct mtd_raid_dev *subdev = NULL; + + addr = instr->addr; + len = 0; + for (i = 0; i < raid->npebs_per_leb; i++) { + for (icopy = 0; icopy < raid->ncopies; icopy++) { + err = raid->ops->logical_to_physical(raid, addr, len, icopy, &devid, &subdev_off, &size); + if (err) + goto out; + + subdev = &raid->devs[devid]; + if (!(subdev->mtd->flags & MTD_WRITEABLE)) { + ret = -EROFS; + goto out; + } + + instr->addr = subdev_off; + + ret = mtd_raid_dev_erase(subdev, instr); + if (ret) + goto out; + + if (instr->state != MTD_ERASE_DONE) { + ret = -EIO; + goto out; + } + } + + addr += raid->substripe_size; + } + +out: + return ret; +} + +int mtd_raid_erase(struct mtd_info *mtd, struct erase_info *instr) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + struct erase_info *erase; + loff_t logical; + loff_t length; + int err = 0; + + if (!(mtd->flags & MTD_WRITEABLE)) + return -EROFS; + + if (check_offs_aligned(mtd, instr->addr, instr->len, mtd->erasesize)) + return -EINVAL; + + instr->fail_addr = MTD_FAIL_ADDR_UNKNOWN; + + erase = kmalloc(sizeof(struct erase_info), GFP_KERNEL); + if (!erase) + return -ENOMEM; + + *erase = *instr; + logical = instr->addr; + length = instr->len; + + err = 0; + while (length > 0) { + erase->addr = logical; + err = raid_erase_lblock(raid, erase); + if (err) + break; + + logical += raid->mtd.erasesize; + length -= raid->mtd.erasesize; + } + + instr->state = erase->state; + kfree(erase); + + if (err) + return err; + + if (instr->callback) + instr->callback(instr); + + return 0; + +} + +void mtd_raid_sync(struct mtd_info *mtd) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int i; + + for (i = 0; i < raid->dev_count; i++) { + struct mtd_info *subdev = raid->devs[i].mtd; + mtd_sync(subdev); + } +} + +int mtd_raid_suspend(struct mtd_info *mtd) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int i, err = 0; + + for (i = 0; i < raid->dev_count; i++) { + struct mtd_info *subdev = raid->devs[i].mtd; + if ((err = mtd_suspend(subdev)) < 0) + return err; + } + return err; +} + +void mtd_raid_resume(struct mtd_info *mtd) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int i; + + for (i = 0; i < raid->dev_count; i++) { + struct mtd_info *subdev = raid->devs[i].mtd; + mtd_resume(subdev); + } +} + +int mtd_raid_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int ret = 0, err = 0, i = 0; + int devid; + loff_t subdev_off; + size_t size; + struct mtd_info *subdev = NULL; + + while (len) { + for (i = 0; i < raid->ncopies; i++) { + err = raid->ops->logical_to_physical(raid, ofs, len, i, &devid, &subdev_off, &size); + if (err) { + ret = err; + goto out; + } + + subdev = raid->devs[devid].mtd; + err = mtd_lock(subdev, subdev_off, size); + if (unlikely(err)) { + ret = err; + goto out; + } + } + + len -= size; + ofs += size; + } +out: + return ret; +} + +int mtd_raid_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int ret = 0, err = 0, i = 0; + int devid; + loff_t subdev_off; + size_t size; + struct mtd_info *subdev = NULL; + + while (len) { + for (i = 0; i < raid->ncopies; i++) { + err = raid->ops->logical_to_physical(raid, ofs, len, i, &devid, &subdev_off, &size); + if (err) { + ret = err; + goto out; + } + + subdev = raid->devs[devid].mtd; + err = mtd_unlock(subdev, subdev_off, size); + if (unlikely(err)) { + ret = err; + goto out; + } + } + + len -= size; + ofs += size; + } + +out: + return ret; +} + +int mtd_raid_block_isbad(struct mtd_info *mtd, loff_t ofs) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int i = 0; + loff_t from = ofs, subdev_off; + size_t len; + int devid; + size_t size; + struct mtd_info *subdev = NULL; + + if (!mtd_can_have_bb(raid->devs[0].mtd)) + return 0; + + ofs -= do_div(from, mtd->erasesize); + from = ofs; + len = mtd->erasesize; + while (len) { + for (i = 0; i < raid->ncopies; i++) { + raid->ops->logical_to_physical(raid, from, len, i, &devid, &subdev_off, &size); + + subdev = raid->devs[devid].mtd; + if (mtd_block_isbad(subdev, subdev_off)) + return 1; + } + + len -= size; + from += size; + } + + return 0; +} + +int mtd_raid_block_markbad(struct mtd_info *mtd, loff_t ofs) +{ + struct mtd_raid *raid = MTD_RAID(mtd); + int i, err = 0; + loff_t from = ofs; + size_t len; + int devid; + loff_t subdev_off; + size_t size; + struct mtd_info *subdev = NULL; + + ofs -= do_div(from, mtd->erasesize); + from = ofs; + len = mtd->erasesize; + while (len) { + for (i = 0; i < raid->ncopies; i++) { + err = raid->ops->logical_to_physical(raid, from, len, i, &devid, &subdev_off, &size); + if (err) + goto out; + + subdev = raid->devs[devid].mtd; + err = mtd_block_markbad(subdev, subdev_off); + if (err) + goto out; + else + mtd->ecc_stats.badblocks++; + } + + len -= size; + from += size; + } + +out: + return err; +} + +int mtd_raid_init(struct mtd_raid *raid, int *mtd_nums, int dev_count, size_t substripe_size) +{ + struct mtd_info *subdev = NULL; + struct mtd_info *mtd = NULL; + int ret = 0; + int i = 0; + + INIT_LIST_HEAD(&raid->node); + raid->substripe_size = substripe_size; + raid->dev_count = dev_count; + + for (i = 0; i < dev_count; i++) { + subdev = get_mtd_device(NULL, mtd_nums[i]); + if (IS_ERR(subdev)) { + ret = PTR_ERR(subdev); + pr_err("error: Cannot get MTD device. mtd_num: %d\n", mtd_nums[i]); + goto out; + } + raid->devs[i].mtd = subdev; + raid->devs[i].id = i; + } + + mtd = &raid->mtd; + subdev = raid->devs[0].mtd; + + if (raid->substripe_size == 0) + raid->substripe_size = subdev->writesize; + + mtd->owner = THIS_MODULE; + mtd->type = subdev->type; + mtd->flags = subdev->flags; + mtd->writesize = subdev->writesize; + mtd->writebufsize = subdev->writebufsize; + mtd->subpage_sft = subdev->subpage_sft; + mtd->oobsize = subdev->oobsize; + mtd->oobavail = subdev->oobavail; + mtd->ecclayout = subdev->ecclayout; + + mtd->_erase = mtd_raid_erase; + mtd->_read = mtd_raid_read; + mtd->_write = mtd_raid_write; + mtd->_sync = mtd_raid_sync; + mtd->_lock = mtd_raid_lock; + mtd->_unlock = mtd_raid_unlock; + mtd->_suspend = mtd_raid_suspend; + mtd->_resume = mtd_raid_resume; + + if (subdev->_read_oob) + mtd->_read_oob = mtd_raid_read_oob; + if (subdev->_write_oob) + mtd->_write_oob = mtd_raid_write_oob; + if (subdev->_block_isbad) + mtd->_block_isbad = mtd_raid_block_isbad; + if (subdev->_block_markbad) + mtd->_block_markbad = mtd_raid_block_markbad; + + for (i = 1; i < dev_count; i++) { + if (mtd->flags != raid->devs[i].mtd->flags) { + /* + * Expect all flags except MTD_WRITEABLE to be + * equal on all subdevices. + */ + if ((mtd->flags ^ raid->devs[i].mtd-> + flags) & ~MTD_WRITEABLE) { + printk("Incompatible device flags on \"%s\"\n", + raid->devs[i].mtd->name); + ret = -EINVAL; + goto out; + } else { + /* if writeable attribute differs, + make super device writeable */ + mtd->flags |= + raid->devs[i].mtd->flags & MTD_WRITEABLE; + } + } + + if (mtd->writesize != raid->devs[i].mtd->writesize || + mtd->subpage_sft != raid->devs[i].mtd->subpage_sft || + mtd->oobsize != raid->devs[i].mtd->oobsize || + !mtd->_read_oob != !raid->devs[i].mtd->_read_oob || + !mtd->_write_oob != !raid->devs[i].mtd->_write_oob) { + printk("Incompatible OOB or ECC data on \"%s\"\n", + raid->devs[i].mtd->name); + ret = -EINVAL; + goto out; + } + + if (mtd->writebufsize != raid->devs[i].mtd->writebufsize) { + pr_err("Incompatible writebufsize on \"%s\"", + raid->devs[i].mtd->name); + ret = -EINVAL; + goto out; + } + } + + if (raid->ops->init) { + ret = raid->ops->init(raid, dev_count, substripe_size); + if (ret) + goto out; + } + + for (i = 0; i < dev_count; i++) { + /* + * Init bg thread for each raid_dev to handle io requests. + */ + INIT_LIST_HEAD(&raid->devs[i].list); + raid->devs[i].thread = kthread_create(mtd_raid_dev_thread, &raid->devs[i], + "%s_thread_%d", raid->name, i); + } + + return 0; +out: + return ret; +} + +int mtd_raid_destroy(struct mtd_raid *raid); + +int mtd_raid_create(enum mtd_raid_level raid_level, int *mtd_nums, int dev_count, int substripe_size) +{ + int ret = 0; + struct mtd_raid *raid = NULL; + + switch (raid_level){ + case MTD_RAID_LEVEL_SINGLE: + { + raid = mtd_raid_single_create(mtd_nums, dev_count, substripe_size); + if (!raid) { + pr_err("MTD RAID: Failed to create raid single device."); + ret = -EINVAL; + goto out; + } + break; + } + case MTD_RAID_LEVEL_RAID0: + { + raid = mtd_raid0_create(mtd_nums, dev_count, substripe_size); + if (!raid) { + pr_err("MTD RAID: Failed to create raid0 device."); + ret = -EINVAL; + goto out; + } + break; + } + case MTD_RAID_LEVEL_RAID1: + { + raid = mtd_raid1_create(mtd_nums, dev_count, substripe_size); + if (!raid) { + pr_err("MTD RAID: Failed to create raid1 device."); + ret = -EINVAL; + goto out; + } + break; + } + default: + pr_err("MTD RAID: Unsupported raid level: %d.", raid_level); + ret = -ENOTSUPP; + goto out; + } + + ret = mtd_raid_init(raid, mtd_nums, dev_count, substripe_size); + if (ret) + goto destroy; + + return mtd_device_register(&raid->mtd, NULL, 0); + +destroy: + mtd_raid_destroy(raid); +out: + return ret; +} + +int mtd_raid_destroy(struct mtd_raid *raid) +{ + int i = 0; + int ret = 0; + + ret = mtd_device_unregister(&raid->mtd); + if (ret) + goto out; + + for (i = 0; i < raid->dev_count; i++) { + if (raid->devs[i].mtd) + put_mtd_device(raid->devs[i].mtd); + + if (raid->devs[i].thread) + kthread_stop(raid->devs[i].thread); + } + + if (raid->ops->destroy) + raid->ops->destroy(raid); +out: + return ret; +} + +static struct miscdevice mtd_raid_ctrl_cdev = { + .minor = MISC_DYNAMIC_MINOR, + .name = "mtd_raid_ctrl", + .fops = &mtd_raid_ctrl_cdev_operations, +}; + +static int __init init_mtd_raid(void) +{ + int err = 0; + + err = mtd_raid_list_init(); + if (err) + goto out; + + err = misc_register(&mtd_raid_ctrl_cdev); + if (err) { + pr_err("MTD RAID error: cannot register device"); + goto out; + } + + return 0; +out: + pr_err("MTD RAID error: cannot initialize MTD RAID, error %d", err); + return err; +} + +static void __exit cleanup_mtd_raid(void) +{ + misc_deregister(&mtd_raid_ctrl_cdev); + mtd_raid_list_destroy(); +} + +module_init(init_mtd_raid); +module_exit(cleanup_mtd_raid); + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Dongsheng Yang <yangds.fnst@cn.fujitsu.com>"); +MODULE_DESCRIPTION("Support for MTD RAID"); diff --git a/drivers/mtd/mtd_raid/mtd_raid.h b/drivers/mtd/mtd_raid/mtd_raid.h new file mode 100644 index 0000000..5300aa1 --- /dev/null +++ b/drivers/mtd/mtd_raid/mtd_raid.h @@ -0,0 +1,264 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +#ifndef __MTD_RAID_H +#define __MTD_RAID_H + +#include <linux/mtd/mtd.h> +#include <mtd/mtd-abi.h> +#include <mtd/mtd-raid-user.h> + +struct mtd_raid; + +/* + * Operations for different mtd raid structure, it + * differs in different raid level. + **/ +struct mtd_raid_operations { + int (*init)(struct mtd_raid *, int, size_t); + void (*destroy)(struct mtd_raid *); + + /* logical_to_physical is the core function in mtd raid, + * when we get a logical address from user, + * we will call ->logical_to_physical() to get the phisical + * devid, address and length.*/ + int (*logical_to_physical)(struct mtd_raid *, loff_t, size_t, int, int *, loff_t *, size_t *); + int (*physical_to_logical)(struct mtd_raid *, int, loff_t, size_t, loff_t *, size_t *); +}; + +/* + * struct mtd_raid_dev, one mtd_raid_dev represent a mtd device + * int the raid array. + **/ +struct mtd_raid_dev { + int id; + struct mtd_raid *raid; + struct mtd_info *mtd; + + spinlock_t lock; + struct list_head list; + struct task_struct *thread; +}; + +/* + * Context of a IO, we will send lots of io request to + * mtd_raid_dev, and mtd_raid_dev will do it in thread. + * But in order to track the status of each request, we + * will register them in a mtd_raid_ctx. + */ +struct mtd_raid_ctx { + spinlock_t lock; + struct list_head all_list; + + struct list_head submit_list; + struct list_head complete_list; + struct list_head failed_list; + struct list_head corrected_list; + struct list_head error_list; + + unsigned int failed; + unsigned int corrected; + unsigned int errored; + + struct task_struct *wait; +}; + + +/* + * Type for each io request; + */ +enum mtd_raid_request_type { + MTD_RAID_REQ_READ = 0, + MTD_RAID_REQ_READ_OOB +}; + +struct mtd_raid_request; +typedef void request_func_t(struct mtd_raid_request *); +typedef void end_func_t(struct mtd_raid_request *); +typedef int retry_func_t(struct mtd_raid_request *request, int i_copy); + +struct mtd_raid_request { + enum mtd_raid_request_type type; + struct mtd_raid_ctx *ctx; + struct mtd_raid_dev *raid_dev; + + struct list_head node; + struct list_head node_all; + struct list_head node_request; + + /* Main action of this request. */ + request_func_t *func; + /* + * If this request failed, how to retry. NULL means + * don't retry for this request. + */ + retry_func_t *retry_func; + /* + * When we want to destroy this request, what we need + * to do. + */ + end_func_t *end_func; +}; + +/* + * Request for mtd_read. + */ +struct mtd_raid_read_request { + struct mtd_raid_request request; + + loff_t from; + size_t len; + size_t retlen; + u_char *buf; + int retval; +}; + +/* + * Request for mtd_read_oob + */ +struct mtd_raid_read_oob_request { + struct mtd_raid_request request; + + loff_t from; + size_t len; + struct mtd_oob_ops ops; + int retval; +}; + +/* + * structure to represent a RAID device + **/ +struct mtd_raid { + char name[32]; + int ncopies; + int dev_count; + int npebs_per_leb; + int substripe_size; + /* + * This is the "superblock" for this RAID device. + * We will fill up it and register it. + **/ + struct mtd_info mtd; + struct list_head node; + const struct mtd_raid_operations *ops; + enum mtd_raid_level raid_level; + struct mtd_raid_dev devs[0]; +}; + +struct mtd_raid_single { + /* + * Please make the raid to be the last member, + * because we will alloc devs appending this structure. + */ + struct mtd_raid raid; +}; + +struct mtd_raid0 { + // XXX Add reada support here. + + /* + * Please make the raid to be the last member, + * because we will alloc devs appending this structure. + */ + struct mtd_raid raid; +}; + +struct mtd_raid1 { + /* + * Please make the raid to be the last member, + * because we will alloc devs appending this structure. + */ + struct mtd_raid raid; +}; + +/* Macros to get specified request pointers from generic request */ +#define READ_REQUEST(req) \ + container_of(req, struct mtd_raid_read_request, request) + +#define READ_OOB_REQUEST(req) \ + container_of(req, struct mtd_raid_read_oob_request, request) + +/* Macros to get specified mtd_raid pointers from mtd_info pointer */ +#define MTD_RAID(mtd) \ + container_of(mtd, struct mtd_raid, mtd) + +#define MTD_RAID_SINGLE(mtd_raid) \ + container_of(mtd_raid, struct mtd_raid_single, raid) + +#define MTD_RAID_RAID0(mtd_raid) \ + container_of(mtd_raid, struct mtd_raid0, raid) + +#define MTD_RAID_RAID1(mtd_raid) \ + container_of(mtd_raid, struct mtd_raid1, raid) + +/* ioctl.c */ +extern const struct file_operations mtd_raid_ctrl_cdev_operations; + +/* core.c */ +int mtd_raid_list_init(void); +void mtd_raid_list_destroy(void); +int mtd_raid_list_register(enum mtd_raid_level raid_level, struct mtd_raid *mtd_raid); +struct mtd_raid *mtd_raid_list_get(int mtd_num); +void mtd_raid_list_unregister(struct mtd_raid *mtd_raid); + +int mtd_raid_create(enum mtd_raid_level raid_level, int *mtd_nums, int dev_count, int substripe_size); +int mtd_raid_destroy(struct mtd_raid *mtd_raid); + +int mtd_raid_read(struct mtd_info *mtd, loff_t from, size_t len, size_t * retlen, u_char * buf); +int mtd_raid_write(struct mtd_info *mtd, loff_t to, size_t len, size_t * retlen, const u_char * buf); +int mtd_raid_erase(struct mtd_info *mtd, struct erase_info *instr); +int mtd_raid_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops); +int mtd_raid_unlock(struct mtd_info *mtd, loff_t ofs, uint64_t len); +int mtd_raid_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len); +void mtd_raid_resume(struct mtd_info *mtd); +int mtd_raid_suspend(struct mtd_info *mtd); +void mtd_raid_sync(struct mtd_info *mtd); +int mtd_raid_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops); +int mtd_raid_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops); +int mtd_raid_block_isbad(struct mtd_info *mtd, loff_t ofs); +int mtd_raid_block_markbad(struct mtd_info *mtd, loff_t ofs); + +/* raid_io.c */ +int mtd_raid_dev_thread(void *u); +int mtd_raid_ctx_init(struct mtd_raid_ctx* ctx); +int mtd_raid_ctx_wait(struct mtd_raid_ctx *ctx); +void mtd_raid_ctx_destroy(struct mtd_raid_ctx *ctx); +int mtd_raid_ctx_retry(struct mtd_raid_ctx *ctx, int i_copy); + +int mtd_raid_dev_read(struct mtd_raid_ctx *ctx, struct mtd_raid_dev *raid_dev, + loff_t from, size_t len, size_t *retlen, u_char *buf); +int mtd_raid_dev_read_oob(struct mtd_raid_ctx *ctx, struct mtd_raid_dev *raid_dev, + loff_t from, size_t len, struct mtd_oob_ops *ops); +int mtd_raid_dev_erase(struct mtd_raid_dev *raid_dev, struct erase_info *erase); + +/* raid_single.c */ +extern const struct mtd_raid_operations mtd_raid_single_ops; +struct mtd_raid *mtd_raid_single_create(int *mtd_nums, int dev_count, size_t substripe_size); + +/* raid0.c */ +extern const struct mtd_raid_operations mtd_raid0_ops; +struct mtd_raid *mtd_raid0_create(int *mtd_nums, int dev_count, size_t substripe_size); + +/* raid1.c */ +extern const struct mtd_raid_operations mtd_raid1_ops; +struct mtd_raid *mtd_raid1_create(int *mtd_nums, int dev_count, size_t substripe_size); + +#endif /* __MTD_RAID_H */ diff --git a/drivers/mtd/mtd_raid/raid_io.c b/drivers/mtd/mtd_raid/raid_io.c new file mode 100644 index 0000000..71a5c6b --- /dev/null +++ b/drivers/mtd/mtd_raid/raid_io.c @@ -0,0 +1,449 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +/* + * This file handles the all io-related work. + */ + +#include <linux/freezer.h> +#include <linux/kthread.h> +#include <linux/slab.h> + +#include "mtd_raid.h" + +/** + * Context related operations: + * mtd_raid_ctx_init() --> Init a ctx + * --> Attach requests to ctx + * mtd_raid_ctx_wait() --> Wait all requests end + * mtd_raid_ctx_retry() --> Retry failed requests + * mtd_raid_ctx_destroy() --> Destroy the ctx + */ +int mtd_raid_ctx_init(struct mtd_raid_ctx* ctx) +{ + spin_lock_init(&ctx->lock); + INIT_LIST_HEAD(&ctx->all_list); + INIT_LIST_HEAD(&ctx->submit_list); + INIT_LIST_HEAD(&ctx->complete_list); + INIT_LIST_HEAD(&ctx->failed_list); + INIT_LIST_HEAD(&ctx->corrected_list); + INIT_LIST_HEAD(&ctx->error_list); + + ctx->failed = ctx->corrected = ctx->errored = 0; + ctx->wait = current; + + return 0; +} + +int mtd_raid_ctx_wait(struct mtd_raid_ctx *ctx) +{ + int ret = 0; + + while (1) { + spin_lock(&ctx->lock); + if (list_empty(&ctx->submit_list)) { + ret = ctx->failed + ctx->corrected + ctx->errored; + spin_unlock(&ctx->lock); + return ret; + } + + spin_unlock(&ctx->lock); + set_current_state(TASK_INTERRUPTIBLE); + schedule(); + set_current_state(TASK_RUNNING); + + cond_resched(); + } +} + +void mtd_raid_ctx_destroy(struct mtd_raid_ctx *ctx) +{ + struct mtd_raid_request *request, *next; + + list_for_each_entry_safe(request, next, &ctx->all_list, node_all) { + if (request->end_func) + request->end_func(request); + } +} + +int mtd_raid_ctx_retry(struct mtd_raid_ctx *ctx, int i_copy) +{ + struct mtd_raid_request *request; + int ret = 0; + + list_for_each_entry(request, &ctx->failed_list, node) { + ret = request->retry_func(request, i_copy); + if (ret) + goto out; + spin_lock(&ctx->lock); + list_move_tail(&request->node, &ctx->submit_list); + spin_unlock(&ctx->lock); + } +out: + return ret; +} + +/* + * hooks for each type of request + * */ +static void read_req_func(struct mtd_raid_request *request) +{ + struct mtd_raid_read_request *read_req; + struct mtd_raid_ctx *ctx; + struct mtd_raid_dev *raid_dev; + int ret = 0; + + read_req = READ_REQUEST(request); + raid_dev = request->raid_dev; + ret = mtd_read(raid_dev->mtd, read_req->from, read_req->len, + &read_req->retlen, read_req->buf); + + read_req->retval = ret; + ctx = request->ctx; + spin_lock(&ctx->lock); + list_del_init(&request->node); + if (unlikely(ret)) { + if (mtd_is_eccerr(ret)) { + ctx->failed++; + list_add_tail(&request->node, &ctx->failed_list); + } else if (mtd_is_bitflip(ret)) { + ctx->corrected++; + list_add_tail(&request->node, &ctx->corrected_list); + } else { + ctx->errored++; + list_add_tail(&request->node, &ctx->error_list); + } + } else { + list_add_tail(&request->node, &ctx->complete_list); + } + spin_unlock(&ctx->lock); + wake_up_process(ctx->wait); + + return; +} + +static void read_oob_req_func(struct mtd_raid_request *request) +{ + struct mtd_raid_read_oob_request *read_oob_req; + struct mtd_raid_ctx *ctx; + struct mtd_raid_dev *raid_dev; + int ret = 0; + + read_oob_req = READ_OOB_REQUEST(request); + raid_dev = request->raid_dev; + ret = mtd_read_oob(raid_dev->mtd, read_oob_req->from, &read_oob_req->ops); + + read_oob_req->retval = ret; + ctx = request->ctx; + spin_lock(&ctx->lock); + list_del_init(&request->node); + if (unlikely(ret)) { + if (mtd_is_eccerr(ret)) { + ctx->failed++; + list_add_tail(&request->node, &ctx->failed_list); + } else if (mtd_is_bitflip(ret)) { + ctx->corrected++; + list_add_tail(&request->node, &ctx->corrected_list); + } else { + ctx->errored++; + list_add_tail(&request->node, &ctx->error_list); + } + } else { + list_add_tail(&request->node, &ctx->complete_list); + } + spin_unlock(&ctx->lock); + wake_up_process(ctx->wait); + + return; +} + +static int read_req_retry_func(struct mtd_raid_request *request, int i_copy) +{ + struct mtd_raid *mtd_raid; + struct mtd_raid_read_request *read_req; + loff_t address, subdev_off; + size_t length, size; + int devid, ret = 0; + + mtd_raid = request->raid_dev->raid; + read_req = READ_REQUEST(request); + if (!mtd_raid->ops->physical_to_logical || !mtd_raid->ops->logical_to_physical) + return -EINVAL; + + subdev_off = read_req->from; + size = read_req->len; + devid = request->raid_dev->id; + ret = mtd_raid->ops->physical_to_logical(mtd_raid, devid, subdev_off, size, + &address, &length); + if (ret) + goto out; + + ret = mtd_raid->ops->logical_to_physical(mtd_raid, address, length, i_copy, + &devid, &subdev_off, &size); + if (ret) + goto out; + + /* Fill request with the address of new copy */ + request->raid_dev = &mtd_raid->devs[devid]; + read_req->from = subdev_off; + read_req->len = size; +out: + return ret; +} + +static int read_oob_req_retry_func(struct mtd_raid_request *request, int i_copy) +{ + struct mtd_raid *mtd_raid; + struct mtd_raid_read_oob_request *read_oob_req; + loff_t address, subdev_off; + size_t length, size; + int devid, ret = 0; + + mtd_raid = request->raid_dev->raid; + read_oob_req = READ_OOB_REQUEST(request); + if (!mtd_raid->ops->physical_to_logical || !mtd_raid->ops->logical_to_physical) + return -EINVAL; + + subdev_off = read_oob_req->from; + size = read_oob_req->len; + devid = request->raid_dev->id; + ret = mtd_raid->ops->physical_to_logical(mtd_raid, devid, subdev_off, size, + &address, &length); + if (ret) + goto out; + + ret = mtd_raid->ops->logical_to_physical(mtd_raid, address, length, i_copy, + &devid, &subdev_off, &size); + if (ret) + goto out; + + /* Fill request with the address of new copy */ + request->raid_dev = &mtd_raid->devs[devid]; + read_oob_req->from = subdev_off; + read_oob_req->len = size; +out: + return ret; +} + +/* Generic end_func for request */ +static void request_end_func(struct mtd_raid_request *request) +{ + struct mtd_raid_ctx *ctx = NULL; + struct mtd_raid_dev *raid_dev = NULL; + + ctx = request->ctx; + spin_lock(&ctx->lock); + list_del(&request->node); + list_del(&request->node_all); + spin_unlock(&ctx->lock); + + raid_dev = request->raid_dev; + spin_lock(&raid_dev->lock); + list_del(&request->node_request); + spin_unlock(&raid_dev->lock); +} + +static void read_req_end_func(struct mtd_raid_request *request) +{ + struct mtd_raid_read_request *read_req; + + read_req = READ_REQUEST(request); + request_end_func(request); + kfree(read_req); +} + +static void read_oob_req_end_func(struct mtd_raid_request *request) +{ + struct mtd_raid_read_oob_request *read_oob_req; + + read_oob_req = READ_OOB_REQUEST(request); + request_end_func(request); + kfree(read_oob_req); +} + +/** + * Thread for each raid_dev. + * + * It get requests from raid_dev->list and do the + * requested work, until raid_dev to be empty. Then + * go to sleep. + */ +int mtd_raid_dev_thread(void *u) +{ + struct mtd_raid_dev *raid_dev = u; + struct mtd_raid_request *request; + + set_freezable(); + for (;;) { + if (kthread_should_stop()) + break; + + if (try_to_freeze()) + continue; + + spin_lock(&raid_dev->lock); + if (list_empty(&raid_dev->list)) { + set_current_state(TASK_INTERRUPTIBLE); + spin_unlock(&raid_dev->lock); + schedule(); + continue; + } + /* + * Get the first request from request list. + **/ + request = list_first_entry(&raid_dev->list, + struct mtd_raid_request, node_request); + list_del_init(&request->node_request); + spin_unlock(&raid_dev->lock); + + if (request->func) + request->func(request); + cond_resched(); + } + + return 0; +} + +/* Interfaces of raid_dev */ + +/* + * read interface for raid_dev. + * */ +int mtd_raid_dev_read(struct mtd_raid_ctx *ctx, struct mtd_raid_dev *raid_dev, + loff_t from, size_t len, size_t *retlen, u_char *buf) +{ + struct mtd_raid_read_request *read_req; + struct mtd_raid_request *request; + int ret = 0; + + /* Alloc a request */ + read_req = kzalloc(sizeof(*read_req), GFP_KERNEL); + if (!read_req) + goto out; + + request = &read_req->request; + /* Init the request */ + INIT_LIST_HEAD(&request->node); + INIT_LIST_HEAD(&request->node_all); + INIT_LIST_HEAD(&request->node_request); + + request->ctx = ctx; + request->raid_dev = raid_dev; + request->type = MTD_RAID_REQ_READ; + request->func = read_req_func; + request->retry_func = read_req_retry_func; + request->end_func = read_req_end_func; + + /* Init read_request */ + read_req->from = from; + read_req->len = len; + read_req->buf = buf; + + /* Add request to context */ + spin_lock(&ctx->lock); + list_add_tail(&request->node, &ctx->submit_list); + list_add_tail(&request->node_all, &ctx->all_list); + spin_unlock(&ctx->lock); + + /* Dispatch request to related raid_dev */ + spin_lock(&raid_dev->lock); + list_add_tail(&request->node_request, &raid_dev->list); + spin_unlock(&raid_dev->lock); + + /* Wakeup background thread to handle requests */ + wake_up_process(raid_dev->thread); + + *retlen = len; +out: + return ret; +} + +/* + * read_oob interface for raid_dev. + */ +int mtd_raid_dev_read_oob(struct mtd_raid_ctx *ctx, struct mtd_raid_dev *raid_dev, + loff_t from, size_t len, struct mtd_oob_ops *ops) +{ + struct mtd_raid_read_oob_request *read_oob_req; + struct mtd_raid_request *request; + int ret = 0; + + /* Alloc a request */ + read_oob_req = kzalloc(sizeof(*read_oob_req), GFP_KERNEL); + if (!read_oob_req) + goto out; + + request = &read_oob_req->request; + /* Init the request */ + INIT_LIST_HEAD(&request->node); + INIT_LIST_HEAD(&request->node_all); + INIT_LIST_HEAD(&request->node_request); + + request->ctx = ctx; + request->raid_dev = raid_dev; + request->type = MTD_RAID_REQ_READ_OOB; + request->func = read_oob_req_func; + request->retry_func = read_oob_req_retry_func; + request->end_func = read_oob_req_end_func; + + /* Init read_request */ + read_oob_req->from = from; + read_oob_req->len = len; + memcpy(&read_oob_req->ops, ops, sizeof(*ops)); + + /* Add request to context */ + spin_lock(&ctx->lock); + list_add_tail(&request->node, &ctx->submit_list); + list_add_tail(&request->node_all, &ctx->all_list); + spin_unlock(&ctx->lock); + + /* Dispatch request to related raid_dev */ + spin_lock(&raid_dev->lock); + list_add_tail(&request->node_request, &raid_dev->list); + spin_unlock(&raid_dev->lock); + + /* Wakeup background thread to handle requests */ + wake_up_process(raid_dev->thread); +out: + return ret; +} + +/* + * erase interface for raid_dev. + * */ +int mtd_raid_dev_erase(struct mtd_raid_dev *raid_dev, struct erase_info *erase) +{ + int err = 0; + struct mtd_info *mtd = raid_dev->mtd; + + erase->mtd = mtd; + erase->len = mtd->erasesize; + err = mtd_erase(mtd, erase); + if (err) + goto out; + + if (erase->state != MTD_ERASE_DONE) { + err = -EIO; + goto out; + } +out: + return err; +} diff --git a/include/uapi/mtd/mtd-raid-user.h b/include/uapi/mtd/mtd-raid-user.h new file mode 100644 index 0000000..8c735ba --- /dev/null +++ b/include/uapi/mtd/mtd-raid-user.h @@ -0,0 +1,33 @@ +/* + * Copyright 2015, see mtd/mtd-raid for licensing and copyright details + */ +#ifndef __MTD_RAID_USER_H__ +#define __MTD_RAID_USER_H__ + +#include <linux/types.h> +#include <linux/magic.h> + +/* ioctl's command */ +#define MTD_RAID_IOC_CREATE _IOW(0xFE, 1, struct mtd_raid_create_req) +#define MTD_RAID_IOC_DESTROY _IOW(0xFE, 2, struct mtd_raid_destroy_req) + +enum mtd_raid_level { + MTD_RAID_LEVEL_SINGLE = 0, + MTD_RAID_LEVEL_RAID0, + MTD_RAID_LEVEL_RAID1, + MTD_RAID_LEVEL_MAX +}; + +struct mtd_raid_create_req { + __u8 raid_level; + __u8 reserved[3]; + __u32 dev_count; + __u64 substripe_size; + __u32 mtd_nums[0]; +}; + +struct mtd_raid_destroy_req { + __u32 mtd_num; +}; + +#endif /* __MTD_RAID_USER_H__ */ -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 4/7] mtd: mtd_raid: introduce raid single implementation 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (2 preceding siblings ...) 2015-12-01 10:52 ` [PATCH v1 3/7] mtd: mtd_raid: Init a new layer of MTD RAID Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 5/7] mtd: mtd_raid: introduce raid0 implementation Dongsheng Yang ` (4 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang This is impletion of raid-single, raid-single works similar with mtd_concat. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/mtd_raid/raid_single.c | 156 +++++++++++++++++++++++++++++++++++++ 1 file changed, 156 insertions(+) create mode 100644 drivers/mtd/mtd_raid/raid_single.c diff --git a/drivers/mtd/mtd_raid/raid_single.c b/drivers/mtd/mtd_raid/raid_single.c new file mode 100644 index 0000000..e1421d0 --- /dev/null +++ b/drivers/mtd/mtd_raid/raid_single.c @@ -0,0 +1,156 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +/* + * This file implement single raild level. + * + * RAID_SINGLE: + * + * Flash A: + * ------------------------------ + * | A1 | A2 | ... | An | + * ------------------------------ + * Flash B: + * ------------------------------ + * | B1 | B2 | ... | Bn | + * ------------------------------ + * Flash C: + * ------------------------------ + * | C1 | C2 | ... | Cn | + * ------------------------------ + * + * RAID_SINGLE: + * --------------------------------------------------------------------- + * | A1 | A2 | ... | An | B1 | ... | Bn | C1 | ... | Cn | + * --------------------------------------------------------------------- + * + * It works similarly with concat. + */ +#include <linux/kthread.h> +#include <linux/slab.h> + +#include "mtd_raid.h" + +/** + * + */ +static int raid_single_logical_to_physical(struct mtd_raid *raid, loff_t from, size_t len, int copy_num, + int *devid, loff_t *subdev_off, size_t *size) +{ + struct mtd_info *subdev = NULL; + int i = 0; + + *subdev_off = from; + for (i = 0; i < raid->dev_count; i++) { + subdev = raid->devs[i].mtd; + if (from >= subdev->size) { + from -= subdev->size; + *subdev_off = 0; + *size = 0; + continue; + } + + *subdev_off = from; + if (from + len > subdev->size) + *size = subdev->size - *subdev_off; + else + *size = len; + + *devid = i; + break; + } + + if (i == raid->dev_count) + return -EINVAL; + + return 0; +} + +struct mtd_raid *mtd_raid_single_create(int *mtd_nums, int dev_count, size_t substripe_size) +{ + struct mtd_raid_single *raid_single = NULL; + struct mtd_raid *raid = NULL; + + raid_single = kzalloc(sizeof(struct mtd_raid_single) + sizeof(struct mtd_raid_dev) * dev_count, GFP_KERNEL); + if (!raid_single) + goto out; + + raid = &raid_single->raid; + raid->raid_level = MTD_RAID_LEVEL_SINGLE; + raid->ops = &mtd_raid_single_ops; + raid->ncopies = 1; + raid->npebs_per_leb = 1; + + return &raid_single->raid; +out: + return NULL; +} + +static int raid_single_init(struct mtd_raid *raid, + int dev_count, size_t substripe_size) +{ + int ret = 0; + int i = 0; + int raid_id = 0; + struct mtd_info *mtd = &raid->mtd; + struct mtd_info *subdev = raid->devs[0].mtd; + + raid_id = mtd_raid_list_register(MTD_RAID_LEVEL_SINGLE, raid); + sprintf(raid->name, "mtdsingle-%d", raid_id); + mtd->name = raid->name; + + /* + * Same erasesize with subdevs + **/ + mtd->erasesize = subdev->erasesize; + for (i = 0; i < dev_count; i++) { + subdev = raid->devs[i].mtd; + if (mtd->erasesize != subdev->erasesize) { + pr_err("Incompatible erasesize on \"%s\"", + subdev->name); + ret = -EINVAL; + goto err; + } + /* + * mtd->size is the sum of all subdevs + **/ + mtd->size += subdev->size; + } + return ret; +err: + mtd_raid_list_unregister(raid); + return ret; +} + + +static void raid_single_destroy(struct mtd_raid *mtd_raid) +{ + struct mtd_raid_single *raid_single = MTD_RAID_SINGLE(mtd_raid); + + mtd_raid_list_unregister(&raid_single->raid); + kfree(raid_single); +} + +const struct mtd_raid_operations mtd_raid_single_ops = { + .init = raid_single_init, + .destroy = raid_single_destroy, + .logical_to_physical = raid_single_logical_to_physical, +}; -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 5/7] mtd: mtd_raid: introduce raid0 implementation 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (3 preceding siblings ...) 2015-12-01 10:52 ` [PATCH v1 4/7] mtd: mtd_raid: introduce raid single implementation Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 6/7] mtd: mtd_raid: introduce raid1 implementation Dongsheng Yang ` (3 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang This is implementation of raid0, the main work of it is mapping logical address to physical address. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/mtd_raid/raid0.c | 136 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 136 insertions(+) create mode 100644 drivers/mtd/mtd_raid/raid0.c diff --git a/drivers/mtd/mtd_raid/raid0.c b/drivers/mtd/mtd_raid/raid0.c new file mode 100644 index 0000000..e95a6a9 --- /dev/null +++ b/drivers/mtd/mtd_raid/raid0.c @@ -0,0 +1,136 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +/** + * This file implement raid0. + * RAID0: + * + * Flash A: + * ------------------------------ + * | A1 | A2 | ... | An | + * ------------------------------ + * Flash B: + * ------------------------------ + * | B1 | B2 | ... | Bn | + * ------------------------------ + * Flash C: + * ------------------------------ + * | C1 | C2 | ... | Cn | + * ------------------------------ + * + * RAID_SINGLE: + * ----------------------------------------------------------------------- + * | A1 | B1 | C1 | A2 | B2 | C2 | ... | An | Bn | Cn | + * ----------------------------------------------------------------------- + * + * Detail to see: + * https://en.wikipedia.org/wiki/Standard_RAID_levels + */ + +#include <linux/slab.h> +#include <linux/kthread.h> + +#include "mtd_raid.h" + +static int raid0_logical_to_physical(struct mtd_raid *raid, loff_t from, size_t len, int copy_num, + int *devid, loff_t *subdev_off, size_t *size) +{ + loff_t stripe_size, stripe_offs, stripe_num, substripe_offs, substripe_num; + + stripe_size = raid->substripe_size * raid->npebs_per_leb; + stripe_offs = do_div(from, stripe_size); + stripe_num = from; + substripe_offs = do_div(stripe_offs, raid->substripe_size); + substripe_num = stripe_offs; + + *devid = substripe_num; + *subdev_off = stripe_num * raid->substripe_size + substripe_offs; + *size = len; + if (*size > raid->substripe_size - substripe_offs) + *size = raid->substripe_size - substripe_offs; + + return 0; +} + +struct mtd_raid *mtd_raid0_create(int *mtd_nums, int dev_count, size_t substripe_size) +{ + struct mtd_raid0 *raid0 = NULL; + struct mtd_raid *raid = NULL; + + raid0 = kzalloc(sizeof(struct mtd_raid0) + sizeof(struct mtd_raid_dev) * dev_count, GFP_KERNEL); + if (!raid0) + goto out; + + raid = &raid0->raid; + raid->raid_level = MTD_RAID_LEVEL_RAID0; + raid->ops = &mtd_raid0_ops; + raid->ncopies = 1; + raid->npebs_per_leb = dev_count; + + return &raid0->raid; +out: + return NULL; +} + +static int raid0_init(struct mtd_raid *raid, + int dev_count, size_t substripe_size) +{ + int ret = 0; + int i = 0; + int raid_id = 0; + struct mtd_info *mtd = &raid->mtd; + struct mtd_info *subdev0 = raid->devs[0].mtd; + struct mtd_info *subdev = NULL; + + raid_id = mtd_raid_list_register(MTD_RAID_LEVEL_RAID0, raid); + sprintf(raid->name, "mtd0-%d", raid_id); + mtd->name = raid->name; + + for (i = 0; i < dev_count; i++) { + subdev = raid->devs[i].mtd; + if (subdev0->size != subdev->size) { + pr_err("Incompatible size on \"%s\"", + subdev->name); + ret = -EINVAL; + goto out; + } + + mtd->size += subdev->size; + mtd->erasesize += subdev->erasesize; + } +out: + return ret; +} + + +void raid0_destroy(struct mtd_raid *mtd_raid) +{ + struct mtd_raid0 *raid0 = MTD_RAID_RAID0(mtd_raid); + + mtd_raid_list_unregister(&raid0->raid); + kfree(raid0); +} + +const struct mtd_raid_operations mtd_raid0_ops = { + .init = raid0_init, + .destroy = raid0_destroy, + .logical_to_physical = raid0_logical_to_physical, +}; -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 6/7] mtd: mtd_raid: introduce raid1 implementation 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (4 preceding siblings ...) 2015-12-01 10:52 ` [PATCH v1 5/7] mtd: mtd_raid: introduce raid0 implementation Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 7/7] mtd: mtd_raid: implement ioctl for mtd raid Dongsheng Yang ` (2 subsequent siblings) 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang This is implementation of raid1, the main work of it is to mapping logical address to physical address and mapping physical address to logical address. That would be used when we meet an error in reading and want to find the mirror in raid1. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/mtd_raid/raid1.c | 156 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 156 insertions(+) create mode 100644 drivers/mtd/mtd_raid/raid1.c diff --git a/drivers/mtd/mtd_raid/raid1.c b/drivers/mtd/mtd_raid/raid1.c new file mode 100644 index 0000000..98c01bd --- /dev/null +++ b/drivers/mtd/mtd_raid/raid1.c @@ -0,0 +1,156 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +/** + * This file implement raid1. + * RAID1: + * + * Flash A: + * ------------------------------ + * | A1 | A2 | ... | An | + * ------------------------------ + * Flash B: + * ------------------------------ + * | B1 | B2 | ... | Bn | + * ------------------------------ + * Flash C: + * ------------------------------ + * | C1 | C2 | ... | Cn | + * ------------------------------ + * + * RAID_SINGLE: + * ------------------------------ + * | A1 | A2 | ... | An | + * ------------------------------ + * ------------------------------ + * | B1 | B2 | ... | Bn | Mirror + * ------------------------------ + * ------------------------------ + * | C1 | C2 | ... | Cn | Mirror + * ------------------------------ + * + * C1 = B1 = A1. They store same data in three different flashes. + * + * Detail to see: + * https://en.wikipedia.org/wiki/Standard_RAID_levels + */ + +#include <linux/slab.h> +#include <linux/kthread.h> + +#include "mtd_raid.h" + +static int raid1_logical_to_physical(struct mtd_raid *raid, loff_t from, size_t len, int copy_num, + int *devid, loff_t *subdev_off, size_t *size) +{ + *devid = copy_num; + *subdev_off = from; + *size = len; + + return 0; +} + +/** + * This function will convert the physical address to logical address. + */ +static int raid1_physical_to_logical(struct mtd_raid *raid, int devid, loff_t subdev_off, size_t size, + loff_t *from, size_t *len) +{ + *from = subdev_off; + *len = size; + + return 0; +} + +struct mtd_raid *mtd_raid1_create(int *mtd_nums, int dev_count, size_t substripe_size) +{ + struct mtd_raid1 *raid1 = NULL; + struct mtd_raid *raid = NULL; + + raid1 = kzalloc(sizeof(struct mtd_raid1) + sizeof(struct mtd_raid_dev) * dev_count, GFP_KERNEL); + if (!raid1) + goto out; + + raid = &raid1->raid; + raid->raid_level = MTD_RAID_LEVEL_RAID1; + raid->ops = &mtd_raid1_ops; + raid->ncopies = dev_count; + raid->npebs_per_leb = 1; + + return &raid1->raid; +out: + return NULL; +} + +static int raid1_init(struct mtd_raid *raid, + int dev_count, size_t substripe_size) +{ + int ret = 0; + int i = 0; + int raid_id = 0; + struct mtd_info *mtd = &raid->mtd; + struct mtd_info *subdev = raid->devs[0].mtd; + + raid_id = mtd_raid_list_register(MTD_RAID_LEVEL_RAID1, raid); + sprintf(raid->name, "mtd1-%d", raid_id); + mtd->name = raid->name; + + /** + * size and erasesize are same with subdev. + */ + mtd->size = subdev->size; + mtd->erasesize = subdev->erasesize; + for (i = 0; i < dev_count; i++) { + subdev = raid->devs[i].mtd; + if (mtd->size != subdev->size) { + pr_err("Incompatible size on \"%s\"", + subdev->name); + ret = -EINVAL; + goto out; + } + + if (mtd->erasesize != subdev->erasesize) { + pr_err("Incompatible erasesize on \"%s\"", + subdev->name); + ret = -EINVAL; + goto out; + } + } + +out: + return ret; +} + + +void raid1_destroy(struct mtd_raid *mtd_raid) +{ + struct mtd_raid1 *raid1 = MTD_RAID_RAID1(mtd_raid); + + mtd_raid_list_unregister(&raid1->raid); + kfree(raid1); +} + +const struct mtd_raid_operations mtd_raid1_ops = { + .init = raid1_init, + .destroy = raid1_destroy, + .logical_to_physical = raid1_logical_to_physical, + .physical_to_logical = raid1_physical_to_logical, +}; -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH v1 7/7] mtd: mtd_raid: implement ioctl for mtd raid 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (5 preceding siblings ...) 2015-12-01 10:52 ` [PATCH v1 6/7] mtd: mtd_raid: introduce raid1 implementation Dongsheng Yang @ 2015-12-01 10:52 ` Dongsheng Yang 2015-12-01 10:53 ` [PATCH] mtd-utils: introduce mtd_raid tool Dongsheng Yang 2015-12-02 1:09 ` [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:52 UTC (permalink / raw) To: computersforpeace, richard.weinberger; +Cc: linux-mtd, Dongsheng Yang This is ioctl interface for mtd raid. Currently we support two operations create and destroy. Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- drivers/mtd/mtd_raid/ioctl.c | 137 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 137 insertions(+) create mode 100644 drivers/mtd/mtd_raid/ioctl.c diff --git a/drivers/mtd/mtd_raid/ioctl.c b/drivers/mtd/mtd_raid/ioctl.c new file mode 100644 index 0000000..22facee --- /dev/null +++ b/drivers/mtd/mtd_raid/ioctl.c @@ -0,0 +1,137 @@ +/* + * This file is part of MTD RAID. + * + * Copyright (C) 2015 Dongsheng Yang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., 51 + * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + */ + +#include <linux/compat.h> +#include <linux/slab.h> +#include <linux/string.h> + +#include "mtd_raid.h" + +static int ioctl_create_check(int *mtd_nums, int dev_count, size_t substripe_size) +{ + int i = 0; + + if (dev_count < 0 || substripe_size < 0) + return -EINVAL; + + for (i = 0; i < dev_count; i++) { + if (mtd_nums[i] < 0) + return -EINVAL; + } + + return 0; +} + +int mtd_raid_ioctl_create(struct mtd_raid_create_req *create_req, int *mtd_nums) +{ + enum mtd_raid_level raid_level = create_req->raid_level; + int dev_count = create_req->dev_count; + int substripe_size = create_req->substripe_size; + + if (ioctl_create_check(mtd_nums, dev_count, substripe_size)) + return -EINVAL; + + return mtd_raid_create(raid_level, mtd_nums, dev_count, substripe_size); +} + +int mtd_raid_ioctl_destroy(struct mtd_raid_destroy_req *destroy_req) +{ + struct mtd_raid *raid = NULL; + + raid = mtd_raid_list_get(destroy_req->mtd_num); + if (!raid) + return -EINVAL; + + return mtd_raid_destroy(raid); +} + +static long ctrl_cdev_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + int err = 0; + void __user *argp = (void __user *)arg; + + if (!capable(CAP_SYS_RESOURCE)) + return -EPERM; + + switch (cmd) { + case MTD_RAID_IOC_CREATE: + { + struct mtd_raid_create_req *req; + int dev_count = 0; + int *mtd_nums = NULL; + + req = memdup_user(argp, sizeof(*req)); + if (IS_ERR(req)) { + err = PTR_ERR(req); + goto out; + } + + dev_count = req->dev_count; + mtd_nums = memdup_user(argp + sizeof(*req), sizeof(int) * dev_count); + if (IS_ERR(mtd_nums)) { + err = PTR_ERR(mtd_nums); + goto out; + } + + err = mtd_raid_ioctl_create(req, mtd_nums); + break; + } + case MTD_RAID_IOC_DESTROY: + { + struct mtd_raid_destroy_req *req; + + req = memdup_user(argp, sizeof(*req)); + if (IS_ERR(req)) { + err = PTR_ERR(req); + goto out; + } + + err = mtd_raid_ioctl_destroy(req); + break; + } + + default: + err = -ENOTTY; + break; + } + +out: + return err; +} + +#ifdef CONFIG_COMPAT +static long ctrl_cdev_compat_ioctl(struct file *file, unsigned int cmd, + unsigned long arg) +{ + unsigned long translated_arg = (unsigned long)compat_ptr(arg); + + return ctrl_cdev_ioctl(file, cmd, translated_arg); +} +#else +#define ctrl_cdev_compat_ioctl NULL +#endif + +const struct file_operations mtd_raid_ctrl_cdev_operations = { + .owner = THIS_MODULE, + .unlocked_ioctl = ctrl_cdev_ioctl, + .compat_ioctl = ctrl_cdev_compat_ioctl, + .llseek = no_llseek, +}; -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH] mtd-utils: introduce mtd_raid tool. 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (6 preceding siblings ...) 2015-12-01 10:52 ` [PATCH v1 7/7] mtd: mtd_raid: implement ioctl for mtd raid Dongsheng Yang @ 2015-12-01 10:53 ` Dongsheng Yang 2015-12-02 1:09 ` [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-01 10:53 UTC (permalink / raw) To: linux-mtd; +Cc: Zeng Linggang, Dongsheng Yang From: Zeng Linggang <zenglg.jy@cn.fujitsu.com> Signed-off-by: Zeng Linggang <zenglg.jy@cn.fujitsu.com> Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> --- .gitignore | 1 + Makefile | 2 +- include/mtd/mtd-raid-user.h | 33 ++++++ misc-utils/mtd_raid.c | 246 ++++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 281 insertions(+), 1 deletion(-) create mode 100644 include/mtd/mtd-raid-user.h create mode 100644 misc-utils/mtd_raid.c diff --git a/.gitignore b/.gitignore index 04d9251..94ebdce 100644 --- a/.gitignore +++ b/.gitignore @@ -31,6 +31,7 @@ /misc-utils/flashcp /misc-utils/ftl_check /misc-utils/ftl_format +/misc-utils/mtd_raid /jffsX-utils/jffs2dump /jffsX-utils/jffs2reader /jffsX-utils/mkfs.jffs2 diff --git a/Makefile b/Makefile index 33d3e34..5282be9 100644 --- a/Makefile +++ b/Makefile @@ -20,7 +20,7 @@ MISC_BINS = \ ftl_format doc_loadbios ftl_check mtd_debug docfdisk \ serve_image recv_image mtdpart flash_erase flash_lock \ flash_unlock flash_otp_info flash_otp_dump flash_otp_lock \ - flash_otp_write flashcp + flash_otp_write flashcp mtd_raid UBI_BINS = \ ubiupdatevol ubimkvol ubirmvol ubicrc32 ubinfo ubiattach \ ubidetach ubinize ubiformat ubirename mtdinfo ubirsvol ubiblock diff --git a/include/mtd/mtd-raid-user.h b/include/mtd/mtd-raid-user.h new file mode 100644 index 0000000..8c735ba --- /dev/null +++ b/include/mtd/mtd-raid-user.h @@ -0,0 +1,33 @@ +/* + * Copyright 2015, see mtd/mtd-raid for licensing and copyright details + */ +#ifndef __MTD_RAID_USER_H__ +#define __MTD_RAID_USER_H__ + +#include <linux/types.h> +#include <linux/magic.h> + +/* ioctl's command */ +#define MTD_RAID_IOC_CREATE _IOW(0xFE, 1, struct mtd_raid_create_req) +#define MTD_RAID_IOC_DESTROY _IOW(0xFE, 2, struct mtd_raid_destroy_req) + +enum mtd_raid_level { + MTD_RAID_LEVEL_SINGLE = 0, + MTD_RAID_LEVEL_RAID0, + MTD_RAID_LEVEL_RAID1, + MTD_RAID_LEVEL_MAX +}; + +struct mtd_raid_create_req { + __u8 raid_level; + __u8 reserved[3]; + __u32 dev_count; + __u64 substripe_size; + __u32 mtd_nums[0]; +}; + +struct mtd_raid_destroy_req { + __u32 mtd_num; +}; + +#endif /* __MTD_RAID_USER_H__ */ diff --git a/misc-utils/mtd_raid.c b/misc-utils/mtd_raid.c new file mode 100644 index 0000000..203676a --- /dev/null +++ b/misc-utils/mtd_raid.c @@ -0,0 +1,246 @@ +/* + * User tool for MTD RAID + * + * Copyright (C) 2015 Zeng Linggang. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * + * Authors: Dongsheng Yang <yangds.fnst@cn.fujitsu.com> + * Zeng Linggang <zenglg.jy@cn.fujitsu.com> + */ + +#define PROGRAM_NAME "mtd_raid" + +#include <fcntl.h> +#include <getopt.h> +#include <limits.h> +#include <linux/blkpg.h> +#include <stdio.h> +#include <string.h> +#include <sys/ioctl.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> +#include <mtd/mtd-raid-user.h> + +#include "common.h" + +static void display_help(int status) +{ + printf( + "Usage:\n" + " %1$s create [OPTION]... <MTD_DEVICE1> ...\n" + " %1$s destroy <MTD_RAID_DEVICE>\n" + "\n" + "Options:\n" + " -l, --level=[single|0|1] RAID level of the device you want to create.\n" + " -s, --substripe-size=<SIZE> Size of substripe, it's not necessary for\n" + " raid-single. Default value is mtd->writesize.\n" + " -h, --help Display this help and exit\n" + "\n" + "Create or destroy a mtd raid device. With the create subcommand, we can create\n" + "a raid device from the devices you requested.\n", + PROGRAM_NAME); + exit(status); +} + +/* Command arguments */ + +typedef enum { + COMMAND_CREATE, + COMMAND_DESTROY +} command_type; + +static command_type command; +static int dev_count; +static int mtddevs[1024]; +static ssize_t substripe_size; +static int raid_level; +static int destroy_dev; + +static const char *optstring = "s:l:h"; + +static const struct option longopts[] = { + {"level", 1, NULL, 'l'}, + {"substripe-size", 1, NULL, 's'}, + {"help", 0, NULL, 'h'}, + {NULL, 0, NULL, 0} +}; + +#ifndef MTD_CHAR_MAJOR +/* + * This is taken from kernel <linux/mtd/mtd.h> and is unlikely to change anytime + * soon. + */ +#define MTD_CHAR_MAJOR 90 +#endif + +/** + * mtd_node_to_num - converts device node to MTD number. + * @mtd_dev_node: path to device node to convert + * + * This function converts given @mtd_dev_node to MTD device number. + * @mtd_dev_node should contain path to the MTD device node. Returns MTD device + * number in case of success and %-1 in case of failure (errno is set). + * + * XXX This is a copy from ubi-utils/libubi.c. + * TODO: + * Move the generic files from ubi-utils to top level and then + * kill the duplication of mtd_node_to_num, even more we can move mtdinfo + * from ubi-utils to misc-utils. + */ +static int mtd_node_to_num(const char *mtd_dev_node) +{ + int major, minor; + struct stat sb; + + if (stat(mtd_dev_node, &sb) < 0) { + sys_errmsg("cannot stat \"%s\"", mtd_dev_node); + return -EINVAL; + } + + if (!S_ISCHR(sb.st_mode)) { + sys_errmsg("\"%s\" is not a character device", + mtd_dev_node); + return -EINVAL; + } + + major = major(sb.st_rdev); + minor = minor(sb.st_rdev); + + if (major != MTD_CHAR_MAJOR) { + sys_errmsg("\"%s\" is not an MTD device", mtd_dev_node); + return -EINVAL; + } + + return minor / 2; +} + +static enum mtd_raid_level get_raid_level(const char *level_str) +{ + if (strcmp(level_str, "single") == 0) { + return MTD_RAID_LEVEL_SINGLE; + } else if (strcmp(level_str, "0") == 0) { + return MTD_RAID_LEVEL_RAID0; + } else if (strcmp(level_str, "1") == 0) { + return MTD_RAID_LEVEL_RAID1; + } else { + return -EINVAL; + } +} + +static void process_options(int argc, char * const argv[]) +{ + int mtd_num = 0; + const char *s_command = argv[optind++]; + + if (strcmp(s_command, "destroy") == 0 && argc == 3) + command = COMMAND_DESTROY; + else if (strcmp(s_command, "create") == 0 && argc > 2) + command = COMMAND_CREATE; + else + goto err; + + while (1) { + int c = getopt_long(argc, argv, optstring, longopts, NULL); + if (c == -1) + break; + + switch (c) { + case 'l': + raid_level = get_raid_level(optarg); + if (raid_level < 0) + goto err; + break; + case 's': + substripe_size = strtoull(optarg, NULL, 0); + break; + case 'h': + display_help(EXIT_SUCCESS); + break; + case '?': + goto err; + } + } + + if (command == COMMAND_CREATE) { + int i = 0; + + while (optind < argc) { + mtd_num = mtd_node_to_num(argv[optind++]); + if (mtd_num < 0) + goto err; + if (i >= 1024) + goto err; + mtddevs[i++] = mtd_num; + } + dev_count = i; + if (dev_count == 0) + goto err; + } else if (command == COMMAND_DESTROY) { + mtd_num = mtd_node_to_num(argv[optind++]); + if (mtd_num < 0) + goto err; + destroy_dev = mtd_num; + } + + return; +err: + display_help(EXIT_FAILURE); +} + + +int main(int argc, char * const argv[]) +{ + int fd, i = 0; + struct mtd_raid_create_req *create_req = NULL; + struct mtd_raid_destroy_req *destroy_req = NULL; + + process_options(argc, argv); + + fd = open("/dev/mtd_raid_ctrl", O_RDWR | O_CLOEXEC); + if (fd == -1) + sys_errmsg_die("Cannot open /dev/mtd_raid_ctrl"); + + switch (command) { + case COMMAND_CREATE: + create_req = malloc(sizeof(*create_req) + sizeof(int) * dev_count); + if (!create_req) + sys_errmsg_die("malloc error"); + memset(create_req, 0, sizeof(*create_req) + sizeof(int) * dev_count); + create_req->raid_level = raid_level; + create_req->dev_count = dev_count; + for (i = 0; i < dev_count; i++) + create_req->mtd_nums[i] = mtddevs[i]; + create_req->substripe_size = substripe_size; + + if (ioctl(fd, MTD_RAID_IOC_CREATE, create_req)) + sys_errmsg_die("Failed to issue MTD_RAID_IOC_CREATE ioctl"); + break; + case COMMAND_DESTROY: + destroy_req = malloc(sizeof(*destroy_req)); + if (!destroy_req) + sys_errmsg_die("malloc error"); + memset(destroy_req, 0, sizeof(*destroy_req)); + destroy_req->mtd_num = destroy_dev; + if (ioctl(fd, MTD_RAID_IOC_DESTROY, destroy_req)) + sys_errmsg_die("Failed to issue MTD_RAID_IOC_DESTROY ioctl, %d", destroy_req->mtd_num); + break; + } + + close(fd); + /* Exit happy */ + return EXIT_SUCCESS; +} -- 1.8.4.2 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* Re: [PATCH v1 0/7] Add a mtd raid layer 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang ` (7 preceding siblings ...) 2015-12-01 10:53 ` [PATCH] mtd-utils: introduce mtd_raid tool Dongsheng Yang @ 2015-12-02 1:09 ` Dongsheng Yang 8 siblings, 0 replies; 10+ messages in thread From: Dongsheng Yang @ 2015-12-02 1:09 UTC (permalink / raw) To: Boris Brezillon Cc: Brian Norris, richard.weinberger, linux-mtd, Dongsheng Yang Adding Boris... Hi Boris, I found you are working on the problem of data lose in MLC, there is another idea to protect the our data, by raid5. What do you think? https://en.wikipedia.org/wiki/Standard_RAID_levels#RAID_5 Yang On 12/01/2015 06:52 PM, Dongsheng Yang wrote: > Hi guys, > I am here to release v1 of the mtd raid layer. It's a layer > shown as below: > > --------- ------------- > | ubi | | mtdchar | > --------- ------------- > | | > v v > ---------------------------- > | MTD RAID | <----- a new layer here (drivers/mtd/mtd_raid/*) > ---------------------------- > | > v > ---------------- > | mtd driver | > ---------------- > > Of course, that's optional. > > There are two ways to use mtd raid. > (1). For multi-chips driver: > > /dev/mtd0 > |--------------------------------| > | ---------------------------- | > | | MTD RAID | | > | ---------------------------- | > | | | | > | v v | > | ----------- ----------- | > | | mtd_info| | mtd_info| | > | ----------- ----------- | > |--------------------------------| > > When a driver have multiple chips, we can use mtd raid framework to build > them into a raid array and expose a mtd device. There are some drivers > are using mtd_concat, we can do the same thing with mtd_single. At the same > time, mtd raid provides more choices to them, such as raid0 or raid1. > > (2). For user: > > /dev/mtd2 > |--------------------------------| > | ---------------------------- | > | | MTD RAID | | > | ---------------------------- | > | | | | > | v v | > | ----------- ----------- | > | |/dev/mtd0| |/dev/mtd1| | > | ----------- ----------- | > |--------------------------------| > > we have a user tool for mtd raid, and user can use it to build raid array. > > Example: > # mtd_raid create --level=0 /dev/mtd0 /dev/mtd1 /dev/mtd2 /dev/mtd3 > # mtdinfo /dev/mtd4 > mtd4 > Name: mtdraid0-1 > Type: nand > Eraseblock size: 65536 bytes, 64.0 KiB > Amount of eraseblocks: 8192 (536870912 bytes, 512.0 MiB) > Minimum input/output unit size: 512 bytes > Sub-page size: 256 bytes > OOB size: 16 bytes > Character device major/minor: 90:8 > Bad blocks are allowed: true > Device is writable: true > > There are 4 mtd devices simulated by nandsim, then we create a raid0 with > them. > --------------------------------------------------------------------------- > | device| size | throughput (dd if=/dev/mtdX of=/dev/null bs=1M count=10)| > |-------------------------------------------------------------------------- > | mtd0 | 128M | 14.0 MB/s | > |-------------------------------------------------------------------------- > | mtd1 | 128M | 14.0 MB/s | > |-------------------------------------------------------------------------- > | mtd2 | 128M | 14.0 MB/s | > |-------------------------------------------------------------------------- > | mtd3 | 128M | 14.0 MB/s | > |-------------------------------------------------------------------------- > | mtd4 | 512M | 51.1 MB/s | > --------------------------------------------------------------------------- > > Also we can use raid1 to get a mirror for robustness. > > TODO: > In this framework, we can implement more raid levels. The most useful > one I think is raid5/6, which would do a parity and fixup work in reading. That's > important for MLC or TLC. Raid-level is refered to: > > https://en.wikipedia.org/wiki/Standard_RAID_levels > > > Code: > kernel: https://github.com/yangdongsheng/linux mtd_raid_v1 > user : https://github.com/yangdongsheng/mtd-utils mtd_raid_v1 > > Thanx > Yang > > Dongsheng Yang (7): > mtd: nandsim: remove the abuse of reference to globle variable. > mtd: nansim: add support of multi devices. > mtd: mtd_raid: Init a new layer of MTD RAID > mtd: mtd_raid: introduce raid single implementation > mtd: mtd_raid: introduce raid0 implementation > mtd: mtd_raid: introduce raid1 implementation > mtd: mtd_raid: implement ioctl for mtd raid > > Documentation/ioctl/ioctl-number.txt | 1 + > drivers/mtd/Kconfig | 2 + > drivers/mtd/Makefile | 1 + > drivers/mtd/mtd_raid/Kconfig | 12 + > drivers/mtd/mtd_raid/Makefile | 3 + > drivers/mtd/mtd_raid/core.c | 1000 ++++++++++++++++++++++++++++++++++ > drivers/mtd/mtd_raid/ioctl.c | 137 +++++ > drivers/mtd/mtd_raid/mtd_raid.h | 264 +++++++++ > drivers/mtd/mtd_raid/raid0.c | 136 +++++ > drivers/mtd/mtd_raid/raid1.c | 156 ++++++ > drivers/mtd/mtd_raid/raid_io.c | 449 +++++++++++++++ > drivers/mtd/mtd_raid/raid_single.c | 156 ++++++ > drivers/mtd/nand/nandsim.c | 312 ++++++----- > include/uapi/mtd/mtd-raid-user.h | 33 ++ > 14 files changed, 2518 insertions(+), 144 deletions(-) > create mode 100644 drivers/mtd/mtd_raid/Kconfig > create mode 100644 drivers/mtd/mtd_raid/Makefile > create mode 100644 drivers/mtd/mtd_raid/core.c > create mode 100644 drivers/mtd/mtd_raid/ioctl.c > create mode 100644 drivers/mtd/mtd_raid/mtd_raid.h > create mode 100644 drivers/mtd/mtd_raid/raid0.c > create mode 100644 drivers/mtd/mtd_raid/raid1.c > create mode 100644 drivers/mtd/mtd_raid/raid_io.c > create mode 100644 drivers/mtd/mtd_raid/raid_single.c > create mode 100644 include/uapi/mtd/mtd-raid-user.h > ^ permalink raw reply [flat|nested] 10+ messages in thread
end of thread, other threads:[~2015-12-02 1:18 UTC | newest] Thread overview: 10+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2015-12-01 10:52 [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 1/7] mtd: nandsim: remove the abuse of reference to globle variable Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 2/7] mtd: nansim: add support of multi devices Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 3/7] mtd: mtd_raid: Init a new layer of MTD RAID Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 4/7] mtd: mtd_raid: introduce raid single implementation Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 5/7] mtd: mtd_raid: introduce raid0 implementation Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 6/7] mtd: mtd_raid: introduce raid1 implementation Dongsheng Yang 2015-12-01 10:52 ` [PATCH v1 7/7] mtd: mtd_raid: implement ioctl for mtd raid Dongsheng Yang 2015-12-01 10:53 ` [PATCH] mtd-utils: introduce mtd_raid tool Dongsheng Yang 2015-12-02 1:09 ` [PATCH v1 0/7] Add a mtd raid layer Dongsheng Yang
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.