From 06d91a5fe0b50c9060e70bdf7786f8a3c66249db Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:12 -0700 Subject: [PATCH] md: improve locking on 'safemode' and move superblock writes When md marks the superblock dirty before a write, it calls generic_make_request (to write the superblock) from within generic_make_request (to write the first dirty block), which could cause problems later. With this patch, the superblock write is always done by the helper thread, and write request are delayed until that write completes. Also, the locking around marking the array dirty and writing the superblock is improved to avoid possible races. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- include/linux/raid/md.h | 2 +- include/linux/raid/md_k.h | 7 +++++++ 2 files changed, 8 insertions(+), 1 deletion(-) (limited to 'include/linux/raid') diff --git a/include/linux/raid/md.h b/include/linux/raid/md.h index a6a67d102bf..cfde8f497d6 100644 --- a/include/linux/raid/md.h +++ b/include/linux/raid/md.h @@ -69,7 +69,7 @@ extern mdk_thread_t * md_register_thread (void (*run) (mddev_t *mddev), extern void md_unregister_thread (mdk_thread_t *thread); extern void md_wakeup_thread(mdk_thread_t *thread); extern void md_check_recovery(mddev_t *mddev); -extern void md_write_start(mddev_t *mddev); +extern int md_write_start(mddev_t *mddev, struct bio *bi); extern void md_write_end(mddev_t *mddev); extern void md_handle_safemode(mddev_t *mddev); extern void md_done_sync(mddev_t *mddev, int blocks, int ok); diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index c9a0d4013be..d92db54255a 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -15,6 +15,9 @@ #ifndef _MD_K_H #define _MD_K_H +/* and dm-bio-list.h is not under include/linux because.... ??? */ +#include "../../../drivers/md/dm-bio-list.h" + #define MD_RESERVED 0UL #define LINEAR 1UL #define RAID0 2UL @@ -252,6 +255,10 @@ struct mddev_s atomic_t recovery_active; /* blocks scheduled, but not written */ wait_queue_head_t recovery_wait; sector_t recovery_cp; + + spinlock_t write_lock; + struct bio_list write_list; + unsigned int safemode; /* if set, update "clean" superblock * when no writes pending. */ -- cgit v1.2.3-70-g09d2 From 57afd89f98a990747445f01c458ecae64263b2f8 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:13 -0700 Subject: [PATCH] md: improve the interface to sync_request 1/ change the return value (which is number-of-sectors synced) from 'int' to 'sector_t'. The number of sectors is usually easily small enough to fit in an int, but if resync needs to abort, it may want to return the total number of remaining sectors, which could be large. Also errors cannot be returned as negative numbers now, so use 0 instead 2/ Add a 'skipped' return parameter to allow the array to report that it skipped the sectors. This allows md to take this into account in the speed calculations. Currently there is no important skipping, but the bitmap-based-resync that is coming will use this. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/md.c | 35 +++++++++++++++++++++++------------ drivers/md/raid1.c | 8 ++++---- drivers/md/raid10.c | 19 ++++++++++++------- drivers/md/raid5.c | 6 +++--- drivers/md/raid6main.c | 6 +++--- include/linux/raid/md_k.h | 2 +- 6 files changed, 46 insertions(+), 30 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/md.c b/drivers/md/md.c index 177d2a7d7ce..fa608a1a5c2 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -3241,12 +3241,13 @@ static void md_do_sync(mddev_t *mddev) mddev_t *mddev2; unsigned int currspeed = 0, window; - sector_t max_sectors,j; + sector_t max_sectors,j, io_sectors; unsigned long mark[SYNC_MARKS]; sector_t mark_cnt[SYNC_MARKS]; int last_mark,m; struct list_head *tmp; sector_t last_check; + int skipped = 0; /* just incase thread restarts... */ if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) @@ -3312,7 +3313,7 @@ static void md_do_sync(mddev_t *mddev) if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) /* resync follows the size requested by the personality, - * which default to physical size, but can be virtual size + * which defaults to physical size, but can be virtual size */ max_sectors = mddev->resync_max_sectors; else @@ -3331,9 +3332,10 @@ static void md_do_sync(mddev_t *mddev) j = mddev->recovery_cp; else j = 0; + io_sectors = 0; for (m = 0; m < SYNC_MARKS; m++) { mark[m] = jiffies; - mark_cnt[m] = j; + mark_cnt[m] = io_sectors; } last_mark = 0; mddev->resync_mark = mark[last_mark]; @@ -3358,21 +3360,29 @@ static void md_do_sync(mddev_t *mddev) } while (j < max_sectors) { - int sectors; + sector_t sectors; - sectors = mddev->pers->sync_request(mddev, j, currspeed < sysctl_speed_limit_min); - if (sectors < 0) { + skipped = 0; + sectors = mddev->pers->sync_request(mddev, j, &skipped, + currspeed < sysctl_speed_limit_min); + if (sectors == 0) { set_bit(MD_RECOVERY_ERR, &mddev->recovery); goto out; } - atomic_add(sectors, &mddev->recovery_active); + + if (!skipped) { /* actual IO requested */ + io_sectors += sectors; + atomic_add(sectors, &mddev->recovery_active); + } + j += sectors; if (j>1) mddev->curr_resync = j; - if (last_check + window > j || j == max_sectors) + + if (last_check + window > io_sectors || j == max_sectors) continue; - last_check = j; + last_check = io_sectors; if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) || test_bit(MD_RECOVERY_ERR, &mddev->recovery)) @@ -3386,7 +3396,7 @@ static void md_do_sync(mddev_t *mddev) mddev->resync_mark = mark[next]; mddev->resync_mark_cnt = mark_cnt[next]; mark[next] = jiffies; - mark_cnt[next] = j - atomic_read(&mddev->recovery_active); + mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active); last_mark = next; } @@ -3413,7 +3423,8 @@ static void md_do_sync(mddev_t *mddev) mddev->queue->unplug_fn(mddev->queue); cond_resched(); - currspeed = ((unsigned long)(j-mddev->resync_mark_cnt))/2/((jiffies-mddev->resync_mark)/HZ +1) +1; + currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2 + /((jiffies-mddev->resync_mark)/HZ +1) +1; if (currspeed > sysctl_speed_limit_min) { if ((currspeed > sysctl_speed_limit_max) || @@ -3433,7 +3444,7 @@ static void md_do_sync(mddev_t *mddev) wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); /* tell personality that we are finished */ - mddev->pers->sync_request(mddev, max_sectors, 1); + mddev->pers->sync_request(mddev, max_sectors, &skipped, 1); if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && mddev->curr_resync > 2 && diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index 3f1280bbaf3..3c5c916cb09 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -1010,7 +1010,7 @@ static int init_resync(conf_t *conf) * that can be installed to exclude normal IO requests. */ -static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) +static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) { conf_t *conf = mddev_to_conf(mddev); mirror_info_t *mirror; @@ -1023,7 +1023,7 @@ static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) if (!conf->r1buf_pool) if (init_resync(conf)) - return -ENOMEM; + return 0; max_sector = mddev->size << 1; if (sector_nr >= max_sector) { @@ -1107,8 +1107,8 @@ static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) /* There is nowhere to write, so all non-sync * drives must be failed - so we are finished */ - int rv = max_sector - sector_nr; - md_done_sync(mddev, rv, 1); + sector_t rv = max_sector - sector_nr; + *skipped = 1; put_buf(r1_bio); rdev_dec_pending(conf->mirrors[disk].rdev, mddev); return rv; diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index bfc9f52f0ec..8476515bfdc 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -1321,7 +1321,7 @@ static int init_resync(conf_t *conf) * */ -static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) +static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) { conf_t *conf = mddev_to_conf(mddev); r10bio_t *r10_bio; @@ -1335,7 +1335,7 @@ static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) if (!conf->r10buf_pool) if (init_resync(conf)) - return -ENOMEM; + return 0; skipped: max_sector = mddev->size << 1; @@ -1343,15 +1343,15 @@ static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) max_sector = mddev->resync_max_sectors; if (sector_nr >= max_sector) { close_sync(conf); + *skipped = 1; return sectors_skipped; } if (chunks_skipped >= conf->raid_disks) { /* if there has been nothing to do on any drive, * then there is nothing to do at all.. */ - sector_t sec = max_sector - sector_nr; - md_done_sync(mddev, sec, 1); - return sec + sectors_skipped; + *skipped = 1; + return (max_sector - sector_nr) + sectors_skipped; } /* make sure whole request will fit in a chunk - if chunks @@ -1565,17 +1565,22 @@ static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) } } + if (sectors_skipped) + /* pretend they weren't skipped, it makes + * no important difference in this case + */ + md_done_sync(mddev, sectors_skipped, 1); + return sectors_skipped + nr_sectors; giveup: /* There is nowhere to write, so all non-sync * drives must be failed, so try the next chunk... */ { - int sec = max_sector - sector_nr; + sector_t sec = max_sector - sector_nr; sectors_skipped += sec; chunks_skipped ++; sector_nr = max_sector; - md_done_sync(mddev, sec, 1); goto skipped; } } diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 677ce49078d..1ce3f5aaa98 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -1477,7 +1477,7 @@ static int make_request (request_queue_t *q, struct bio * bi) } /* FIXME go_faster isn't used */ -static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) +static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) { raid5_conf_t *conf = (raid5_conf_t *) mddev->private; struct stripe_head *sh; @@ -1500,8 +1500,8 @@ static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) * nothing we can do. */ if (mddev->degraded >= 1 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { - int rv = (mddev->size << 1) - sector_nr; - md_done_sync(mddev, rv, 1); + sector_t rv = (mddev->size << 1) - sector_nr; + *skipped = 1; return rv; } diff --git a/drivers/md/raid6main.c b/drivers/md/raid6main.c index fede16c4e8f..d9c385496dc 100644 --- a/drivers/md/raid6main.c +++ b/drivers/md/raid6main.c @@ -1636,7 +1636,7 @@ static int make_request (request_queue_t *q, struct bio * bi) } /* FIXME go_faster isn't used */ -static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) +static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) { raid6_conf_t *conf = (raid6_conf_t *) mddev->private; struct stripe_head *sh; @@ -1659,8 +1659,8 @@ static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) * nothing we can do. */ if (mddev->degraded >= 2 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { - int rv = (mddev->size << 1) - sector_nr; - md_done_sync(mddev, rv, 1); + sector_t rv = (mddev->size << 1) - sector_nr; + *skipped = 1; return rv; } diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index d92db54255a..bce0032decf 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -298,7 +298,7 @@ struct mdk_personality_s int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev); int (*hot_remove_disk) (mddev_t *mddev, int number); int (*spare_active) (mddev_t *mddev); - int (*sync_request)(mddev_t *mddev, sector_t sector_nr, int go_faster); + sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster); int (*resize) (mddev_t *mddev, sector_t sectors); int (*reshape) (mddev_t *mddev, int raid_disks); int (*reconfig) (mddev_t *mddev, int layout, int chunk_size); -- cgit v1.2.3-70-g09d2 From 32a7627cf3a35396a8e834faf34e38ae9f3b1309 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:14 -0700 Subject: [PATCH] md: optimised resync using Bitmap based intent logging With this patch, the intent to write to some block in the array can be logged to a bitmap file. Each bit represents some number of sectors and is set before any update happens, and only cleared when all writes relating to all sectors are complete. After an unclean shutdown, information in this bitmap can be used to optimise resync - only sectors which could be out-of-sync need to be updated. Also if a drive is removed and then added back into an array, the recovery can make use of the bitmap to optimise reconstruction. This is not implemented in this patch. Currently the bitmap is stored in a file which must (obviously) be stored on a separate device. The patch only provided infrastructure. It does not update any personalities to bitmap intent logging. Md arrays can still be used with no bitmap file. This patch has minimal impact on such arrays. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/Makefile | 3 +- drivers/md/bitmap.c | 1519 +++++++++++++++++++++++++++++++++++++++++++ drivers/md/md.c | 172 ++++- include/linux/raid/bitmap.h | 280 ++++++++ include/linux/raid/md_k.h | 4 + include/linux/raid/md_u.h | 7 + 6 files changed, 1970 insertions(+), 15 deletions(-) create mode 100644 drivers/md/bitmap.c create mode 100644 include/linux/raid/bitmap.h (limited to 'include/linux/raid') diff --git a/drivers/md/Makefile b/drivers/md/Makefile index 90de9c146a5..d3efedf6a6a 100644 --- a/drivers/md/Makefile +++ b/drivers/md/Makefile @@ -7,6 +7,7 @@ dm-mod-objs := dm.o dm-table.o dm-target.o dm-linear.o dm-stripe.o \ dm-multipath-objs := dm-hw-handler.o dm-path-selector.o dm-mpath.o dm-snapshot-objs := dm-snap.o dm-exception-store.o dm-mirror-objs := dm-log.o dm-raid1.o +md-mod-objs := md.o bitmap.o raid6-objs := raid6main.o raid6algos.o raid6recov.o raid6tables.o \ raid6int1.o raid6int2.o raid6int4.o \ raid6int8.o raid6int16.o raid6int32.o \ @@ -28,7 +29,7 @@ obj-$(CONFIG_MD_RAID5) += raid5.o xor.o obj-$(CONFIG_MD_RAID6) += raid6.o xor.o obj-$(CONFIG_MD_MULTIPATH) += multipath.o obj-$(CONFIG_MD_FAULTY) += faulty.o -obj-$(CONFIG_BLK_DEV_MD) += md.o +obj-$(CONFIG_BLK_DEV_MD) += md-mod.o obj-$(CONFIG_BLK_DEV_DM) += dm-mod.o obj-$(CONFIG_DM_CRYPT) += dm-crypt.o obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c new file mode 100644 index 00000000000..34ffc133db0 --- /dev/null +++ b/drivers/md/bitmap.c @@ -0,0 +1,1519 @@ +/* + * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 + * + * bitmap_create - sets up the bitmap structure + * bitmap_destroy - destroys the bitmap structure + * + * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.: + * - added disk storage for bitmap + * - changes to allow various bitmap chunk sizes + * - added bitmap daemon (to asynchronously clear bitmap bits from disk) + */ + +/* + * Still to do: + * + * flush after percent set rather than just time based. (maybe both). + * wait if count gets too high, wake when it drops to half. + * allow bitmap to be mirrored with superblock (before or after...) + * allow hot-add to re-instate a current device. + * allow hot-add of bitmap after quiessing device + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* debug macros */ + +#define DEBUG 0 + +#if DEBUG +/* these are for debugging purposes only! */ + +/* define one and only one of these */ +#define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */ +#define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/ +#define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */ +#define INJECT_FAULTS_4 0 /* undef */ +#define INJECT_FAULTS_5 0 /* undef */ +#define INJECT_FAULTS_6 0 + +/* if these are defined, the driver will fail! debug only */ +#define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */ +#define INJECT_FATAL_FAULT_2 0 /* undef */ +#define INJECT_FATAL_FAULT_3 0 /* undef */ +#endif + +//#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */ +#define DPRINTK(x...) do { } while(0) + +#ifndef PRINTK +# if DEBUG > 0 +# define PRINTK(x...) printk(KERN_DEBUG x) +# else +# define PRINTK(x...) +# endif +#endif + +static inline char * bmname(struct bitmap *bitmap) +{ + return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; +} + + +/* + * test if the bitmap is active + */ +int bitmap_active(struct bitmap *bitmap) +{ + unsigned long flags; + int res = 0; + + if (!bitmap) + return res; + spin_lock_irqsave(&bitmap->lock, flags); + res = bitmap->flags & BITMAP_ACTIVE; + spin_unlock_irqrestore(&bitmap->lock, flags); + return res; +} + +#define WRITE_POOL_SIZE 256 +/* mempool for queueing pending writes on the bitmap file */ +static void *write_pool_alloc(unsigned int gfp_flags, void *data) +{ + return kmalloc(sizeof(struct page_list), gfp_flags); +} + +static void write_pool_free(void *ptr, void *data) +{ + kfree(ptr); +} + +/* + * just a placeholder - calls kmalloc for bitmap pages + */ +static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) +{ + unsigned char *page; + +#if INJECT_FAULTS_1 + page = NULL; +#else + page = kmalloc(PAGE_SIZE, GFP_NOIO); +#endif + if (!page) + printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); + else + printk("%s: bitmap_alloc_page: allocated page at %p\n", + bmname(bitmap), page); + return page; +} + +/* + * for now just a placeholder -- just calls kfree for bitmap pages + */ +static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) +{ + PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); + kfree(page); +} + +/* + * check a page and, if necessary, allocate it (or hijack it if the alloc fails) + * + * 1) check to see if this page is allocated, if it's not then try to alloc + * 2) if the alloc fails, set the page's hijacked flag so we'll use the + * page pointer directly as a counter + * + * if we find our page, we increment the page's refcount so that it stays + * allocated while we're using it + */ +static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create) +{ + unsigned char *mappage; + + if (page >= bitmap->pages) { + printk(KERN_ALERT + "%s: invalid bitmap page request: %lu (> %lu)\n", + bmname(bitmap), page, bitmap->pages-1); + return -EINVAL; + } + + + if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ + return 0; + + if (bitmap->bp[page].map) /* page is already allocated, just return */ + return 0; + + if (!create) + return -ENOENT; + + spin_unlock_irq(&bitmap->lock); + + /* this page has not been allocated yet */ + + if ((mappage = bitmap_alloc_page(bitmap)) == NULL) { + PRINTK("%s: bitmap map page allocation failed, hijacking\n", + bmname(bitmap)); + /* failed - set the hijacked flag so that we can use the + * pointer as a counter */ + spin_lock_irq(&bitmap->lock); + if (!bitmap->bp[page].map) + bitmap->bp[page].hijacked = 1; + goto out; + } + + /* got a page */ + + spin_lock_irq(&bitmap->lock); + + /* recheck the page */ + + if (bitmap->bp[page].map || bitmap->bp[page].hijacked) { + /* somebody beat us to getting the page */ + bitmap_free_page(bitmap, mappage); + return 0; + } + + /* no page was in place and we have one, so install it */ + + memset(mappage, 0, PAGE_SIZE); + bitmap->bp[page].map = mappage; + bitmap->missing_pages--; +out: + return 0; +} + + +/* if page is completely empty, put it back on the free list, or dealloc it */ +/* if page was hijacked, unmark the flag so it might get alloced next time */ +/* Note: lock should be held when calling this */ +static inline void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) +{ + char *ptr; + + if (bitmap->bp[page].count) /* page is still busy */ + return; + + /* page is no longer in use, it can be released */ + + if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ + bitmap->bp[page].hijacked = 0; + bitmap->bp[page].map = NULL; + return; + } + + /* normal case, free the page */ + +#if 0 +/* actually ... let's not. We will probably need the page again exactly when + * memory is tight and we are flusing to disk + */ + return; +#else + ptr = bitmap->bp[page].map; + bitmap->bp[page].map = NULL; + bitmap->missing_pages++; + bitmap_free_page(bitmap, ptr); + return; +#endif +} + + +/* + * bitmap file handling - read and write the bitmap file and its superblock + */ + +/* copy the pathname of a file to a buffer */ +char *file_path(struct file *file, char *buf, int count) +{ + struct dentry *d; + struct vfsmount *v; + + if (!buf) + return NULL; + + d = file->f_dentry; + v = file->f_vfsmnt; + + buf = d_path(d, v, buf, count); + + return IS_ERR(buf) ? NULL : buf; +} + +/* + * basic page I/O operations + */ + +/* + * write out a page + */ +static int write_page(struct page *page, int wait) +{ + int ret = -ENOMEM; + + lock_page(page); + + if (page->mapping == NULL) + goto unlock_out; + else if (i_size_read(page->mapping->host) < page->index << PAGE_SHIFT) { + ret = -ENOENT; + goto unlock_out; + } + + ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE); + if (!ret) + ret = page->mapping->a_ops->commit_write(NULL, page, 0, + PAGE_SIZE); + if (ret) { +unlock_out: + unlock_page(page); + return ret; + } + + set_page_dirty(page); /* force it to be written out */ + return write_one_page(page, wait); +} + +/* read a page from a file, pinning it into cache, and return bytes_read */ +static struct page *read_page(struct file *file, unsigned long index, + unsigned long *bytes_read) +{ + struct inode *inode = file->f_mapping->host; + struct page *page = NULL; + loff_t isize = i_size_read(inode); + unsigned long end_index = isize >> PAGE_CACHE_SHIFT; + + PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_CACHE_SIZE, + (unsigned long long)index << PAGE_CACHE_SHIFT); + + page = read_cache_page(inode->i_mapping, index, + (filler_t *)inode->i_mapping->a_ops->readpage, file); + if (IS_ERR(page)) + goto out; + wait_on_page_locked(page); + if (!PageUptodate(page) || PageError(page)) { + page_cache_release(page); + page = ERR_PTR(-EIO); + goto out; + } + + if (index > end_index) /* we have read beyond EOF */ + *bytes_read = 0; + else if (index == end_index) /* possible short read */ + *bytes_read = isize & ~PAGE_CACHE_MASK; + else + *bytes_read = PAGE_CACHE_SIZE; /* got a full page */ +out: + if (IS_ERR(page)) + printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", + (int)PAGE_CACHE_SIZE, + (unsigned long long)index << PAGE_CACHE_SHIFT, + PTR_ERR(page)); + return page; +} + +/* + * bitmap file superblock operations + */ + +/* update the event counter and sync the superblock to disk */ +int bitmap_update_sb(struct bitmap *bitmap) +{ + bitmap_super_t *sb; + unsigned long flags; + + if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ + return 0; + spin_lock_irqsave(&bitmap->lock, flags); + if (!bitmap->sb_page) { /* no superblock */ + spin_unlock_irqrestore(&bitmap->lock, flags); + return 0; + } + page_cache_get(bitmap->sb_page); + spin_unlock_irqrestore(&bitmap->lock, flags); + sb = (bitmap_super_t *)kmap(bitmap->sb_page); + sb->events = cpu_to_le64(bitmap->mddev->events); + if (!bitmap->mddev->degraded) + sb->events_cleared = cpu_to_le64(bitmap->mddev->events); + kunmap(bitmap->sb_page); + write_page(bitmap->sb_page, 0); + return 0; +} + +/* print out the bitmap file superblock */ +void bitmap_print_sb(struct bitmap *bitmap) +{ + bitmap_super_t *sb; + + if (!bitmap || !bitmap->sb_page) + return; + sb = (bitmap_super_t *)kmap(bitmap->sb_page); + printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); + printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); + printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); + printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", + *(__u32 *)(sb->uuid+0), + *(__u32 *)(sb->uuid+4), + *(__u32 *)(sb->uuid+8), + *(__u32 *)(sb->uuid+12)); + printk(KERN_DEBUG " events: %llu\n", + (unsigned long long) le64_to_cpu(sb->events)); + printk(KERN_DEBUG "events_clred: %llu\n", + (unsigned long long) le64_to_cpu(sb->events_cleared)); + printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); + printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); + printk(KERN_DEBUG "daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); + printk(KERN_DEBUG " sync size: %llu KB\n", le64_to_cpu(sb->sync_size)); + kunmap(bitmap->sb_page); +} + +/* read the superblock from the bitmap file and initialize some bitmap fields */ +static int bitmap_read_sb(struct bitmap *bitmap) +{ + char *reason = NULL; + bitmap_super_t *sb; + unsigned long chunksize, daemon_sleep; + unsigned long bytes_read; + unsigned long long events; + int err = -EINVAL; + + /* page 0 is the superblock, read it... */ + bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read); + if (IS_ERR(bitmap->sb_page)) { + err = PTR_ERR(bitmap->sb_page); + bitmap->sb_page = NULL; + return err; + } + + sb = (bitmap_super_t *)kmap(bitmap->sb_page); + + if (bytes_read < sizeof(*sb)) { /* short read */ + printk(KERN_INFO "%s: bitmap file superblock truncated\n", + bmname(bitmap)); + err = -ENOSPC; + goto out; + } + + chunksize = le32_to_cpu(sb->chunksize); + daemon_sleep = le32_to_cpu(sb->daemon_sleep); + + /* verify that the bitmap-specific fields are valid */ + if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) + reason = "bad magic"; + else if (sb->version != cpu_to_le32(BITMAP_MAJOR)) + reason = "unrecognized superblock version"; + else if (chunksize < 512 || chunksize > (1024 * 1024 * 4)) + reason = "bitmap chunksize out of range (512B - 4MB)"; + else if ((1 << ffz(~chunksize)) != chunksize) + reason = "bitmap chunksize not a power of 2"; + else if (daemon_sleep < 1 || daemon_sleep > 15) + reason = "daemon sleep period out of range"; + if (reason) { + printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", + bmname(bitmap), reason); + goto out; + } + + /* keep the array size field of the bitmap superblock up to date */ + sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); + + if (!bitmap->mddev->persistent) + goto success; + + /* + * if we have a persistent array superblock, compare the + * bitmap's UUID and event counter to the mddev's + */ + if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { + printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", + bmname(bitmap)); + goto out; + } + events = le64_to_cpu(sb->events); + if (events < bitmap->mddev->events) { + printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " + "-- forcing full recovery\n", bmname(bitmap), events, + (unsigned long long) bitmap->mddev->events); + sb->state |= BITMAP_STALE; + } +success: + /* assign fields using values from superblock */ + bitmap->chunksize = chunksize; + bitmap->daemon_sleep = daemon_sleep; + bitmap->flags |= sb->state; + bitmap->events_cleared = le64_to_cpu(sb->events_cleared); + err = 0; +out: + kunmap(bitmap->sb_page); + if (err) + bitmap_print_sb(bitmap); + return err; +} + +enum bitmap_mask_op { + MASK_SET, + MASK_UNSET +}; + +/* record the state of the bitmap in the superblock */ +static void bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, + enum bitmap_mask_op op) +{ + bitmap_super_t *sb; + unsigned long flags; + + spin_lock_irqsave(&bitmap->lock, flags); + if (!bitmap || !bitmap->sb_page) { /* can't set the state */ + spin_unlock_irqrestore(&bitmap->lock, flags); + return; + } + page_cache_get(bitmap->sb_page); + spin_unlock_irqrestore(&bitmap->lock, flags); + sb = (bitmap_super_t *)kmap(bitmap->sb_page); + switch (op) { + case MASK_SET: sb->state |= bits; + break; + case MASK_UNSET: sb->state &= ~bits; + break; + default: BUG(); + } + kunmap(bitmap->sb_page); + page_cache_release(bitmap->sb_page); +} + +/* + * general bitmap file operations + */ + +/* calculate the index of the page that contains this bit */ +static inline unsigned long file_page_index(unsigned long chunk) +{ + return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT; +} + +/* calculate the (bit) offset of this bit within a page */ +static inline unsigned long file_page_offset(unsigned long chunk) +{ + return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1); +} + +/* + * return a pointer to the page in the filemap that contains the given bit + * + * this lookup is complicated by the fact that the bitmap sb might be exactly + * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page + * 0 or page 1 + */ +static inline struct page *filemap_get_page(struct bitmap *bitmap, + unsigned long chunk) +{ + return bitmap->filemap[file_page_index(chunk) - file_page_index(0)]; +} + + +static void bitmap_file_unmap(struct bitmap *bitmap) +{ + struct page **map, *sb_page; + unsigned long *attr; + int pages; + unsigned long flags; + + spin_lock_irqsave(&bitmap->lock, flags); + map = bitmap->filemap; + bitmap->filemap = NULL; + attr = bitmap->filemap_attr; + bitmap->filemap_attr = NULL; + pages = bitmap->file_pages; + bitmap->file_pages = 0; + sb_page = bitmap->sb_page; + bitmap->sb_page = NULL; + spin_unlock_irqrestore(&bitmap->lock, flags); + + while (pages--) + if (map[pages]->index != 0) /* 0 is sb_page, release it below */ + page_cache_release(map[pages]); + kfree(map); + kfree(attr); + + if (sb_page) + page_cache_release(sb_page); +} + +static void bitmap_stop_daemons(struct bitmap *bitmap); + +/* dequeue the next item in a page list -- don't call from irq context */ +static struct page_list *dequeue_page(struct bitmap *bitmap, + struct list_head *head) +{ + struct page_list *item = NULL; + + spin_lock(&bitmap->write_lock); + if (list_empty(head)) + goto out; + item = list_entry(head->prev, struct page_list, list); + list_del(head->prev); +out: + spin_unlock(&bitmap->write_lock); + return item; +} + +static void drain_write_queues(struct bitmap *bitmap) +{ + struct list_head *queues[] = { &bitmap->complete_pages, NULL }; + struct list_head *head; + struct page_list *item; + int i; + + for (i = 0; queues[i]; i++) { + head = queues[i]; + while ((item = dequeue_page(bitmap, head))) { + page_cache_release(item->page); + mempool_free(item, bitmap->write_pool); + } + } + + spin_lock(&bitmap->write_lock); + bitmap->writes_pending = 0; /* make sure waiters continue */ + wake_up(&bitmap->write_wait); + spin_unlock(&bitmap->write_lock); +} + +static void bitmap_file_put(struct bitmap *bitmap) +{ + struct file *file; + struct inode *inode; + unsigned long flags; + + spin_lock_irqsave(&bitmap->lock, flags); + file = bitmap->file; + bitmap->file = NULL; + spin_unlock_irqrestore(&bitmap->lock, flags); + + bitmap_stop_daemons(bitmap); + + drain_write_queues(bitmap); + + bitmap_file_unmap(bitmap); + + if (file) { + inode = file->f_mapping->host; + spin_lock(&inode->i_lock); + atomic_set(&inode->i_writecount, 1); /* allow writes again */ + spin_unlock(&inode->i_lock); + fput(file); + } +} + + +/* + * bitmap_file_kick - if an error occurs while manipulating the bitmap file + * then it is no longer reliable, so we stop using it and we mark the file + * as failed in the superblock + */ +static void bitmap_file_kick(struct bitmap *bitmap) +{ + char *path, *ptr = NULL; + + bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET); + bitmap_update_sb(bitmap); + + path = kmalloc(PAGE_SIZE, GFP_KERNEL); + if (path) + ptr = file_path(bitmap->file, path, PAGE_SIZE); + + printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n", + bmname(bitmap), ptr ? ptr : ""); + + kfree(path); + + bitmap_file_put(bitmap); + + return; +} + +enum bitmap_page_attr { + BITMAP_PAGE_DIRTY = 1, // there are set bits that need to be synced + BITMAP_PAGE_CLEAN = 2, // there are bits that might need to be cleared + BITMAP_PAGE_NEEDWRITE=4, // there are cleared bits that need to be synced +}; + +static inline void set_page_attr(struct bitmap *bitmap, struct page *page, + enum bitmap_page_attr attr) +{ + bitmap->filemap_attr[page->index] |= attr; +} + +static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, + enum bitmap_page_attr attr) +{ + bitmap->filemap_attr[page->index] &= ~attr; +} + +static inline unsigned long get_page_attr(struct bitmap *bitmap, struct page *page) +{ + return bitmap->filemap_attr[page->index]; +} + +/* + * bitmap_file_set_bit -- called before performing a write to the md device + * to set (and eventually sync) a particular bit in the bitmap file + * + * we set the bit immediately, then we record the page number so that + * when an unplug occurs, we can flush the dirty pages out to disk + */ +static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) +{ + unsigned long bit; + struct page *page; + void *kaddr; + unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); + + if (!bitmap->file || !bitmap->filemap) { + return; + } + + page = filemap_get_page(bitmap, chunk); + bit = file_page_offset(chunk); + + + /* make sure the page stays cached until it gets written out */ + if (! (get_page_attr(bitmap, page) & BITMAP_PAGE_DIRTY)) + page_cache_get(page); + + /* set the bit */ + kaddr = kmap_atomic(page, KM_USER0); + set_bit(bit, kaddr); + kunmap_atomic(kaddr, KM_USER0); + PRINTK("set file bit %lu page %lu\n", bit, page->index); + + /* record page number so it gets flushed to disk when unplug occurs */ + set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); + +} + +/* this gets called when the md device is ready to unplug its underlying + * (slave) device queues -- before we let any writes go down, we need to + * sync the dirty pages of the bitmap file to disk */ +int bitmap_unplug(struct bitmap *bitmap) +{ + unsigned long i, attr, flags; + struct page *page; + int wait = 0; + + if (!bitmap) + return 0; + + /* look at each page to see if there are any set bits that need to be + * flushed out to disk */ + for (i = 0; i < bitmap->file_pages; i++) { + spin_lock_irqsave(&bitmap->lock, flags); + if (!bitmap->file || !bitmap->filemap) { + spin_unlock_irqrestore(&bitmap->lock, flags); + return 0; + } + page = bitmap->filemap[i]; + attr = get_page_attr(bitmap, page); + clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); + clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); + if ((attr & BITMAP_PAGE_DIRTY)) + wait = 1; + spin_unlock_irqrestore(&bitmap->lock, flags); + + if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) + write_page(page, 0); + } + if (wait) { /* if any writes were performed, we need to wait on them */ + spin_lock_irq(&bitmap->write_lock); + wait_event_lock_irq(bitmap->write_wait, + bitmap->writes_pending == 0, bitmap->write_lock, + wake_up_process(bitmap->writeback_daemon->tsk)); + spin_unlock_irq(&bitmap->write_lock); + } + return 0; +} + +static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, + unsigned long sectors, int set); +/* * bitmap_init_from_disk -- called at bitmap_create time to initialize + * the in-memory bitmap from the on-disk bitmap -- also, sets up the + * memory mapping of the bitmap file + * Special cases: + * if there's no bitmap file, or if the bitmap file had been + * previously kicked from the array, we mark all the bits as + * 1's in order to cause a full resync. + */ +static int bitmap_init_from_disk(struct bitmap *bitmap) +{ + unsigned long i, chunks, index, oldindex, bit; + struct page *page = NULL, *oldpage = NULL; + unsigned long num_pages, bit_cnt = 0; + struct file *file; + unsigned long bytes, offset, dummy; + int outofdate; + int ret = -ENOSPC; + + chunks = bitmap->chunks; + file = bitmap->file; + + if (!file) { /* no file, dirty all the in-memory bits */ + printk(KERN_INFO "%s: no bitmap file, doing full recovery\n", + bmname(bitmap)); + bitmap_set_memory_bits(bitmap, 0, + chunks << CHUNK_BLOCK_SHIFT(bitmap), 1); + return 0; + } + +#if INJECT_FAULTS_3 + outofdate = 1; +#else + outofdate = bitmap->flags & BITMAP_STALE; +#endif + if (outofdate) + printk(KERN_INFO "%s: bitmap file is out of date, doing full " + "recovery\n", bmname(bitmap)); + + bytes = (chunks + 7) / 8; + num_pages = (bytes + PAGE_SIZE - 1) / PAGE_SIZE; + if (i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { + printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", + bmname(bitmap), + (unsigned long) i_size_read(file->f_mapping->host), + bytes + sizeof(bitmap_super_t)); + goto out; + } + num_pages++; + bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); + if (!bitmap->filemap) { + ret = -ENOMEM; + goto out; + } + + bitmap->filemap_attr = kmalloc(sizeof(long) * num_pages, GFP_KERNEL); + if (!bitmap->filemap_attr) { + ret = -ENOMEM; + goto out; + } + + memset(bitmap->filemap_attr, 0, sizeof(long) * num_pages); + + oldindex = ~0L; + + for (i = 0; i < chunks; i++) { + index = file_page_index(i); + bit = file_page_offset(i); + if (index != oldindex) { /* this is a new page, read it in */ + /* unmap the old page, we're done with it */ + if (oldpage != NULL) + kunmap(oldpage); + if (index == 0) { + /* + * if we're here then the superblock page + * contains some bits (PAGE_SIZE != sizeof sb) + * we've already read it in, so just use it + */ + page = bitmap->sb_page; + offset = sizeof(bitmap_super_t); + } else { + page = read_page(file, index, &dummy); + if (IS_ERR(page)) { /* read error */ + ret = PTR_ERR(page); + goto out; + } + offset = 0; + } + oldindex = index; + oldpage = page; + kmap(page); + + if (outofdate) { + /* + * if bitmap is out of date, dirty the + * whole page and write it out + */ + memset(page_address(page) + offset, 0xff, + PAGE_SIZE - offset); + ret = write_page(page, 1); + if (ret) { + kunmap(page); + /* release, page not in filemap yet */ + page_cache_release(page); + goto out; + } + } + + bitmap->filemap[bitmap->file_pages++] = page; + } + if (test_bit(bit, page_address(page))) { + /* if the disk bit is set, set the memory bit */ + bitmap_set_memory_bits(bitmap, + i << CHUNK_BLOCK_SHIFT(bitmap), 1, 1); + bit_cnt++; + } +#if 0 + else + bitmap_set_memory_bits(bitmap, + i << CHUNK_BLOCK_SHIFT(bitmap), 1, 0); +#endif + } + + /* everything went OK */ + ret = 0; + bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); + + if (page) /* unmap the last page */ + kunmap(page); + + if (bit_cnt) { /* Kick recovery if any bits were set */ + set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); + md_wakeup_thread(bitmap->mddev->thread); + } + +out: + printk(KERN_INFO "%s: bitmap initialized from disk: " + "read %lu/%lu pages, set %lu bits, status: %d\n", + bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, ret); + + return ret; +} + + +static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) +{ + sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); + unsigned long page = chunk >> PAGE_COUNTER_SHIFT; + bitmap->bp[page].count += inc; +/* + if (page == 0) printk("count page 0, offset %llu: %d gives %d\n", + (unsigned long long)offset, inc, bitmap->bp[page].count); +*/ + bitmap_checkfree(bitmap, page); +} +static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, + sector_t offset, int *blocks, + int create); + +/* + * bitmap daemon -- periodically wakes up to clean bits and flush pages + * out to disk + */ + +int bitmap_daemon_work(struct bitmap *bitmap) +{ + unsigned long bit, j; + unsigned long flags; + struct page *page = NULL, *lastpage = NULL; + int err = 0; + int blocks; + int attr; + + if (bitmap == NULL) + return 0; + if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ)) + return 0; + bitmap->daemon_lastrun = jiffies; + + for (j = 0; j < bitmap->chunks; j++) { + bitmap_counter_t *bmc; + spin_lock_irqsave(&bitmap->lock, flags); + if (!bitmap->file || !bitmap->filemap) { + /* error or shutdown */ + spin_unlock_irqrestore(&bitmap->lock, flags); + break; + } + + page = filemap_get_page(bitmap, j); + /* skip this page unless it's marked as needing cleaning */ + if (!((attr=get_page_attr(bitmap, page)) & BITMAP_PAGE_CLEAN)) { + if (attr & BITMAP_PAGE_NEEDWRITE) { + page_cache_get(page); + clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); + } + spin_unlock_irqrestore(&bitmap->lock, flags); + if (attr & BITMAP_PAGE_NEEDWRITE) { + if (write_page(page, 0)) + bitmap_file_kick(bitmap); + page_cache_release(page); + } + continue; + } + + bit = file_page_offset(j); + + if (page != lastpage) { + /* grab the new page, sync and release the old */ + page_cache_get(page); + if (lastpage != NULL) { + if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) { + clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); + spin_unlock_irqrestore(&bitmap->lock, flags); + write_page(lastpage, 0); + } else { + set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); + spin_unlock_irqrestore(&bitmap->lock, flags); + } + kunmap(lastpage); + page_cache_release(lastpage); + if (err) + bitmap_file_kick(bitmap); + } else + spin_unlock_irqrestore(&bitmap->lock, flags); + lastpage = page; + kmap(page); +/* + printk("bitmap clean at page %lu\n", j); +*/ + spin_lock_irqsave(&bitmap->lock, flags); + clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); + } + bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), + &blocks, 0); + if (bmc) { +/* + if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); +*/ + if (*bmc == 2) { + *bmc=1; /* maybe clear the bit next time */ + set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); + } else if (*bmc == 1) { + /* we can clear the bit */ + *bmc = 0; + bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), + -1); + + /* clear the bit */ + clear_bit(bit, page_address(page)); + } + } + spin_unlock_irqrestore(&bitmap->lock, flags); + } + + /* now sync the final page */ + if (lastpage != NULL) { + kunmap(lastpage); + spin_lock_irqsave(&bitmap->lock, flags); + if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) { + clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); + spin_unlock_irqrestore(&bitmap->lock, flags); + write_page(lastpage, 0); + } else { + set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); + spin_unlock_irqrestore(&bitmap->lock, flags); + } + + page_cache_release(lastpage); + } + + return err; +} + +static void daemon_exit(struct bitmap *bitmap, mdk_thread_t **daemon) +{ + mdk_thread_t *dmn; + unsigned long flags; + + /* if no one is waiting on us, we'll free the md thread struct + * and exit, otherwise we let the waiter clean things up */ + spin_lock_irqsave(&bitmap->lock, flags); + if ((dmn = *daemon)) { /* no one is waiting, cleanup and exit */ + *daemon = NULL; + spin_unlock_irqrestore(&bitmap->lock, flags); + kfree(dmn); + complete_and_exit(NULL, 0); /* do_exit not exported */ + } + spin_unlock_irqrestore(&bitmap->lock, flags); +} + +static void bitmap_writeback_daemon(mddev_t *mddev) +{ + struct bitmap *bitmap = mddev->bitmap; + struct page *page; + struct page_list *item; + int err = 0; + + while (1) { + PRINTK("%s: bitmap writeback daemon waiting...\n", bmname(bitmap)); + down_interruptible(&bitmap->write_done); + if (signal_pending(current)) { + printk(KERN_INFO + "%s: bitmap writeback daemon got signal, exiting...\n", + bmname(bitmap)); + break; + } + + PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap)); + /* wait on bitmap page writebacks */ + while ((item = dequeue_page(bitmap, &bitmap->complete_pages))) { + page = item->page; + mempool_free(item, bitmap->write_pool); + PRINTK("wait on page writeback: %p %lu\n", page, bitmap->writes_pending); + wait_on_page_writeback(page); + PRINTK("finished page writeback: %p %lu\n", page, bitmap->writes_pending); + spin_lock(&bitmap->write_lock); + if (!--bitmap->writes_pending) + wake_up(&bitmap->write_wait); + spin_unlock(&bitmap->write_lock); + err = PageError(page); + page_cache_release(page); + if (err) { + printk(KERN_WARNING "%s: bitmap file writeback " + "failed (page %lu): %d\n", + bmname(bitmap), page->index, err); + bitmap_file_kick(bitmap); + goto out; + } + } + } +out: + if (err) { + printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n", + bmname(bitmap), err); + daemon_exit(bitmap, &bitmap->writeback_daemon); + } + return; +} + +static int bitmap_start_daemon(struct bitmap *bitmap, mdk_thread_t **ptr, + void (*func)(mddev_t *), char *name) +{ + mdk_thread_t *daemon; + unsigned long flags; + char namebuf[32]; + + spin_lock_irqsave(&bitmap->lock, flags); + *ptr = NULL; + if (!bitmap->file) /* no need for daemon if there's no backing file */ + goto out_unlock; + + spin_unlock_irqrestore(&bitmap->lock, flags); + +#if INJECT_FATAL_FAULT_2 + daemon = NULL; +#else + sprintf(namebuf, "%%s_%s", name); + daemon = md_register_thread(func, bitmap->mddev, namebuf); +#endif + if (!daemon) { + printk(KERN_ERR "%s: failed to start bitmap daemon\n", + bmname(bitmap)); + return -ECHILD; + } + + spin_lock_irqsave(&bitmap->lock, flags); + *ptr = daemon; + + md_wakeup_thread(daemon); /* start it running */ + + PRINTK("%s: %s daemon (pid %d) started...\n", + bmname(bitmap), name, bitmap->daemon->tsk->pid); +out_unlock: + spin_unlock_irqrestore(&bitmap->lock, flags); + return 0; +} + +static int bitmap_start_daemons(struct bitmap *bitmap) +{ + int err = bitmap_start_daemon(bitmap, &bitmap->writeback_daemon, + bitmap_writeback_daemon, "bitmap_wb"); + return err; +} + +static void bitmap_stop_daemon(struct bitmap *bitmap, mdk_thread_t **ptr) +{ + mdk_thread_t *daemon; + unsigned long flags; + + spin_lock_irqsave(&bitmap->lock, flags); + daemon = *ptr; + *ptr = NULL; + spin_unlock_irqrestore(&bitmap->lock, flags); + if (daemon) + md_unregister_thread(daemon); /* destroy the thread */ +} + +static void bitmap_stop_daemons(struct bitmap *bitmap) +{ + /* the daemons can't stop themselves... they'll just exit instead... */ + if (bitmap->writeback_daemon && + current->pid != bitmap->writeback_daemon->tsk->pid) + bitmap_stop_daemon(bitmap, &bitmap->writeback_daemon); +} + +static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, + sector_t offset, int *blocks, + int create) +{ + /* If 'create', we might release the lock and reclaim it. + * The lock must have been taken with interrupts enabled. + * If !create, we don't release the lock. + */ + sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); + unsigned long page = chunk >> PAGE_COUNTER_SHIFT; + unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; + sector_t csize; + + if (bitmap_checkpage(bitmap, page, create) < 0) { + csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); + *blocks = csize - (offset & (csize- 1)); + return NULL; + } + /* now locked ... */ + + if (bitmap->bp[page].hijacked) { /* hijacked pointer */ + /* should we use the first or second counter field + * of the hijacked pointer? */ + int hi = (pageoff > PAGE_COUNTER_MASK); + csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + + PAGE_COUNTER_SHIFT - 1); + *blocks = csize - (offset & (csize- 1)); + return &((bitmap_counter_t *) + &bitmap->bp[page].map)[hi]; + } else { /* page is allocated */ + csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); + *blocks = csize - (offset & (csize- 1)); + return (bitmap_counter_t *) + &(bitmap->bp[page].map[pageoff]); + } +} + +int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors) +{ + if (!bitmap) return 0; + while (sectors) { + int blocks; + bitmap_counter_t *bmc; + + spin_lock_irq(&bitmap->lock); + bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); + if (!bmc) { + spin_unlock_irq(&bitmap->lock); + return 0; + } + + switch(*bmc) { + case 0: + bitmap_file_set_bit(bitmap, offset); + bitmap_count_page(bitmap,offset, 1); + blk_plug_device(bitmap->mddev->queue); + /* fall through */ + case 1: + *bmc = 2; + } + if ((*bmc & COUNTER_MAX) == COUNTER_MAX) BUG(); + (*bmc)++; + + spin_unlock_irq(&bitmap->lock); + + offset += blocks; + if (sectors > blocks) + sectors -= blocks; + else sectors = 0; + } + return 0; +} + +void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, + int success) +{ + if (!bitmap) return; + while (sectors) { + int blocks; + unsigned long flags; + bitmap_counter_t *bmc; + + spin_lock_irqsave(&bitmap->lock, flags); + bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); + if (!bmc) { + spin_unlock_irqrestore(&bitmap->lock, flags); + return; + } + + if (!success && ! (*bmc & NEEDED_MASK)) + *bmc |= NEEDED_MASK; + + (*bmc)--; + if (*bmc <= 2) { + set_page_attr(bitmap, + filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), + BITMAP_PAGE_CLEAN); + } + spin_unlock_irqrestore(&bitmap->lock, flags); + offset += blocks; + if (sectors > blocks) + sectors -= blocks; + else sectors = 0; + } +} + +int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks) +{ + bitmap_counter_t *bmc; + int rv; + if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ + *blocks = 1024; + return 1; /* always resync if no bitmap */ + } + spin_lock_irq(&bitmap->lock); + bmc = bitmap_get_counter(bitmap, offset, blocks, 0); + rv = 0; + if (bmc) { + /* locked */ + if (RESYNC(*bmc)) + rv = 1; + else if (NEEDED(*bmc)) { + rv = 1; + *bmc |= RESYNC_MASK; + *bmc &= ~NEEDED_MASK; + } + } + spin_unlock_irq(&bitmap->lock); + return rv; +} + +void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) +{ + bitmap_counter_t *bmc; + unsigned long flags; +/* + if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); +*/ if (bitmap == NULL) { + *blocks = 1024; + return; + } + spin_lock_irqsave(&bitmap->lock, flags); + bmc = bitmap_get_counter(bitmap, offset, blocks, 0); + if (bmc == NULL) + goto unlock; + /* locked */ +/* + if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); +*/ + if (RESYNC(*bmc)) { + *bmc &= ~RESYNC_MASK; + + if (!NEEDED(*bmc) && aborted) + *bmc |= NEEDED_MASK; + else { + if (*bmc <= 2) { + set_page_attr(bitmap, + filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), + BITMAP_PAGE_CLEAN); + } + } + } + unlock: + spin_unlock_irqrestore(&bitmap->lock, flags); +} + +void bitmap_close_sync(struct bitmap *bitmap) +{ + /* Sync has finished, and any bitmap chunks that weren't synced + * properly have been aborted. It remains to us to clear the + * RESYNC bit wherever it is still on + */ + sector_t sector = 0; + int blocks; + if (!bitmap) return; + while (sector < bitmap->mddev->resync_max_sectors) { + bitmap_end_sync(bitmap, sector, &blocks, 0); +/* + if (sector < 500) printk("bitmap_close_sync: sec %llu blks %d\n", + (unsigned long long)sector, blocks); +*/ sector += blocks; + } +} + +static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, + unsigned long sectors, int set) +{ + /* For each chunk covered by any of these sectors, set the + * resync needed bit, and the counter to 1. They should all + * be 0 at this point + */ + while (sectors) { + int secs; + bitmap_counter_t *bmc; + spin_lock_irq(&bitmap->lock); + bmc = bitmap_get_counter(bitmap, offset, &secs, 1); + if (!bmc) { + spin_unlock_irq(&bitmap->lock); + return; + } + if (set && !NEEDED(*bmc)) { + BUG_ON(*bmc); + *bmc = NEEDED_MASK | 1; + bitmap_count_page(bitmap, offset, 1); + } + spin_unlock_irq(&bitmap->lock); + if (sectors > secs) + sectors -= secs; + else + sectors = 0; + } +} + +/* dirty the entire bitmap */ +int bitmap_setallbits(struct bitmap *bitmap) +{ + unsigned long flags; + unsigned long j; + + /* dirty the in-memory bitmap */ + bitmap_set_memory_bits(bitmap, 0, bitmap->chunks << CHUNK_BLOCK_SHIFT(bitmap), 1); + + /* dirty the bitmap file */ + for (j = 0; j < bitmap->file_pages; j++) { + struct page *page = bitmap->filemap[j]; + + spin_lock_irqsave(&bitmap->lock, flags); + page_cache_get(page); + spin_unlock_irqrestore(&bitmap->lock, flags); + memset(kmap(page), 0xff, PAGE_SIZE); + kunmap(page); + write_page(page, 0); + } + + return 0; +} + +/* + * free memory that was allocated + */ +void bitmap_destroy(mddev_t *mddev) +{ + unsigned long k, pages; + struct bitmap_page *bp; + struct bitmap *bitmap = mddev->bitmap; + + if (!bitmap) /* there was no bitmap */ + return; + + mddev->bitmap = NULL; /* disconnect from the md device */ + + /* release the bitmap file and kill the daemon */ + bitmap_file_put(bitmap); + + bp = bitmap->bp; + pages = bitmap->pages; + + /* free all allocated memory */ + + mempool_destroy(bitmap->write_pool); + + if (bp) /* deallocate the page memory */ + for (k = 0; k < pages; k++) + if (bp[k].map && !bp[k].hijacked) + kfree(bp[k].map); + kfree(bp); + kfree(bitmap); +} + +/* + * initialize the bitmap structure + * if this returns an error, bitmap_destroy must be called to do clean up + */ +int bitmap_create(mddev_t *mddev) +{ + struct bitmap *bitmap; + unsigned long blocks = mddev->resync_max_sectors; + unsigned long chunks; + unsigned long pages; + struct file *file = mddev->bitmap_file; + int err; + + BUG_ON(sizeof(bitmap_super_t) != 256); + + if (!file) /* bitmap disabled, nothing to do */ + return 0; + + bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL); + if (!bitmap) + return -ENOMEM; + + memset(bitmap, 0, sizeof(*bitmap)); + + spin_lock_init(&bitmap->lock); + bitmap->mddev = mddev; + mddev->bitmap = bitmap; + + spin_lock_init(&bitmap->write_lock); + init_MUTEX_LOCKED(&bitmap->write_done); + INIT_LIST_HEAD(&bitmap->complete_pages); + init_waitqueue_head(&bitmap->write_wait); + bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc, + write_pool_free, NULL); + if (!bitmap->write_pool) + return -ENOMEM; + + bitmap->file = file; + get_file(file); + /* read superblock from bitmap file (this sets bitmap->chunksize) */ + err = bitmap_read_sb(bitmap); + if (err) + return err; + + bitmap->chunkshift = find_first_bit(&bitmap->chunksize, + sizeof(bitmap->chunksize)); + + /* now that chunksize and chunkshift are set, we can use these macros */ + chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) / + CHUNK_BLOCK_RATIO(bitmap); + pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; + + BUG_ON(!pages); + + bitmap->chunks = chunks; + bitmap->pages = pages; + bitmap->missing_pages = pages; + bitmap->counter_bits = COUNTER_BITS; + + bitmap->syncchunk = ~0UL; + +#if INJECT_FATAL_FAULT_1 + bitmap->bp = NULL; +#else + bitmap->bp = kmalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); +#endif + if (!bitmap->bp) + return -ENOMEM; + memset(bitmap->bp, 0, pages * sizeof(*bitmap->bp)); + + bitmap->flags |= BITMAP_ACTIVE; + + /* now that we have some pages available, initialize the in-memory + * bitmap from the on-disk bitmap */ + err = bitmap_init_from_disk(bitmap); + if (err) + return err; + + printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", + pages, bmname(bitmap)); + + /* kick off the bitmap daemons */ + err = bitmap_start_daemons(bitmap); + if (err) + return err; + return bitmap_update_sb(bitmap); +} + +/* the bitmap API -- for raid personalities */ +EXPORT_SYMBOL(bitmap_startwrite); +EXPORT_SYMBOL(bitmap_endwrite); +EXPORT_SYMBOL(bitmap_start_sync); +EXPORT_SYMBOL(bitmap_end_sync); +EXPORT_SYMBOL(bitmap_unplug); +EXPORT_SYMBOL(bitmap_close_sync); +EXPORT_SYMBOL(bitmap_daemon_work); diff --git a/drivers/md/md.c b/drivers/md/md.c index fa608a1a5c2..c402f6cc704 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -19,6 +19,9 @@ Neil Brown . + - persistent bitmap code + Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. + 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, or (at your option) @@ -33,6 +36,7 @@ #include #include #include +#include #include #include #include /* for invalidate_bdev */ @@ -40,6 +44,8 @@ #include +#include + #ifdef CONFIG_KMOD #include #endif @@ -1198,8 +1204,11 @@ void md_print_devices(void) printk("md: * *\n"); printk("md: **********************************\n"); ITERATE_MDDEV(mddev,tmp) { - printk("%s: ", mdname(mddev)); + if (mddev->bitmap) + bitmap_print_sb(mddev->bitmap); + else + printk("%s: ", mdname(mddev)); ITERATE_RDEV(mddev,rdev,tmp2) printk("<%s>", bdevname(rdev->bdev,b)); printk("\n"); @@ -1287,7 +1296,7 @@ repeat: "md: updating %s RAID superblock on device (in sync %d)\n", mdname(mddev),mddev->in_sync); - err = 0; + err = bitmap_update_sb(mddev->bitmap); ITERATE_RDEV(mddev,rdev,tmp) { char b[BDEVNAME_SIZE]; dprintk(KERN_INFO "md: "); @@ -1624,12 +1633,19 @@ static int do_md_run(mddev_t * mddev) mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ - err = mddev->pers->run(mddev); + /* before we start the array running, initialise the bitmap */ + err = bitmap_create(mddev); + if (err) + printk(KERN_ERR "%s: failed to create bitmap (%d)\n", + mdname(mddev), err); + else + err = mddev->pers->run(mddev); if (err) { printk(KERN_ERR "md: pers->run() failed ...\n"); module_put(mddev->pers->owner); mddev->pers = NULL; - return -EINVAL; + bitmap_destroy(mddev); + return err; } atomic_set(&mddev->writes_pending,0); mddev->safemode = 0; @@ -1742,6 +1758,14 @@ static int do_md_stop(mddev_t * mddev, int ro) if (ro) set_disk_ro(disk, 1); } + + bitmap_destroy(mddev); + if (mddev->bitmap_file) { + atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1); + fput(mddev->bitmap_file); + mddev->bitmap_file = NULL; + } + /* * Free resources if final stop */ @@ -2000,6 +2024,42 @@ static int get_array_info(mddev_t * mddev, void __user * arg) return 0; } +static int get_bitmap_file(mddev_t * mddev, void * arg) +{ + mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */ + char *ptr, *buf = NULL; + int err = -ENOMEM; + + file = kmalloc(sizeof(*file), GFP_KERNEL); + if (!file) + goto out; + + /* bitmap disabled, zero the first byte and copy out */ + if (!mddev->bitmap || !mddev->bitmap->file) { + file->pathname[0] = '\0'; + goto copy_out; + } + + buf = kmalloc(sizeof(file->pathname), GFP_KERNEL); + if (!buf) + goto out; + + ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname)); + if (!ptr) + goto out; + + strcpy(file->pathname, ptr); + +copy_out: + err = 0; + if (copy_to_user(arg, file, sizeof(*file))) + err = -EFAULT; +out: + kfree(buf); + kfree(file); + return err; +} + static int get_disk_info(mddev_t * mddev, void __user * arg) { mdu_disk_info_t info; @@ -2275,6 +2335,48 @@ abort_export: return err; } +/* similar to deny_write_access, but accounts for our holding a reference + * to the file ourselves */ +static int deny_bitmap_write_access(struct file * file) +{ + struct inode *inode = file->f_mapping->host; + + spin_lock(&inode->i_lock); + if (atomic_read(&inode->i_writecount) > 1) { + spin_unlock(&inode->i_lock); + return -ETXTBSY; + } + atomic_set(&inode->i_writecount, -1); + spin_unlock(&inode->i_lock); + + return 0; +} + +static int set_bitmap_file(mddev_t *mddev, int fd) +{ + int err; + + if (mddev->pers) + return -EBUSY; + + mddev->bitmap_file = fget(fd); + + if (mddev->bitmap_file == NULL) { + printk(KERN_ERR "%s: error: failed to get bitmap file\n", + mdname(mddev)); + return -EBADF; + } + + err = deny_bitmap_write_access(mddev->bitmap_file); + if (err) { + printk(KERN_ERR "%s: error: bitmap file is already in use\n", + mdname(mddev)); + fput(mddev->bitmap_file); + mddev->bitmap_file = NULL; + } + return err; +} + /* * set_array_info is used two different ways * The original usage is when creating a new array. @@ -2586,8 +2688,10 @@ static int md_ioctl(struct inode *inode, struct file *file, /* * Commands querying/configuring an existing array: */ - /* if we are initialised yet, only ADD_NEW_DISK or STOP_ARRAY is allowed */ - if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY && cmd != RUN_ARRAY) { + /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY, + * RUN_ARRAY, and SET_BITMAP_FILE are allowed */ + if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY + && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) { err = -ENODEV; goto abort_unlock; } @@ -2601,6 +2705,10 @@ static int md_ioctl(struct inode *inode, struct file *file, err = get_array_info(mddev, argp); goto done_unlock; + case GET_BITMAP_FILE: + err = get_bitmap_file(mddev, (void *)arg); + goto done_unlock; + case GET_DISK_INFO: err = get_disk_info(mddev, argp); goto done_unlock; @@ -2681,6 +2789,10 @@ static int md_ioctl(struct inode *inode, struct file *file, err = do_md_run (mddev); goto done_unlock; + case SET_BITMAP_FILE: + err = set_bitmap_file(mddev, (int)arg); + goto done_unlock; + default: if (_IOC_TYPE(cmd) == MD_MAJOR) printk(KERN_WARNING "md: %s(pid %d) used" @@ -2792,8 +2904,9 @@ static int md_thread(void * arg) while (thread->run) { void (*run)(mddev_t *); - wait_event_interruptible(thread->wqueue, - test_bit(THREAD_WAKEUP, &thread->flags)); + wait_event_interruptible_timeout(thread->wqueue, + test_bit(THREAD_WAKEUP, &thread->flags), + thread->timeout); if (current->flags & PF_FREEZE) refrigerator(PF_FREEZE); @@ -2839,6 +2952,7 @@ mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, thread->run = run; thread->mddev = mddev; thread->name = name; + thread->timeout = MAX_SCHEDULE_TIMEOUT; ret = kernel_thread(md_thread, thread, 0); if (ret < 0) { kfree(thread); @@ -2877,13 +2991,13 @@ void md_error(mddev_t *mddev, mdk_rdev_t *rdev) if (!rdev || rdev->faulty) return; - +/* dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", mdname(mddev), MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), __builtin_return_address(0),__builtin_return_address(1), __builtin_return_address(2),__builtin_return_address(3)); - +*/ if (!mddev->pers->error_handler) return; mddev->pers->error_handler(mddev,rdev); @@ -3037,6 +3151,7 @@ static int md_seq_show(struct seq_file *seq, void *v) struct list_head *tmp2; mdk_rdev_t *rdev; int i; + struct bitmap *bitmap; if (v == (void*)1) { seq_printf(seq, "Personalities : "); @@ -3089,10 +3204,36 @@ static int md_seq_show(struct seq_file *seq, void *v) if (mddev->pers) { mddev->pers->status (seq, mddev); seq_printf(seq, "\n "); - if (mddev->curr_resync > 2) + if (mddev->curr_resync > 2) { status_resync (seq, mddev); - else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) - seq_printf(seq, " resync=DELAYED"); + seq_printf(seq, "\n "); + } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) + seq_printf(seq, " resync=DELAYED\n "); + } else + seq_printf(seq, "\n "); + + if ((bitmap = mddev->bitmap)) { + char *buf, *path; + unsigned long chunk_kb; + unsigned long flags; + buf = kmalloc(PAGE_SIZE, GFP_KERNEL); + spin_lock_irqsave(&bitmap->lock, flags); + chunk_kb = bitmap->chunksize >> 10; + seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " + "%lu%s chunk", + bitmap->pages - bitmap->missing_pages, + bitmap->pages, + (bitmap->pages - bitmap->missing_pages) + << (PAGE_SHIFT - 10), + chunk_kb ? chunk_kb : bitmap->chunksize, + chunk_kb ? "KB" : "B"); + if (bitmap->file && buf) { + path = file_path(bitmap->file, buf, PAGE_SIZE); + seq_printf(seq, ", file: %s", path ? path : ""); + } + seq_printf(seq, "\n"); + spin_unlock_irqrestore(&bitmap->lock, flags); + kfree(buf); } seq_printf(seq, "\n"); @@ -3328,7 +3469,8 @@ static void md_do_sync(mddev_t *mddev) sysctl_speed_limit_max); is_mddev_idle(mddev); /* this also initializes IO event counters */ - if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) + /* we don't use the checkpoint if there's a bitmap */ + if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap) j = mddev->recovery_cp; else j = 0; @@ -3673,6 +3815,8 @@ static int __init md_init(void) " MD_SB_DISKS=%d\n", MD_MAJOR_VERSION, MD_MINOR_VERSION, MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS); + printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR, + BITMAP_MINOR); if (register_blkdev(MAJOR_NR, "md")) return -1; diff --git a/include/linux/raid/bitmap.h b/include/linux/raid/bitmap.h new file mode 100644 index 00000000000..f785cf26cba --- /dev/null +++ b/include/linux/raid/bitmap.h @@ -0,0 +1,280 @@ +/* + * bitmap.h: Copyright (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 + * + * additions: Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. + */ +#ifndef BITMAP_H +#define BITMAP_H 1 + +#define BITMAP_MAJOR 3 +#define BITMAP_MINOR 38 + +/* + * in-memory bitmap: + * + * Use 16 bit block counters to track pending writes to each "chunk". + * The 2 high order bits are special-purpose, the first is a flag indicating + * whether a resync is needed. The second is a flag indicating whether a + * resync is active. + * This means that the counter is actually 14 bits: + * + * +--------+--------+------------------------------------------------+ + * | resync | resync | counter | + * | needed | active | | + * | (0-1) | (0-1) | (0-16383) | + * +--------+--------+------------------------------------------------+ + * + * The "resync needed" bit is set when: + * a '1' bit is read from storage at startup. + * a write request fails on some drives + * a resync is aborted on a chunk with 'resync active' set + * It is cleared (and resync-active set) when a resync starts across all drives + * of the chunk. + * + * + * The "resync active" bit is set when: + * a resync is started on all drives, and resync_needed is set. + * resync_needed will be cleared (as long as resync_active wasn't already set). + * It is cleared when a resync completes. + * + * The counter counts pending write requests, plus the on-disk bit. + * When the counter is '1' and the resync bits are clear, the on-disk + * bit can be cleared aswell, thus setting the counter to 0. + * When we set a bit, or in the counter (to start a write), if the fields is + * 0, we first set the disk bit and set the counter to 1. + * + * If the counter is 0, the on-disk bit is clear and the stipe is clean + * Anything that dirties the stipe pushes the counter to 2 (at least) + * and sets the on-disk bit (lazily). + * If a periodic sweep find the counter at 2, it is decremented to 1. + * If the sweep find the counter at 1, the on-disk bit is cleared and the + * counter goes to zero. + * + * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block + * counters as a fallback when "page" memory cannot be allocated: + * + * Normal case (page memory allocated): + * + * page pointer (32-bit) + * + * [ ] ------+ + * | + * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters) + * c1 c2 c2048 + * + * Hijacked case (page memory allocation failed): + * + * hijacked page pointer (32-bit) + * + * [ ][ ] (no page memory allocated) + * counter #1 (16-bit) counter #2 (16-bit) + * + */ + +#ifdef __KERNEL__ + +#define PAGE_BITS (PAGE_SIZE << 3) +#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3) + +typedef __u16 bitmap_counter_t; +#define COUNTER_BITS 16 +#define COUNTER_BIT_SHIFT 4 +#define COUNTER_BYTE_RATIO (COUNTER_BITS / 8) +#define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3) + +#define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1))) +#define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2))) +#define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1) +#define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK) +#define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK) +#define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX) + +/* how many counters per page? */ +#define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS) +/* same, except a shift value for more efficient bitops */ +#define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT) +/* same, except a mask value for more efficient bitops */ +#define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1) + +#define BITMAP_BLOCK_SIZE 512 +#define BITMAP_BLOCK_SHIFT 9 + +/* how many blocks per chunk? (this is variable) */ +#define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT) +#define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT) +#define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1) + +/* when hijacked, the counters and bits represent even larger "chunks" */ +/* there will be 1024 chunks represented by each counter in the page pointers */ +#define PAGEPTR_BLOCK_RATIO(bitmap) \ + (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1) +#define PAGEPTR_BLOCK_SHIFT(bitmap) \ + (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1) +#define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1) + +/* + * on-disk bitmap: + * + * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap + * file a page at a time. There's a superblock at the start of the file. + */ + +/* map chunks (bits) to file pages - offset by the size of the superblock */ +#define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3)) + +#endif + +/* + * bitmap structures: + */ + +#define BITMAP_MAGIC 0x6d746962 + +/* use these for bitmap->flags and bitmap->sb->state bit-fields */ +enum bitmap_state { + BITMAP_ACTIVE = 0x001, /* the bitmap is in use */ + BITMAP_STALE = 0x002 /* the bitmap file is out of date or had -EIO */ +}; + +/* the superblock at the front of the bitmap file -- little endian */ +typedef struct bitmap_super_s { + __u32 magic; /* 0 BITMAP_MAGIC */ + __u32 version; /* 4 the bitmap major for now, could change... */ + __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ + __u64 events; /* 24 event counter for the bitmap (1)*/ + __u64 events_cleared;/*32 event counter when last bit cleared (2) */ + __u64 sync_size; /* 40 the size of the md device's sync range(3) */ + __u32 state; /* 48 bitmap state information */ + __u32 chunksize; /* 52 the bitmap chunk size in bytes */ + __u32 daemon_sleep; /* 56 seconds between disk flushes */ + + __u8 pad[256 - 60]; /* set to zero */ +} bitmap_super_t; + +/* notes: + * (1) This event counter is updated before the eventcounter in the md superblock + * When a bitmap is loaded, it is only accepted if this event counter is equal + * to, or one greater than, the event counter in the superblock. + * (2) This event counter is updated when the other one is *if*and*only*if* the + * array is not degraded. As bits are not cleared when the array is degraded, + * this represents the last time that any bits were cleared. + * If a device is being added that has an event count with this value or + * higher, it is accepted as conforming to the bitmap. + * (3)This is the number of sectors represented by the bitmap, and is the range that + * resync happens across. For raid1 and raid5/6 it is the size of individual + * devices. For raid10 it is the size of the array. + */ + +#ifdef __KERNEL__ + +/* the in-memory bitmap is represented by bitmap_pages */ +struct bitmap_page { + /* + * map points to the actual memory page + */ + char *map; + /* + * in emergencies (when map cannot be alloced), hijack the map + * pointer and use it as two counters itself + */ + unsigned int hijacked:1; + /* + * count of dirty bits on the page + */ + unsigned int count:31; +}; + +/* keep track of bitmap file pages that have pending writes on them */ +struct page_list { + struct list_head list; + struct page *page; +}; + +/* the main bitmap structure - one per mddev */ +struct bitmap { + struct bitmap_page *bp; + unsigned long pages; /* total number of pages in the bitmap */ + unsigned long missing_pages; /* number of pages not yet allocated */ + + mddev_t *mddev; /* the md device that the bitmap is for */ + + int counter_bits; /* how many bits per block counter */ + + /* bitmap chunksize -- how much data does each bit represent? */ + unsigned long chunksize; + unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */ + unsigned long chunks; /* total number of data chunks for the array */ + + /* We hold a count on the chunk currently being synced, and drop + * it when the last block is started. If the resync is aborted + * midway, we need to be able to drop that count, so we remember + * the counted chunk.. + */ + unsigned long syncchunk; + + __u64 events_cleared; + + /* bitmap spinlock */ + spinlock_t lock; + + struct file *file; /* backing disk file */ + struct page *sb_page; /* cached copy of the bitmap file superblock */ + struct page **filemap; /* list of cache pages for the file */ + unsigned long *filemap_attr; /* attributes associated w/ filemap pages */ + unsigned long file_pages; /* number of pages in the file */ + + unsigned long flags; + + /* + * the bitmap daemon - periodically wakes up and sweeps the bitmap + * file, cleaning up bits and flushing out pages to disk as necessary + */ + unsigned long daemon_lastrun; /* jiffies of last run */ + unsigned long daemon_sleep; /* how many seconds between updates? */ + + /* + * bitmap write daemon - this daemon performs writes to the bitmap file + * this thread is only needed because of a limitation in ext3 (jbd) + * that does not allow a task to have two journal transactions ongoing + * simultaneously (even if the transactions are for two different + * filesystems) -- in the case of bitmap, that would be the filesystem + * that the bitmap file resides on and the filesystem that is mounted + * on the md device -- see current->journal_info in jbd/transaction.c + */ + mdk_thread_t *writeback_daemon; + spinlock_t write_lock; + struct semaphore write_ready; + struct semaphore write_done; + unsigned long writes_pending; + wait_queue_head_t write_wait; + struct list_head write_pages; + struct list_head complete_pages; + mempool_t *write_pool; +}; + +/* the bitmap API */ + +/* these are used only by md/bitmap */ +int bitmap_create(mddev_t *mddev); +void bitmap_destroy(mddev_t *mddev); +int bitmap_active(struct bitmap *bitmap); + +char *file_path(struct file *file, char *buf, int count); +void bitmap_print_sb(struct bitmap *bitmap); +int bitmap_update_sb(struct bitmap *bitmap); + +int bitmap_setallbits(struct bitmap *bitmap); + +/* these are exported */ +int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors); +void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, + int success); +int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks); +void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted); +void bitmap_close_sync(struct bitmap *bitmap); + +int bitmap_unplug(struct bitmap *bitmap); +int bitmap_daemon_work(struct bitmap *bitmap); +#endif + +#endif diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index bce0032decf..16e94a9f0f8 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -267,6 +267,9 @@ struct mddev_s atomic_t writes_pending; request_queue_t *queue; /* for plugging ... */ + struct bitmap *bitmap; /* the bitmap for the device */ + struct file *bitmap_file; /* the bitmap file */ + struct list_head all_mddevs; }; @@ -341,6 +344,7 @@ typedef struct mdk_thread_s { unsigned long flags; struct completion *event; struct task_struct *tsk; + unsigned long timeout; const char *name; } mdk_thread_t; diff --git a/include/linux/raid/md_u.h b/include/linux/raid/md_u.h index a2df5c2a42a..81da20ccec4 100644 --- a/include/linux/raid/md_u.h +++ b/include/linux/raid/md_u.h @@ -23,6 +23,7 @@ #define GET_DISK_INFO _IOR (MD_MAJOR, 0x12, mdu_disk_info_t) #define PRINT_RAID_DEBUG _IO (MD_MAJOR, 0x13) #define RAID_AUTORUN _IO (MD_MAJOR, 0x14) +#define GET_BITMAP_FILE _IOR (MD_MAJOR, 0x15, mdu_bitmap_file_t) /* configuration */ #define CLEAR_ARRAY _IO (MD_MAJOR, 0x20) @@ -36,6 +37,7 @@ #define HOT_ADD_DISK _IO (MD_MAJOR, 0x28) #define SET_DISK_FAULTY _IO (MD_MAJOR, 0x29) #define HOT_GENERATE_ERROR _IO (MD_MAJOR, 0x2a) +#define SET_BITMAP_FILE _IOW (MD_MAJOR, 0x2b, int) /* usage */ #define RUN_ARRAY _IOW (MD_MAJOR, 0x30, mdu_param_t) @@ -106,6 +108,11 @@ typedef struct mdu_start_info_s { } mdu_start_info_t; +typedef struct mdu_bitmap_file_s +{ + char pathname[4096]; +} mdu_bitmap_file_t; + typedef struct mdu_param_s { int personality; /* 1,2,3,4 */ -- cgit v1.2.3-70-g09d2 From 77ad4bc706fe6c52ab953f31c287a6af712d080c Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:21 -0700 Subject: [PATCH] md: enable the bitmap write-back daemon and wait for it. Currently we don't wait for updates to the bitmap to be flushed to disk properly. The infrastructure all there, but it isn't being used.... A separate kernel thread (bitmap_writeback_daemon) is needed to wait for each page as we cannot get callbacks when a page write completes. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/bitmap.c | 119 ++++++++++++++++++++------------------------ include/linux/raid/bitmap.h | 13 +---- 2 files changed, 55 insertions(+), 77 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c index 9462fdd517c..86b6b037fa4 100644 --- a/drivers/md/bitmap.c +++ b/drivers/md/bitmap.c @@ -261,30 +261,33 @@ char *file_path(struct file *file, char *buf, int count) /* * write out a page */ -static int write_page(struct page *page, int wait) +static int write_page(struct bitmap *bitmap, struct page *page, int wait) { int ret = -ENOMEM; lock_page(page); - if (page->mapping == NULL) - goto unlock_out; - else if (i_size_read(page->mapping->host) < page->index << PAGE_SHIFT) { - ret = -ENOENT; - goto unlock_out; - } - ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE); if (!ret) ret = page->mapping->a_ops->commit_write(NULL, page, 0, PAGE_SIZE); if (ret) { -unlock_out: unlock_page(page); return ret; } set_page_dirty(page); /* force it to be written out */ + + if (!wait) { + /* add to list to be waited for by daemon */ + struct page_list *item = mempool_alloc(bitmap->write_pool, GFP_NOIO); + item->page = page; + page_cache_get(page); + spin_lock(&bitmap->write_lock); + list_add(&item->list, &bitmap->complete_pages); + spin_unlock(&bitmap->write_lock); + md_wakeup_thread(bitmap->writeback_daemon); + } return write_one_page(page, wait); } @@ -343,14 +346,13 @@ int bitmap_update_sb(struct bitmap *bitmap) spin_unlock_irqrestore(&bitmap->lock, flags); return 0; } - page_cache_get(bitmap->sb_page); spin_unlock_irqrestore(&bitmap->lock, flags); sb = (bitmap_super_t *)kmap(bitmap->sb_page); sb->events = cpu_to_le64(bitmap->mddev->events); if (!bitmap->mddev->degraded) sb->events_cleared = cpu_to_le64(bitmap->mddev->events); kunmap(bitmap->sb_page); - return write_page(bitmap->sb_page, 0); + return write_page(bitmap, bitmap->sb_page, 0); } /* print out the bitmap file superblock */ @@ -556,10 +558,10 @@ static void bitmap_file_unmap(struct bitmap *bitmap) static void bitmap_stop_daemons(struct bitmap *bitmap); /* dequeue the next item in a page list -- don't call from irq context */ -static struct page_list *dequeue_page(struct bitmap *bitmap, - struct list_head *head) +static struct page_list *dequeue_page(struct bitmap *bitmap) { struct page_list *item = NULL; + struct list_head *head = &bitmap->complete_pages; spin_lock(&bitmap->write_lock); if (list_empty(head)) @@ -573,23 +575,15 @@ out: static void drain_write_queues(struct bitmap *bitmap) { - struct list_head *queues[] = { &bitmap->complete_pages, NULL }; - struct list_head *head; struct page_list *item; - int i; - for (i = 0; queues[i]; i++) { - head = queues[i]; - while ((item = dequeue_page(bitmap, head))) { - page_cache_release(item->page); - mempool_free(item, bitmap->write_pool); - } + while ((item = dequeue_page(bitmap))) { + /* don't bother to wait */ + page_cache_release(item->page); + mempool_free(item, bitmap->write_pool); } - spin_lock(&bitmap->write_lock); - bitmap->writes_pending = 0; /* make sure waiters continue */ wake_up(&bitmap->write_wait); - spin_unlock(&bitmap->write_lock); } static void bitmap_file_put(struct bitmap *bitmap) @@ -734,13 +728,13 @@ int bitmap_unplug(struct bitmap *bitmap) spin_unlock_irqrestore(&bitmap->lock, flags); if (attr & (BITMAP_PAGE_DIRTY | BITMAP_PAGE_NEEDWRITE)) - if (write_page(page, 0)) + if (write_page(bitmap, page, 0)) return 1; } if (wait) { /* if any writes were performed, we need to wait on them */ spin_lock_irq(&bitmap->write_lock); wait_event_lock_irq(bitmap->write_wait, - bitmap->writes_pending == 0, bitmap->write_lock, + list_empty(&bitmap->complete_pages), bitmap->write_lock, wake_up_process(bitmap->writeback_daemon->tsk)); spin_unlock_irq(&bitmap->write_lock); } @@ -841,7 +835,7 @@ static int bitmap_init_from_disk(struct bitmap *bitmap, int in_sync) */ memset(page_address(page) + offset, 0xff, PAGE_SIZE - offset); - ret = write_page(page, 1); + ret = write_page(bitmap, page, 1); if (ret) { kunmap(page); /* release, page not in filemap yet */ @@ -934,7 +928,7 @@ int bitmap_daemon_work(struct bitmap *bitmap) } spin_unlock_irqrestore(&bitmap->lock, flags); if (attr & BITMAP_PAGE_NEEDWRITE) { - if (write_page(page, 0)) + if (write_page(bitmap, page, 0)) bitmap_file_kick(bitmap); page_cache_release(page); } @@ -950,7 +944,7 @@ int bitmap_daemon_work(struct bitmap *bitmap) if (get_page_attr(bitmap, lastpage) & BITMAP_PAGE_NEEDWRITE) { clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); spin_unlock_irqrestore(&bitmap->lock, flags); - err = write_page(lastpage, 0); + err = write_page(bitmap, lastpage, 0); } else { set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); spin_unlock_irqrestore(&bitmap->lock, flags); @@ -998,7 +992,7 @@ int bitmap_daemon_work(struct bitmap *bitmap) if (get_page_attr(bitmap, lastpage) &BITMAP_PAGE_NEEDWRITE) { clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); spin_unlock_irqrestore(&bitmap->lock, flags); - err = write_page(lastpage, 0); + err = write_page(bitmap, lastpage, 0); } else { set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); spin_unlock_irqrestore(&bitmap->lock, flags); @@ -1034,46 +1028,40 @@ static void bitmap_writeback_daemon(mddev_t *mddev) struct page_list *item; int err = 0; - while (1) { - PRINTK("%s: bitmap writeback daemon waiting...\n", bmname(bitmap)); - down_interruptible(&bitmap->write_done); - if (signal_pending(current)) { - printk(KERN_INFO - "%s: bitmap writeback daemon got signal, exiting...\n", - bmname(bitmap)); - break; - } + if (signal_pending(current)) { + printk(KERN_INFO + "%s: bitmap writeback daemon got signal, exiting...\n", + bmname(bitmap)); + err = -EINTR; + goto out; + } - PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap)); - /* wait on bitmap page writebacks */ - while ((item = dequeue_page(bitmap, &bitmap->complete_pages))) { - page = item->page; - mempool_free(item, bitmap->write_pool); - PRINTK("wait on page writeback: %p %lu\n", page, bitmap->writes_pending); - wait_on_page_writeback(page); - PRINTK("finished page writeback: %p %lu\n", page, bitmap->writes_pending); - spin_lock(&bitmap->write_lock); - if (!--bitmap->writes_pending) - wake_up(&bitmap->write_wait); - spin_unlock(&bitmap->write_lock); - err = PageError(page); - page_cache_release(page); - if (err) { - printk(KERN_WARNING "%s: bitmap file writeback " - "failed (page %lu): %d\n", - bmname(bitmap), page->index, err); - bitmap_file_kick(bitmap); - goto out; - } + PRINTK("%s: bitmap writeback daemon woke up...\n", bmname(bitmap)); + /* wait on bitmap page writebacks */ + while ((item = dequeue_page(bitmap))) { + page = item->page; + mempool_free(item, bitmap->write_pool); + PRINTK("wait on page writeback: %p\n", page); + wait_on_page_writeback(page); + PRINTK("finished page writeback: %p\n", page); + + err = PageError(page); + page_cache_release(page); + if (err) { + printk(KERN_WARNING "%s: bitmap file writeback " + "failed (page %lu): %d\n", + bmname(bitmap), page->index, err); + bitmap_file_kick(bitmap); + goto out; } } -out: + out: + wake_up(&bitmap->write_wait); if (err) { printk(KERN_INFO "%s: bitmap writeback daemon exiting (%d)\n", - bmname(bitmap), err); + bmname(bitmap), err); daemon_exit(bitmap, &bitmap->writeback_daemon); } - return; } static int bitmap_start_daemon(struct bitmap *bitmap, mdk_thread_t **ptr, @@ -1375,7 +1363,7 @@ int bitmap_setallbits(struct bitmap *bitmap) spin_unlock_irqrestore(&bitmap->lock, flags); memset(kmap(page), 0xff, PAGE_SIZE); kunmap(page); - if (write_page(page, 0)) + if (write_page(bitmap, page, 0)) return 1; } @@ -1443,7 +1431,6 @@ int bitmap_create(mddev_t *mddev) mddev->bitmap = bitmap; spin_lock_init(&bitmap->write_lock); - init_MUTEX_LOCKED(&bitmap->write_done); INIT_LIST_HEAD(&bitmap->complete_pages); init_waitqueue_head(&bitmap->write_wait); bitmap->write_pool = mempool_create(WRITE_POOL_SIZE, write_pool_alloc, diff --git a/include/linux/raid/bitmap.h b/include/linux/raid/bitmap.h index f785cf26cba..cfe60cfc8f3 100644 --- a/include/linux/raid/bitmap.h +++ b/include/linux/raid/bitmap.h @@ -233,21 +233,12 @@ struct bitmap { unsigned long daemon_sleep; /* how many seconds between updates? */ /* - * bitmap write daemon - this daemon performs writes to the bitmap file - * this thread is only needed because of a limitation in ext3 (jbd) - * that does not allow a task to have two journal transactions ongoing - * simultaneously (even if the transactions are for two different - * filesystems) -- in the case of bitmap, that would be the filesystem - * that the bitmap file resides on and the filesystem that is mounted - * on the md device -- see current->journal_info in jbd/transaction.c + * bitmap_writeback_daemon waits for file-pages that have been written, + * as there is no way to get a call-back when a page write completes. */ mdk_thread_t *writeback_daemon; spinlock_t write_lock; - struct semaphore write_ready; - struct semaphore write_done; - unsigned long writes_pending; wait_queue_head_t write_wait; - struct list_head write_pages; struct list_head complete_pages; mempool_t *write_pool; }; -- cgit v1.2.3-70-g09d2 From 191ea9b2c7cc3ebbe0678834ab710d7d95ad3f9a Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:23 -0700 Subject: [PATCH] md: raid1 support for bitmap intent logging Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/raid1.c | 182 ++++++++++++++++++++++++++++++++++++++------- include/linux/raid/raid1.h | 16 +++- 2 files changed, 168 insertions(+), 30 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index 3c5c916cb09..0fd4c3bfc85 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -12,6 +12,15 @@ * Fixes to reconstruction by Jakob Østergaard" * Various fixes by Neil Brown * + * Changes by Peter T. Breuer 31/1/2003 to support + * bitmapped intelligence in resync: + * + * - bitmap marked during normal i/o + * - bitmap used to skip nondirty blocks during sync + * + * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: + * - persistent bitmap code + * * 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, or (at your option) @@ -22,7 +31,16 @@ * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ +#include "dm-bio-list.h" #include +#include + +#define DEBUG 0 +#if DEBUG +#define PRINTK(x...) printk(x) +#else +#define PRINTK(x...) +#endif /* * Number of guaranteed r1bios in case of extreme VM load: @@ -287,9 +305,11 @@ static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int /* * this branch is our 'one mirror IO has finished' event handler: */ - if (!uptodate) + if (!uptodate) { md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); - else + /* an I/O failed, we can't clear the bitmap */ + set_bit(R1BIO_Degraded, &r1_bio->state); + } else /* * Set R1BIO_Uptodate in our master bio, so that * we will return a good error code for to the higher @@ -309,6 +329,10 @@ static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int * already. */ if (atomic_dec_and_test(&r1_bio->remaining)) { + /* clear the bitmap if all writes complete successfully */ + bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, + r1_bio->sectors, + !test_bit(R1BIO_Degraded, &r1_bio->state)); md_write_end(r1_bio->mddev); raid_end_bio_io(r1_bio); } @@ -458,7 +482,10 @@ static void unplug_slaves(mddev_t *mddev) static void raid1_unplug(request_queue_t *q) { - unplug_slaves(q->queuedata); + mddev_t *mddev = q->queuedata; + + unplug_slaves(mddev); + md_wakeup_thread(mddev->thread); } static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk, @@ -501,16 +528,16 @@ static void device_barrier(conf_t *conf, sector_t sect) { spin_lock_irq(&conf->resync_lock); wait_event_lock_irq(conf->wait_idle, !waitqueue_active(&conf->wait_resume), - conf->resync_lock, unplug_slaves(conf->mddev)); + conf->resync_lock, raid1_unplug(conf->mddev->queue)); if (!conf->barrier++) { wait_event_lock_irq(conf->wait_idle, !conf->nr_pending, - conf->resync_lock, unplug_slaves(conf->mddev)); + conf->resync_lock, raid1_unplug(conf->mddev->queue)); if (conf->nr_pending) BUG(); } wait_event_lock_irq(conf->wait_resume, conf->barrier < RESYNC_DEPTH, - conf->resync_lock, unplug_slaves(conf->mddev)); + conf->resync_lock, raid1_unplug(conf->mddev->queue)); conf->next_resync = sect; spin_unlock_irq(&conf->resync_lock); } @@ -522,8 +549,12 @@ static int make_request(request_queue_t *q, struct bio * bio) mirror_info_t *mirror; r1bio_t *r1_bio; struct bio *read_bio; - int i, disks; + int i, targets = 0, disks; mdk_rdev_t *rdev; + struct bitmap *bitmap = mddev->bitmap; + unsigned long flags; + struct bio_list bl; + /* * Register the new request and wait if the reconstruction @@ -554,7 +585,7 @@ static int make_request(request_queue_t *q, struct bio * bio) r1_bio->master_bio = bio; r1_bio->sectors = bio->bi_size >> 9; - + r1_bio->state = 0; r1_bio->mddev = mddev; r1_bio->sector = bio->bi_sector; @@ -597,6 +628,13 @@ static int make_request(request_queue_t *q, struct bio * bio) * bios[x] to bio */ disks = conf->raid_disks; +#if 0 + { static int first=1; + if (first) printk("First Write sector %llu disks %d\n", + (unsigned long long)r1_bio->sector, disks); + first = 0; + } +#endif rcu_read_lock(); for (i = 0; i < disks; i++) { if ((rdev=conf->mirrors[i].rdev) != NULL && @@ -607,13 +645,21 @@ static int make_request(request_queue_t *q, struct bio * bio) r1_bio->bios[i] = NULL; } else r1_bio->bios[i] = bio; + targets++; } else r1_bio->bios[i] = NULL; } rcu_read_unlock(); - atomic_set(&r1_bio->remaining, 1); + if (targets < conf->raid_disks) { + /* array is degraded, we will not clear the bitmap + * on I/O completion (see raid1_end_write_request) */ + set_bit(R1BIO_Degraded, &r1_bio->state); + } + + atomic_set(&r1_bio->remaining, 0); + bio_list_init(&bl); for (i = 0; i < disks; i++) { struct bio *mbio; if (!r1_bio->bios[i]) @@ -629,14 +675,23 @@ static int make_request(request_queue_t *q, struct bio * bio) mbio->bi_private = r1_bio; atomic_inc(&r1_bio->remaining); - generic_make_request(mbio); - } - if (atomic_dec_and_test(&r1_bio->remaining)) { - md_write_end(mddev); - raid_end_bio_io(r1_bio); + bio_list_add(&bl, mbio); } + bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors); + spin_lock_irqsave(&conf->device_lock, flags); + bio_list_merge(&conf->pending_bio_list, &bl); + bio_list_init(&bl); + + blk_plug_device(mddev->queue); + spin_unlock_irqrestore(&conf->device_lock, flags); + +#if 0 + while ((bio = bio_list_pop(&bl)) != NULL) + generic_make_request(bio); +#endif + return 0; } @@ -716,7 +771,7 @@ static void close_sync(conf_t *conf) { spin_lock_irq(&conf->resync_lock); wait_event_lock_irq(conf->wait_resume, !conf->barrier, - conf->resync_lock, unplug_slaves(conf->mddev)); + conf->resync_lock, raid1_unplug(conf->mddev->queue)); spin_unlock_irq(&conf->resync_lock); if (conf->barrier) BUG(); @@ -830,10 +885,11 @@ static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error) * or re-read if the read failed. * We don't do much here, just schedule handling by raid1d */ - if (!uptodate) + if (!uptodate) { md_error(r1_bio->mddev, conf->mirrors[r1_bio->read_disk].rdev); - else + set_bit(R1BIO_Degraded, &r1_bio->state); + } else set_bit(R1BIO_Uptodate, &r1_bio->state); rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); reschedule_retry(r1_bio); @@ -857,8 +913,10 @@ static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error) mirror = i; break; } - if (!uptodate) + if (!uptodate) { md_error(mddev, conf->mirrors[mirror].rdev); + set_bit(R1BIO_Degraded, &r1_bio->state); + } update_head_pos(mirror, r1_bio); if (atomic_dec_and_test(&r1_bio->remaining)) { @@ -878,6 +936,9 @@ static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) bio = r1_bio->bios[r1_bio->read_disk]; +/* + if (r1_bio->sector == 0) printk("First sync write startss\n"); +*/ /* * schedule writes */ @@ -905,10 +966,12 @@ static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) atomic_inc(&conf->mirrors[i].rdev->nr_pending); atomic_inc(&r1_bio->remaining); md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); + generic_make_request(wbio); } if (atomic_dec_and_test(&r1_bio->remaining)) { + /* if we're here, all write(s) have completed, so clean up */ md_done_sync(mddev, r1_bio->sectors, 1); put_buf(r1_bio); } @@ -937,6 +1000,26 @@ static void raid1d(mddev_t *mddev) for (;;) { char b[BDEVNAME_SIZE]; spin_lock_irqsave(&conf->device_lock, flags); + + if (conf->pending_bio_list.head) { + bio = bio_list_get(&conf->pending_bio_list); + blk_remove_plug(mddev->queue); + spin_unlock_irqrestore(&conf->device_lock, flags); + /* flush any pending bitmap writes to disk before proceeding w/ I/O */ + if (bitmap_unplug(mddev->bitmap) != 0) + printk("%s: bitmap file write failed!\n", mdname(mddev)); + + while (bio) { /* submit pending writes */ + struct bio *next = bio->bi_next; + bio->bi_next = NULL; + generic_make_request(bio); + bio = next; + } + unplug = 1; + + continue; + } + if (list_empty(head)) break; r1_bio = list_entry(head->prev, r1bio_t, retry_list); @@ -1020,17 +1103,43 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i int disk; int i; int write_targets = 0; + int sync_blocks; if (!conf->r1buf_pool) + { +/* + printk("sync start - bitmap %p\n", mddev->bitmap); +*/ if (init_resync(conf)) return 0; + } max_sector = mddev->size << 1; if (sector_nr >= max_sector) { + /* If we aborted, we need to abort the + * sync on the 'current' bitmap chunk (there will + * only be one in raid1 resync. + * We can find the current addess in mddev->curr_resync + */ + if (!conf->fullsync) { + if (mddev->curr_resync < max_sector) + bitmap_end_sync(mddev->bitmap, + mddev->curr_resync, + &sync_blocks, 1); + bitmap_close_sync(mddev->bitmap); + } + if (mddev->curr_resync >= max_sector) + conf->fullsync = 0; close_sync(conf); return 0; } + if (!conf->fullsync && + !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks)) { + /* We can skip this block, and probably several more */ + *skipped = 1; + return sync_blocks; + } /* * If there is non-resync activity waiting for us then * put in a delay to throttle resync. @@ -1069,6 +1178,7 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i r1_bio->mddev = mddev; r1_bio->sector = sector_nr; + r1_bio->state = 0; set_bit(R1BIO_IsSync, &r1_bio->state); r1_bio->read_disk = disk; @@ -1103,6 +1213,11 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i bio->bi_bdev = conf->mirrors[i].rdev->bdev; bio->bi_private = r1_bio; } + + if (write_targets + 1 < conf->raid_disks) + /* array degraded, can't clear bitmap */ + set_bit(R1BIO_Degraded, &r1_bio->state); + if (write_targets == 0) { /* There is nowhere to write, so all non-sync * drives must be failed - so we are finished @@ -1122,6 +1237,14 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i len = (max_sector - sector_nr) << 9; if (len == 0) break; + if (!conf->fullsync && sync_blocks == 0) + if (!bitmap_start_sync(mddev->bitmap, + sector_nr, &sync_blocks)) + break; + if (sync_blocks < (PAGE_SIZE>>9)) + BUG(); + if (len > (sync_blocks<<9)) len = sync_blocks<<9; + for (i=0 ; i < conf->raid_disks; i++) { bio = r1_bio->bios[i]; if (bio->bi_end_io) { @@ -1144,6 +1267,7 @@ static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, i } nr_sectors += len>>9; sector_nr += len>>9; + sync_blocks -= (len>>9); } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); bio_full: bio = r1_bio->bios[disk]; @@ -1232,6 +1356,9 @@ static int run(mddev_t *mddev) init_waitqueue_head(&conf->wait_idle); init_waitqueue_head(&conf->wait_resume); + bio_list_init(&conf->pending_bio_list); + bio_list_init(&conf->flushing_bio_list); + if (!conf->working_disks) { printk(KERN_ERR "raid1: no operational mirrors for %s\n", mdname(mddev)); @@ -1260,16 +1387,15 @@ static int run(mddev_t *mddev) conf->last_used = j; - - { - mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); - if (!mddev->thread) { - printk(KERN_ERR - "raid1: couldn't allocate thread for %s\n", - mdname(mddev)); - goto out_free_conf; - } + mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); + if (!mddev->thread) { + printk(KERN_ERR + "raid1: couldn't allocate thread for %s\n", + mdname(mddev)); + goto out_free_conf; } + if (mddev->bitmap) mddev->thread->timeout = mddev->bitmap->daemon_sleep * HZ; + printk(KERN_INFO "raid1: raid set %s active with %d out of %d mirrors\n", mdname(mddev), mddev->raid_disks - mddev->degraded, @@ -1394,7 +1520,7 @@ static int raid1_reshape(mddev_t *mddev, int raid_disks) spin_lock_irq(&conf->resync_lock); conf->barrier++; wait_event_lock_irq(conf->wait_idle, !conf->nr_pending, - conf->resync_lock, unplug_slaves(mddev)); + conf->resync_lock, raid1_unplug(mddev->queue)); spin_unlock_irq(&conf->resync_lock); /* ok, everything is stopped */ diff --git a/include/linux/raid/raid1.h b/include/linux/raid/raid1.h index abbfdd9afe1..9d93cf12e89 100644 --- a/include/linux/raid/raid1.h +++ b/include/linux/raid/raid1.h @@ -36,12 +36,21 @@ struct r1_private_data_s { spinlock_t device_lock; struct list_head retry_list; + /* queue pending writes and submit them on unplug */ + struct bio_list pending_bio_list; + /* queue of writes that have been unplugged */ + struct bio_list flushing_bio_list; + /* for use when syncing mirrors: */ spinlock_t resync_lock; - int nr_pending; - int barrier; + int nr_pending; + int barrier; sector_t next_resync; + int fullsync; /* set to 1 if a full sync is needed, + * (fresh device added). + * Cleared when a sync completes. + */ wait_queue_head_t wait_idle; wait_queue_head_t wait_resume; @@ -85,14 +94,17 @@ struct r1bio_s { int read_disk; struct list_head retry_list; + struct bitmap_update *bitmap_update; /* * if the IO is in WRITE direction, then multiple bios are used. * We choose the number when they are allocated. */ struct bio *bios[0]; + /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/ }; /* bits for r1bio.state */ #define R1BIO_Uptodate 0 #define R1BIO_IsSync 1 +#define R1BIO_Degraded 2 #endif -- cgit v1.2.3-70-g09d2 From 41158c7eb22312cfaa256744e1553bb4042ff085 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:25 -0700 Subject: [PATCH] md: optimise reconstruction when re-adding a recently failed drive. When an array is degraded, bit in the intent-bitmap are never cleared. So if a recently failed drive is re-added, we only need to reconstruct the block that are still reflected in the bitmap. This patch adds support for this re-adding. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/md.c | 71 ++++++++++++++++++++++++++++++++++++----------- drivers/md/raid1.c | 7 ++++- include/linux/raid/md_k.h | 4 +++ 3 files changed, 65 insertions(+), 17 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/md.c b/drivers/md/md.c index b02f8d1d77e..789b114f860 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -577,6 +577,8 @@ static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) mdp_disk_t *desc; mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); + rdev->raid_disk = -1; + rdev->in_sync = 0; if (mddev->raid_disks == 0) { mddev->major_version = 0; mddev->minor_version = sb->minor_version; @@ -607,16 +609,24 @@ static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) memcpy(mddev->uuid+12,&sb->set_uuid3, 4); mddev->max_disks = MD_SB_DISKS; - } else { - __u64 ev1; - ev1 = md_event(sb); + } else if (mddev->pers == NULL) { + /* Insist on good event counter while assembling */ + __u64 ev1 = md_event(sb); ++ev1; if (ev1 < mddev->events) return -EINVAL; - } + } else if (mddev->bitmap) { + /* if adding to array with a bitmap, then we can accept an + * older device ... but not too old. + */ + __u64 ev1 = md_event(sb); + if (ev1 < mddev->bitmap->events_cleared) + return 0; + } else /* just a hot-add of a new device, leave raid_disk at -1 */ + return 0; + if (mddev->level != LEVEL_MULTIPATH) { - rdev->raid_disk = -1; - rdev->in_sync = rdev->faulty = 0; + rdev->faulty = 0; desc = sb->disks + rdev->desc_nr; if (desc->state & (1<in_sync = 1; rdev->raid_disk = desc->raid_disk; } - } + } else /* MULTIPATH are always insync */ + rdev->in_sync = 1; return 0; } @@ -868,6 +879,8 @@ static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) { struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); + rdev->raid_disk = -1; + rdev->in_sync = 0; if (mddev->raid_disks == 0) { mddev->major_version = 1; mddev->patch_version = 0; @@ -885,13 +898,21 @@ static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) memcpy(mddev->uuid, sb->set_uuid, 16); mddev->max_disks = (4096-256)/2; - } else { - __u64 ev1; - ev1 = le64_to_cpu(sb->events); + } else if (mddev->pers == NULL) { + /* Insist of good event counter while assembling */ + __u64 ev1 = le64_to_cpu(sb->events); ++ev1; if (ev1 < mddev->events) return -EINVAL; - } + } else if (mddev->bitmap) { + /* If adding to array with a bitmap, then we can accept an + * older device, but not too old. + */ + __u64 ev1 = le64_to_cpu(sb->events); + if (ev1 < mddev->bitmap->events_cleared) + return 0; + } else /* just a hot-add of a new device, leave raid_disk at -1 */ + return 0; if (mddev->level != LEVEL_MULTIPATH) { int role; @@ -899,14 +920,10 @@ static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); switch(role) { case 0xffff: /* spare */ - rdev->in_sync = 0; rdev->faulty = 0; - rdev->raid_disk = -1; break; case 0xfffe: /* faulty */ - rdev->in_sync = 0; rdev->faulty = 1; - rdev->raid_disk = -1; break; default: rdev->in_sync = 1; @@ -914,7 +931,9 @@ static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) rdev->raid_disk = role; break; } - } + } else /* MULTIPATH are always insync */ + rdev->in_sync = 1; + return 0; } @@ -2155,6 +2174,18 @@ static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) PTR_ERR(rdev)); return PTR_ERR(rdev); } + /* set save_raid_disk if appropriate */ + if (!mddev->persistent) { + if (info->state & (1<raid_disk < mddev->raid_disks) + rdev->raid_disk = info->raid_disk; + else + rdev->raid_disk = -1; + } else + super_types[mddev->major_version]. + validate_super(mddev, rdev); + rdev->saved_raid_disk = rdev->raid_disk; + rdev->in_sync = 0; /* just to be sure */ rdev->raid_disk = -1; err = bind_rdev_to_array(rdev, mddev); @@ -3706,6 +3737,14 @@ void md_check_recovery(mddev_t *mddev) mddev->pers->spare_active(mddev); } md_update_sb(mddev); + + /* if array is no-longer degraded, then any saved_raid_disk + * information must be scrapped + */ + if (!mddev->degraded) + ITERATE_RDEV(mddev,rdev,rtmp) + rdev->saved_raid_disk = -1; + mddev->recovery = 0; /* flag recovery needed just to double check */ set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index c3b4772cfae..3f5234fe359 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -811,9 +811,12 @@ static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) { conf_t *conf = mddev->private; int found = 0; - int mirror; + int mirror = 0; mirror_info_t *p; + if (rdev->saved_raid_disk >= 0 && + conf->mirrors[rdev->saved_raid_disk].rdev == NULL) + mirror = rdev->saved_raid_disk; for (mirror=0; mirror < mddev->raid_disks; mirror++) if ( !(p=conf->mirrors+mirror)->rdev) { @@ -830,6 +833,8 @@ static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) p->head_position = 0; rdev->raid_disk = mirror; found = 1; + if (rdev->saved_raid_disk != mirror) + conf->fullsync = 1; p->rdev = rdev; break; } diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index 16e94a9f0f8..6cdcb4434c6 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -183,6 +183,10 @@ struct mdk_rdev_s int desc_nr; /* descriptor index in the superblock */ int raid_disk; /* role of device in array */ + int saved_raid_disk; /* role that device used to have in the + * array and could again if we did a partial + * resync from the bitmap + */ atomic_t nr_pending; /* number of pending requests. * only maintained for arrays that -- cgit v1.2.3-70-g09d2 From 3d310eb7b3df1252e8595d059d982b0a9825a137 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:26 -0700 Subject: [PATCH] md: fix deadlock due to md thread processing delayed requests. Before completing a 'write' the md superblock might need to be updated. This is best done by the md_thread. The current code schedules this up and queues the write request for later handling by the md_thread. However some personalities (Raid5/raid6) will deadlock if the md_thread tries to submit requests to its own array. So this patch changes things so the processes submitting the request waits for the superblock to be written and then submits the request itself. This fixes a recently-created deadlock in raid5/raid6 Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/md.c | 45 ++++++++++++++++----------------------------- drivers/md/raid1.c | 4 ++-- drivers/md/raid10.c | 3 +-- drivers/md/raid5.c | 3 +-- drivers/md/raid6main.c | 3 +-- include/linux/raid/md.h | 2 +- include/linux/raid/md_k.h | 2 +- 7 files changed, 23 insertions(+), 39 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/md.c b/drivers/md/md.c index 789b114f860..7075bebb7f3 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -224,8 +224,8 @@ static mddev_t * mddev_find(dev_t unit) INIT_LIST_HEAD(&new->all_mddevs); init_timer(&new->safemode_timer); atomic_set(&new->active, 1); - bio_list_init(&new->write_list); spin_lock_init(&new->write_lock); + init_waitqueue_head(&new->sb_wait); new->queue = blk_alloc_queue(GFP_KERNEL); if (!new->queue) { @@ -1307,6 +1307,7 @@ repeat: if (!mddev->persistent) { mddev->sb_dirty = 0; spin_unlock(&mddev->write_lock); + wake_up(&mddev->sb_wait); return; } spin_unlock(&mddev->write_lock); @@ -1348,6 +1349,7 @@ repeat: } mddev->sb_dirty = 0; spin_unlock(&mddev->write_lock); + wake_up(&mddev->sb_wait); } @@ -3368,29 +3370,26 @@ void md_done_sync(mddev_t *mddev, int blocks, int ok) /* md_write_start(mddev, bi) * If we need to update some array metadata (e.g. 'active' flag - * in superblock) before writing, queue bi for later writing - * and return 0, else return 1 and it will be written now + * in superblock) before writing, schedule a superblock update + * and wait for it to complete. */ -int md_write_start(mddev_t *mddev, struct bio *bi) +void md_write_start(mddev_t *mddev, struct bio *bi) { + DEFINE_WAIT(w); if (bio_data_dir(bi) != WRITE) - return 1; + return; atomic_inc(&mddev->writes_pending); - spin_lock(&mddev->write_lock); - if (mddev->in_sync == 0 && mddev->sb_dirty == 0) { - spin_unlock(&mddev->write_lock); - return 1; - } - bio_list_add(&mddev->write_list, bi); - if (mddev->in_sync) { - mddev->in_sync = 0; - mddev->sb_dirty = 1; + spin_lock(&mddev->write_lock); + if (mddev->in_sync) { + mddev->in_sync = 0; + mddev->sb_dirty = 1; + md_wakeup_thread(mddev->thread); + } + spin_unlock(&mddev->write_lock); } - spin_unlock(&mddev->write_lock); - md_wakeup_thread(mddev->thread); - return 0; + wait_event(mddev->sb_wait, mddev->sb_dirty==0); } void md_write_end(mddev_t *mddev) @@ -3685,7 +3684,6 @@ void md_check_recovery(mddev_t *mddev) mddev->sb_dirty || test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || test_bit(MD_RECOVERY_DONE, &mddev->recovery) || - mddev->write_list.head || (mddev->safemode == 1) || (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending) && !mddev->in_sync && mddev->recovery_cp == MaxSector) @@ -3694,7 +3692,6 @@ void md_check_recovery(mddev_t *mddev) if (mddev_trylock(mddev)==0) { int spares =0; - struct bio *blist; spin_lock(&mddev->write_lock); if (mddev->safemode && !atomic_read(&mddev->writes_pending) && @@ -3704,21 +3701,11 @@ void md_check_recovery(mddev_t *mddev) } if (mddev->safemode == 1) mddev->safemode = 0; - blist = bio_list_get(&mddev->write_list); spin_unlock(&mddev->write_lock); if (mddev->sb_dirty) md_update_sb(mddev); - while (blist) { - struct bio *b = blist; - blist = blist->bi_next; - b->bi_next = NULL; - generic_make_request(b); - /* we already counted this, so need to un-count */ - md_write_end(mddev); - } - if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index 3f5234fe359..98b09773e79 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -561,8 +561,8 @@ static int make_request(request_queue_t *q, struct bio * bio) * thread has put up a bar for new requests. * Continue immediately if no resync is active currently. */ - if (md_write_start(mddev, bio)==0) - return 0; + md_write_start(mddev, bio); /* wait on superblock update early */ + spin_lock_irq(&conf->resync_lock); wait_event_lock_irq(conf->wait_resume, !conf->barrier, conf->resync_lock, ); conf->nr_pending++; diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index 8476515bfdc..fd7324a86d1 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -700,8 +700,7 @@ static int make_request(request_queue_t *q, struct bio * bio) return 0; } - if (md_write_start(mddev, bio) == 0) - return 0; + md_write_start(mddev, bio); /* * Register the new request and wait if the reconstruction diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 1ce3f5aaa98..93a9726cc2d 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -1411,8 +1411,7 @@ static int make_request (request_queue_t *q, struct bio * bi) sector_t logical_sector, last_sector; struct stripe_head *sh; - if (md_write_start(mddev, bi)==0) - return 0; + md_write_start(mddev, bi); if (bio_data_dir(bi)==WRITE) { disk_stat_inc(mddev->gendisk, writes); diff --git a/drivers/md/raid6main.c b/drivers/md/raid6main.c index d9c385496dc..f62ea1a73d0 100644 --- a/drivers/md/raid6main.c +++ b/drivers/md/raid6main.c @@ -1570,8 +1570,7 @@ static int make_request (request_queue_t *q, struct bio * bi) sector_t logical_sector, last_sector; struct stripe_head *sh; - if (md_write_start(mddev, bi)==0) - return 0; + md_write_start(mddev, bi); if (bio_data_dir(bi)==WRITE) { disk_stat_inc(mddev->gendisk, writes); diff --git a/include/linux/raid/md.h b/include/linux/raid/md.h index cfde8f497d6..75f41d8faed 100644 --- a/include/linux/raid/md.h +++ b/include/linux/raid/md.h @@ -69,7 +69,7 @@ extern mdk_thread_t * md_register_thread (void (*run) (mddev_t *mddev), extern void md_unregister_thread (mdk_thread_t *thread); extern void md_wakeup_thread(mdk_thread_t *thread); extern void md_check_recovery(mddev_t *mddev); -extern int md_write_start(mddev_t *mddev, struct bio *bi); +extern void md_write_start(mddev_t *mddev, struct bio *bi); extern void md_write_end(mddev_t *mddev); extern void md_handle_safemode(mddev_t *mddev); extern void md_done_sync(mddev_t *mddev, int blocks, int ok); diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index 6cdcb4434c6..3e977025cf4 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -261,7 +261,7 @@ struct mddev_s sector_t recovery_cp; spinlock_t write_lock; - struct bio_list write_list; + wait_queue_head_t sb_wait; /* for waiting on superblock updates */ unsigned int safemode; /* if set, update "clean" superblock * when no writes pending. -- cgit v1.2.3-70-g09d2 From a654b9d8f851f4ca02649d5825cbe6c608adb10c Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:27 -0700 Subject: [PATCH] md: allow md intent bitmap to be stored near the superblock. This provides an alternate to storing the bitmap in a separate file. The bitmap can be stored at a given offset from the superblock. Obviously the creator of the array must make sure this doesn't intersect with data.... After is good for version-0.90 superblocks. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/bitmap.c | 132 +++++++++++++++++++++++++++++++++++--------- drivers/md/md.c | 40 +++++++++++++- include/linux/raid/bitmap.h | 2 + include/linux/raid/md.h | 15 ++++- include/linux/raid/md_k.h | 4 ++ include/linux/raid/md_p.h | 7 ++- 6 files changed, 170 insertions(+), 30 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c index 204564dc6a0..030d6861051 100644 --- a/drivers/md/bitmap.c +++ b/drivers/md/bitmap.c @@ -116,7 +116,7 @@ static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) if (!page) printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); else - printk("%s: bitmap_alloc_page: allocated page at %p\n", + PRINTK("%s: bitmap_alloc_page: allocated page at %p\n", bmname(bitmap), page); return page; } @@ -258,13 +258,61 @@ char *file_path(struct file *file, char *buf, int count) * basic page I/O operations */ +/* IO operations when bitmap is stored near all superblocks */ +static struct page *read_sb_page(mddev_t *mddev, long offset, unsigned long index) +{ + /* choose a good rdev and read the page from there */ + + mdk_rdev_t *rdev; + struct list_head *tmp; + struct page *page = alloc_page(GFP_KERNEL); + sector_t target; + + if (!page) + return ERR_PTR(-ENOMEM); + do { + ITERATE_RDEV(mddev, rdev, tmp) + if (rdev->in_sync && !rdev->faulty) + goto found; + return ERR_PTR(-EIO); + + found: + target = (rdev->sb_offset << 1) + offset + index * (PAGE_SIZE/512); + + } while (!sync_page_io(rdev->bdev, target, PAGE_SIZE, page, READ)); + + page->index = index; + return page; +} + +static int write_sb_page(mddev_t *mddev, long offset, struct page *page, int wait) +{ + mdk_rdev_t *rdev; + struct list_head *tmp; + + ITERATE_RDEV(mddev, rdev, tmp) + if (rdev->in_sync && !rdev->faulty) + md_super_write(mddev, rdev, + (rdev->sb_offset<<1) + offset + + page->index * (PAGE_SIZE/512), + PAGE_SIZE, + page); + + if (wait) + wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); + return 0; +} + /* - * write out a page + * write out a page to a file */ static int write_page(struct bitmap *bitmap, struct page *page, int wait) { int ret = -ENOMEM; + if (bitmap->file == NULL) + return write_sb_page(bitmap->mddev, bitmap->offset, page, wait); + lock_page(page); ret = page->mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE); @@ -394,7 +442,12 @@ static int bitmap_read_sb(struct bitmap *bitmap) int err = -EINVAL; /* page 0 is the superblock, read it... */ - bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read); + if (bitmap->file) + bitmap->sb_page = read_page(bitmap->file, 0, &bytes_read); + else { + bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, 0); + bytes_read = PAGE_SIZE; + } if (IS_ERR(bitmap->sb_page)) { err = PTR_ERR(bitmap->sb_page); bitmap->sb_page = NULL; @@ -625,14 +678,16 @@ static void bitmap_file_kick(struct bitmap *bitmap) bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET); bitmap_update_sb(bitmap); - path = kmalloc(PAGE_SIZE, GFP_KERNEL); - if (path) - ptr = file_path(bitmap->file, path, PAGE_SIZE); + if (bitmap->file) { + path = kmalloc(PAGE_SIZE, GFP_KERNEL); + if (path) + ptr = file_path(bitmap->file, path, PAGE_SIZE); - printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n", - bmname(bitmap), ptr ? ptr : ""); + printk(KERN_ALERT "%s: kicking failed bitmap file %s from array!\n", + bmname(bitmap), ptr ? ptr : ""); - kfree(path); + kfree(path); + } bitmap_file_put(bitmap); @@ -676,7 +731,7 @@ static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) void *kaddr; unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); - if (!bitmap->file || !bitmap->filemap) { + if (!bitmap->filemap) { return; } @@ -715,7 +770,7 @@ int bitmap_unplug(struct bitmap *bitmap) * flushed out to disk */ for (i = 0; i < bitmap->file_pages; i++) { spin_lock_irqsave(&bitmap->lock, flags); - if (!bitmap->file || !bitmap->filemap) { + if (!bitmap->filemap) { spin_unlock_irqrestore(&bitmap->lock, flags); return 0; } @@ -732,11 +787,15 @@ int bitmap_unplug(struct bitmap *bitmap) return 1; } if (wait) { /* if any writes were performed, we need to wait on them */ - spin_lock_irq(&bitmap->write_lock); - wait_event_lock_irq(bitmap->write_wait, - list_empty(&bitmap->complete_pages), bitmap->write_lock, - wake_up_process(bitmap->writeback_daemon->tsk)); - spin_unlock_irq(&bitmap->write_lock); + if (bitmap->file) { + spin_lock_irq(&bitmap->write_lock); + wait_event_lock_irq(bitmap->write_wait, + list_empty(&bitmap->complete_pages), bitmap->write_lock, + wake_up_process(bitmap->writeback_daemon->tsk)); + spin_unlock_irq(&bitmap->write_lock); + } else + wait_event(bitmap->mddev->sb_wait, + atomic_read(&bitmap->mddev->pending_writes)==0); } return 0; } @@ -764,7 +823,7 @@ static int bitmap_init_from_disk(struct bitmap *bitmap, int in_sync) chunks = bitmap->chunks; file = bitmap->file; - BUG_ON(!file); + BUG_ON(!file && !bitmap->offset); #if INJECT_FAULTS_3 outofdate = 1; @@ -779,7 +838,7 @@ static int bitmap_init_from_disk(struct bitmap *bitmap, int in_sync) num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE; - if (i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { + if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", bmname(bitmap), (unsigned long) i_size_read(file->f_mapping->host), @@ -816,14 +875,18 @@ static int bitmap_init_from_disk(struct bitmap *bitmap, int in_sync) */ page = bitmap->sb_page; offset = sizeof(bitmap_super_t); - } else { + } else if (file) { page = read_page(file, index, &dummy); - if (IS_ERR(page)) { /* read error */ - ret = PTR_ERR(page); - goto out; - } + offset = 0; + } else { + page = read_sb_page(bitmap->mddev, bitmap->offset, index); offset = 0; } + if (IS_ERR(page)) { /* read error */ + ret = PTR_ERR(page); + goto out; + } + oldindex = index; oldpage = page; kmap(page); @@ -874,6 +937,19 @@ out: return ret; } +void bitmap_write_all(struct bitmap *bitmap) +{ + /* We don't actually write all bitmap blocks here, + * just flag them as needing to be written + */ + + unsigned long chunks = bitmap->chunks; + unsigned long bytes = (chunks+7)/8 + sizeof(bitmap_super_t); + unsigned long num_pages = (bytes + PAGE_SIZE-1) / PAGE_SIZE; + while (num_pages--) + bitmap->filemap_attr[num_pages] |= BITMAP_PAGE_NEEDWRITE; +} + static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) { @@ -913,7 +989,7 @@ int bitmap_daemon_work(struct bitmap *bitmap) for (j = 0; j < bitmap->chunks; j++) { bitmap_counter_t *bmc; spin_lock_irqsave(&bitmap->lock, flags); - if (!bitmap->file || !bitmap->filemap) { + if (!bitmap->filemap) { /* error or shutdown */ spin_unlock_irqrestore(&bitmap->lock, flags); break; @@ -1072,6 +1148,7 @@ static int bitmap_start_daemon(struct bitmap *bitmap, mdk_thread_t **ptr, spin_lock_irqsave(&bitmap->lock, flags); *ptr = NULL; + if (!bitmap->file) /* no need for daemon if there's no backing file */ goto out_unlock; @@ -1416,9 +1493,11 @@ int bitmap_create(mddev_t *mddev) BUG_ON(sizeof(bitmap_super_t) != 256); - if (!file) /* bitmap disabled, nothing to do */ + if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */ return 0; + BUG_ON(file && mddev->bitmap_offset); + bitmap = kmalloc(sizeof(*bitmap), GFP_KERNEL); if (!bitmap) return -ENOMEM; @@ -1438,7 +1517,8 @@ int bitmap_create(mddev_t *mddev) return -ENOMEM; bitmap->file = file; - get_file(file); + bitmap->offset = mddev->bitmap_offset; + if (file) get_file(file); /* read superblock from bitmap file (this sets bitmap->chunksize) */ err = bitmap_read_sb(bitmap); if (err) diff --git a/drivers/md/md.c b/drivers/md/md.c index 7075bebb7f3..fde8acfac32 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -337,7 +337,7 @@ static int bi_complete(struct bio *bio, unsigned int bytes_done, int error) return 0; } -static int sync_page_io(struct block_device *bdev, sector_t sector, int size, +int sync_page_io(struct block_device *bdev, sector_t sector, int size, struct page *page, int rw) { struct bio *bio = bio_alloc(GFP_NOIO, 1); @@ -609,6 +609,17 @@ static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) memcpy(mddev->uuid+12,&sb->set_uuid3, 4); mddev->max_disks = MD_SB_DISKS; + + if (sb->state & (1<bitmap_file == NULL) { + if (mddev->level != 1) { + /* FIXME use a better test */ + printk(KERN_WARNING "md: bitmaps only support for raid1\n"); + return -EINVAL; + } + mddev->bitmap_offset = (MD_SB_BYTES >> 9); + } + } else if (mddev->pers == NULL) { /* Insist on good event counter while assembling */ __u64 ev1 = md_event(sb); @@ -702,6 +713,9 @@ static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) sb->layout = mddev->layout; sb->chunk_size = mddev->chunk_size; + if (mddev->bitmap && mddev->bitmap_file == NULL) + sb->state |= (1<disks[0].state = (1<uuid, sb->set_uuid, 16); mddev->max_disks = (4096-256)/2; + + if ((le32_to_cpu(sb->feature_map) & 1) && + mddev->bitmap_file == NULL ) { + if (mddev->level != 1) { + printk(KERN_WARNING "md: bitmaps only supported for raid1\n"); + return -EINVAL; + } + mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset); + } } else if (mddev->pers == NULL) { /* Insist of good event counter while assembling */ __u64 ev1 = le64_to_cpu(sb->events); @@ -960,6 +983,11 @@ static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) else sb->resync_offset = cpu_to_le64(0); + if (mddev->bitmap && mddev->bitmap_file == NULL) { + sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset); + sb->feature_map = cpu_to_le32(1); + } + max_dev = 0; ITERATE_RDEV(mddev,rdev2,tmp) if (rdev2->desc_nr+1 > max_dev) @@ -2406,7 +2434,8 @@ static int set_bitmap_file(mddev_t *mddev, int fd) mdname(mddev)); fput(mddev->bitmap_file); mddev->bitmap_file = NULL; - } + } else + mddev->bitmap_offset = 0; /* file overrides offset */ return err; } @@ -3774,6 +3803,13 @@ void md_check_recovery(mddev_t *mddev) set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); if (!spares) set_bit(MD_RECOVERY_SYNC, &mddev->recovery); + if (spares && mddev->bitmap && ! mddev->bitmap->file) { + /* We are adding a device or devices to an array + * which has the bitmap stored on all devices. + * So make sure all bitmap pages get written + */ + bitmap_write_all(mddev->bitmap); + } mddev->sync_thread = md_register_thread(md_do_sync, mddev, "%s_resync"); diff --git a/include/linux/raid/bitmap.h b/include/linux/raid/bitmap.h index cfe60cfc8f3..e24b74b1115 100644 --- a/include/linux/raid/bitmap.h +++ b/include/linux/raid/bitmap.h @@ -217,6 +217,7 @@ struct bitmap { /* bitmap spinlock */ spinlock_t lock; + long offset; /* offset from superblock if file is NULL */ struct file *file; /* backing disk file */ struct page *sb_page; /* cached copy of the bitmap file superblock */ struct page **filemap; /* list of cache pages for the file */ @@ -255,6 +256,7 @@ void bitmap_print_sb(struct bitmap *bitmap); int bitmap_update_sb(struct bitmap *bitmap); int bitmap_setallbits(struct bitmap *bitmap); +void bitmap_write_all(struct bitmap *bitmap); /* these are exported */ int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors); diff --git a/include/linux/raid/md.h b/include/linux/raid/md.h index 75f41d8faed..ffa316ce4dc 100644 --- a/include/linux/raid/md.h +++ b/include/linux/raid/md.h @@ -60,7 +60,14 @@ */ #define MD_MAJOR_VERSION 0 #define MD_MINOR_VERSION 90 -#define MD_PATCHLEVEL_VERSION 1 +/* + * MD_PATCHLEVEL_VERSION indicates kernel functionality. + * >=1 means different superblock formats are selectable using SET_ARRAY_INFO + * and major_version/minor_version accordingly + * >=2 means that Internal bitmaps are supported by setting MD_SB_BITMAP_PRESENT + * in the super status byte + */ +#define MD_PATCHLEVEL_VERSION 2 extern int register_md_personality (int p_num, mdk_personality_t *p); extern int unregister_md_personality (int p_num); @@ -78,6 +85,12 @@ extern void md_unplug_mddev(mddev_t *mddev); extern void md_print_devices (void); +extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, + sector_t sector, int size, struct page *page); +extern int sync_page_io(struct block_device *bdev, sector_t sector, int size, + struct page *page, int rw); + + #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); } #endif diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index 3e977025cf4..a3725b57fb7 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -273,6 +273,10 @@ struct mddev_s struct bitmap *bitmap; /* the bitmap for the device */ struct file *bitmap_file; /* the bitmap file */ + long bitmap_offset; /* offset from superblock of + * start of bitmap. May be + * negative, but not '0' + */ struct list_head all_mddevs; }; diff --git a/include/linux/raid/md_p.h b/include/linux/raid/md_p.h index 8ba95d67329..8e592a25a8b 100644 --- a/include/linux/raid/md_p.h +++ b/include/linux/raid/md_p.h @@ -96,6 +96,7 @@ typedef struct mdp_device_descriptor_s { #define MD_SB_CLEAN 0 #define MD_SB_ERRORS 1 +#define MD_SB_BITMAP_PRESENT 8 /* bitmap may be present nearby */ typedef struct mdp_superblock_s { /* * Constant generic information @@ -184,7 +185,7 @@ struct mdp_superblock_1 { /* constant array information - 128 bytes */ __u32 magic; /* MD_SB_MAGIC: 0xa92b4efc - little endian */ __u32 major_version; /* 1 */ - __u32 feature_map; /* 0 for now */ + __u32 feature_map; /* bit 0 set if 'bitmap_offset' is meaningful */ __u32 pad0; /* always set to 0 when writing */ __u8 set_uuid[16]; /* user-space generated. */ @@ -197,6 +198,10 @@ struct mdp_superblock_1 { __u32 chunksize; /* in 512byte sectors */ __u32 raid_disks; + __u32 bitmap_offset; /* sectors after start of superblock that bitmap starts + * NOTE: signed, so bitmap can be before superblock + * only meaningful of feature_map[0] is set. + */ __u8 pad1[128-96]; /* set to 0 when written */ /* constant this-device information - 64 bytes */ -- cgit v1.2.3-70-g09d2 From 7bfa19f2748000d646dbdf8f48258cfe1d257b52 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:28 -0700 Subject: [PATCH] md: allow md to update multiple superblocks in parallel. currently, md updates all superblocks (one on each device) in series. It waits for one write to complete before starting the next. This isn't a big problem as superblock updates don't happen that often. However it is neater to do it in parallel, and if the drives in the array have gone to "sleep" after a period of idleness, then waking them is parallel is faster (and someone else should be worrying about power drain). Futher, we will need parallel superblock updates for a future patch which keeps the intent-logging bitmap near the superblock. Also remove the silly code that retired superblock updates 100 times. This simply never made sense. Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/md.c | 85 ++++++++++++++++++++++++++--------------------- include/linux/raid/md_k.h | 1 + 2 files changed, 49 insertions(+), 37 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/md.c b/drivers/md/md.c index fde8acfac32..ef3ad99562c 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -328,6 +328,40 @@ static void free_disk_sb(mdk_rdev_t * rdev) } +static int super_written(struct bio *bio, unsigned int bytes_done, int error) +{ + mdk_rdev_t *rdev = bio->bi_private; + if (bio->bi_size) + return 1; + + if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) + md_error(rdev->mddev, rdev); + + if (atomic_dec_and_test(&rdev->mddev->pending_writes)) + wake_up(&rdev->mddev->sb_wait); + return 0; +} + +void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, + sector_t sector, int size, struct page *page) +{ + /* write first size bytes of page to sector of rdev + * Increment mddev->pending_writes before returning + * and decrement it on completion, waking up sb_wait + * if zero is reached. + * If an error occurred, call md_error + */ + struct bio *bio = bio_alloc(GFP_NOIO, 1); + + bio->bi_bdev = rdev->bdev; + bio->bi_sector = sector; + bio_add_page(bio, page, size, 0); + bio->bi_private = rdev; + bio->bi_end_io = super_written; + atomic_inc(&mddev->pending_writes); + submit_bio((1<bi_size) @@ -1268,30 +1302,6 @@ void md_print_devices(void) } -static int write_disk_sb(mdk_rdev_t * rdev) -{ - char b[BDEVNAME_SIZE]; - if (!rdev->sb_loaded) { - MD_BUG(); - return 1; - } - if (rdev->faulty) { - MD_BUG(); - return 1; - } - - dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", - bdevname(rdev->bdev,b), - (unsigned long long)rdev->sb_offset); - - if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE)) - return 0; - - printk("md: write_disk_sb failed for device %s\n", - bdevname(rdev->bdev,b)); - return 1; -} - static void sync_sbs(mddev_t * mddev) { mdk_rdev_t *rdev; @@ -1306,7 +1316,7 @@ static void sync_sbs(mddev_t * mddev) static void md_update_sb(mddev_t * mddev) { - int err, count = 100; + int err; struct list_head *tmp; mdk_rdev_t *rdev; int sync_req; @@ -1326,6 +1336,7 @@ repeat: MD_BUG(); mddev->events --; } + mddev->sb_dirty = 2; sync_sbs(mddev); /* @@ -1353,24 +1364,24 @@ repeat: dprintk("%s ", bdevname(rdev->bdev,b)); if (!rdev->faulty) { - err += write_disk_sb(rdev); + md_super_write(mddev,rdev, + rdev->sb_offset<<1, MD_SB_BYTES, + rdev->sb_page); + dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", + bdevname(rdev->bdev,b), + (unsigned long long)rdev->sb_offset); + } else dprintk(")\n"); - if (!err && mddev->level == LEVEL_MULTIPATH) + if (mddev->level == LEVEL_MULTIPATH) /* only need to write one superblock... */ break; } - if (err) { - if (--count) { - printk(KERN_ERR "md: errors occurred during superblock" - " update, repeating\n"); - goto repeat; - } - printk(KERN_ERR \ - "md: excessive errors occurred during superblock update, exiting\n"); - } + wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0); + /* if there was a failure, sb_dirty was set to 1, and we re-write super */ + spin_lock(&mddev->write_lock); - if (mddev->in_sync != sync_req) { + if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) { /* have to write it out again */ spin_unlock(&mddev->write_lock); goto repeat; diff --git a/include/linux/raid/md_k.h b/include/linux/raid/md_k.h index a3725b57fb7..8c14ba565a4 100644 --- a/include/linux/raid/md_k.h +++ b/include/linux/raid/md_k.h @@ -262,6 +262,7 @@ struct mddev_s spinlock_t write_lock; wait_queue_head_t sb_wait; /* for waiting on superblock updates */ + atomic_t pending_writes; /* number of active superblock writes */ unsigned int safemode; /* if set, update "clean" superblock * when no writes pending. -- cgit v1.2.3-70-g09d2 From 39730960d94306d7be414e8d54f4e5c071af1278 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 21 Jun 2005 17:17:28 -0700 Subject: [PATCH] Two small fixes for md verion-1 superblocks. 1/ Must typecast int to (sector_t) before inverting or we might not invert enough bits. 2/ When "bitmap_offset" was added to mdp_superblock_1, we didn't increase the count of words-used (96 to 100). Signed-off-by: Neil Brown Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/md/md.c | 2 +- include/linux/raid/md_p.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'include/linux/raid') diff --git a/drivers/md/md.c b/drivers/md/md.c index ef3ad99562c..65fab74ad73 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -847,7 +847,7 @@ static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) case 0: sb_offset = rdev->bdev->bd_inode->i_size >> 9; sb_offset -= 8*2; - sb_offset &= ~(4*2-1); + sb_offset &= ~(sector_t)(4*2-1); /* convert from sectors to K */ sb_offset /= 2; break; diff --git a/include/linux/raid/md_p.h b/include/linux/raid/md_p.h index 8e592a25a8b..dc65cd43549 100644 --- a/include/linux/raid/md_p.h +++ b/include/linux/raid/md_p.h @@ -202,7 +202,7 @@ struct mdp_superblock_1 { * NOTE: signed, so bitmap can be before superblock * only meaningful of feature_map[0] is set. */ - __u8 pad1[128-96]; /* set to 0 when written */ + __u8 pad1[128-100]; /* set to 0 when written */ /* constant this-device information - 64 bytes */ __u64 data_offset; /* sector start of data, often 0 */ -- cgit v1.2.3-70-g09d2