dm kcopyd: rename
authorAlasdair G Kergon <agk@redhat.com>
Thu, 24 Apr 2008 20:55:00 +0000 (21:55 +0100)
committerAlasdair G Kergon <agk@redhat.com>
Fri, 25 Apr 2008 12:26:54 +0000 (13:26 +0100)
Rename kcopyd.[ch] to dm-kcopyd.[ch].

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
drivers/md/dm-kcopyd.c [new file with mode: 0644]
drivers/md/dm-kcopyd.h [new file with mode: 0644]
drivers/md/kcopyd.c [deleted file]
drivers/md/kcopyd.h [deleted file]

diff --git a/drivers/md/dm-kcopyd.c b/drivers/md/dm-kcopyd.c
new file mode 100644 (file)
index 0000000..1734584
--- /dev/null
@@ -0,0 +1,654 @@
+/*
+ * Copyright (C) 2002 Sistina Software (UK) Limited.
+ * Copyright (C) 2006 Red Hat GmbH
+ *
+ * This file is released under the GPL.
+ *
+ * Kcopyd provides a simple interface for copying an area of one
+ * block-device to one or more other block-devices, with an asynchronous
+ * completion notification.
+ */
+
+#include <linux/types.h>
+#include <asm/atomic.h>
+#include <linux/blkdev.h>
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/mempool.h>
+#include <linux/module.h>
+#include <linux/pagemap.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/workqueue.h>
+#include <linux/mutex.h>
+
+#include "kcopyd.h"
+#include "dm.h"
+
+/*-----------------------------------------------------------------
+ * Each kcopyd client has its own little pool of preallocated
+ * pages for kcopyd io.
+ *---------------------------------------------------------------*/
+struct dm_kcopyd_client {
+       spinlock_t lock;
+       struct page_list *pages;
+       unsigned int nr_pages;
+       unsigned int nr_free_pages;
+
+       struct dm_io_client *io_client;
+
+       wait_queue_head_t destroyq;
+       atomic_t nr_jobs;
+
+       mempool_t *job_pool;
+
+       struct workqueue_struct *kcopyd_wq;
+       struct work_struct kcopyd_work;
+
+/*
+ * We maintain three lists of jobs:
+ *
+ * i)   jobs waiting for pages
+ * ii)  jobs that have pages, and are waiting for the io to be issued.
+ * iii) jobs that have completed.
+ *
+ * All three of these are protected by job_lock.
+ */
+       spinlock_t job_lock;
+       struct list_head complete_jobs;
+       struct list_head io_jobs;
+       struct list_head pages_jobs;
+};
+
+static void wake(struct dm_kcopyd_client *kc)
+{
+       queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
+}
+
+static struct page_list *alloc_pl(void)
+{
+       struct page_list *pl;
+
+       pl = kmalloc(sizeof(*pl), GFP_KERNEL);
+       if (!pl)
+               return NULL;
+
+       pl->page = alloc_page(GFP_KERNEL);
+       if (!pl->page) {
+               kfree(pl);
+               return NULL;
+       }
+
+       return pl;
+}
+
+static void free_pl(struct page_list *pl)
+{
+       __free_page(pl->page);
+       kfree(pl);
+}
+
+static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
+                           unsigned int nr, struct page_list **pages)
+{
+       struct page_list *pl;
+
+       spin_lock(&kc->lock);
+       if (kc->nr_free_pages < nr) {
+               spin_unlock(&kc->lock);
+               return -ENOMEM;
+       }
+
+       kc->nr_free_pages -= nr;
+       for (*pages = pl = kc->pages; --nr; pl = pl->next)
+               ;
+
+       kc->pages = pl->next;
+       pl->next = NULL;
+
+       spin_unlock(&kc->lock);
+
+       return 0;
+}
+
+static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
+{
+       struct page_list *cursor;
+
+       spin_lock(&kc->lock);
+       for (cursor = pl; cursor->next; cursor = cursor->next)
+               kc->nr_free_pages++;
+
+       kc->nr_free_pages++;
+       cursor->next = kc->pages;
+       kc->pages = pl;
+       spin_unlock(&kc->lock);
+}
+
+/*
+ * These three functions resize the page pool.
+ */
+static void drop_pages(struct page_list *pl)
+{
+       struct page_list *next;
+
+       while (pl) {
+               next = pl->next;
+               free_pl(pl);
+               pl = next;
+       }
+}
+
+static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
+{
+       unsigned int i;
+       struct page_list *pl = NULL, *next;
+
+       for (i = 0; i < nr; i++) {
+               next = alloc_pl();
+               if (!next) {
+                       if (pl)
+                               drop_pages(pl);
+                       return -ENOMEM;
+               }
+               next->next = pl;
+               pl = next;
+       }
+
+       kcopyd_put_pages(kc, pl);
+       kc->nr_pages += nr;
+       return 0;
+}
+
+static void client_free_pages(struct dm_kcopyd_client *kc)
+{
+       BUG_ON(kc->nr_free_pages != kc->nr_pages);
+       drop_pages(kc->pages);
+       kc->pages = NULL;
+       kc->nr_free_pages = kc->nr_pages = 0;
+}
+
+/*-----------------------------------------------------------------
+ * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
+ * for this reason we use a mempool to prevent the client from
+ * ever having to do io (which could cause a deadlock).
+ *---------------------------------------------------------------*/
+struct kcopyd_job {
+       struct dm_kcopyd_client *kc;
+       struct list_head list;
+       unsigned long flags;
+
+       /*
+        * Error state of the job.
+        */
+       int read_err;
+       unsigned long write_err;
+
+       /*
+        * Either READ or WRITE
+        */
+       int rw;
+       struct dm_io_region source;
+
+       /*
+        * The destinations for the transfer.
+        */
+       unsigned int num_dests;
+       struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
+
+       sector_t offset;
+       unsigned int nr_pages;
+       struct page_list *pages;
+
+       /*
+        * Set this to ensure you are notified when the job has
+        * completed.  'context' is for callback to use.
+        */
+       dm_kcopyd_notify_fn fn;
+       void *context;
+
+       /*
+        * These fields are only used if the job has been split
+        * into more manageable parts.
+        */
+       struct mutex lock;
+       atomic_t sub_jobs;
+       sector_t progress;
+};
+
+/* FIXME: this should scale with the number of pages */
+#define MIN_JOBS 512
+
+static struct kmem_cache *_job_cache;
+
+int __init dm_kcopyd_init(void)
+{
+       _job_cache = KMEM_CACHE(kcopyd_job, 0);
+       if (!_job_cache)
+               return -ENOMEM;
+
+       return 0;
+}
+
+void dm_kcopyd_exit(void)
+{
+       kmem_cache_destroy(_job_cache);
+       _job_cache = NULL;
+}
+
+/*
+ * Functions to push and pop a job onto the head of a given job
+ * list.
+ */
+static struct kcopyd_job *pop(struct list_head *jobs,
+                             struct dm_kcopyd_client *kc)
+{
+       struct kcopyd_job *job = NULL;
+       unsigned long flags;
+
+       spin_lock_irqsave(&kc->job_lock, flags);
+
+       if (!list_empty(jobs)) {
+               job = list_entry(jobs->next, struct kcopyd_job, list);
+               list_del(&job->list);
+       }
+       spin_unlock_irqrestore(&kc->job_lock, flags);
+
+       return job;
+}
+
+static void push(struct list_head *jobs, struct kcopyd_job *job)
+{
+       unsigned long flags;
+       struct dm_kcopyd_client *kc = job->kc;
+
+       spin_lock_irqsave(&kc->job_lock, flags);
+       list_add_tail(&job->list, jobs);
+       spin_unlock_irqrestore(&kc->job_lock, flags);
+}
+
+/*
+ * These three functions process 1 item from the corresponding
+ * job list.
+ *
+ * They return:
+ * < 0: error
+ *   0: success
+ * > 0: can't process yet.
+ */
+static int run_complete_job(struct kcopyd_job *job)
+{
+       void *context = job->context;
+       int read_err = job->read_err;
+       unsigned long write_err = job->write_err;
+       dm_kcopyd_notify_fn fn = job->fn;
+       struct dm_kcopyd_client *kc = job->kc;
+
+       kcopyd_put_pages(kc, job->pages);
+       mempool_free(job, kc->job_pool);
+       fn(read_err, write_err, context);
+
+       if (atomic_dec_and_test(&kc->nr_jobs))
+               wake_up(&kc->destroyq);
+
+       return 0;
+}
+
+static void complete_io(unsigned long error, void *context)
+{
+       struct kcopyd_job *job = (struct kcopyd_job *) context;
+       struct dm_kcopyd_client *kc = job->kc;
+
+       if (error) {
+               if (job->rw == WRITE)
+                       job->write_err |= error;
+               else
+                       job->read_err = 1;
+
+               if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
+                       push(&kc->complete_jobs, job);
+                       wake(kc);
+                       return;
+               }
+       }
+
+       if (job->rw == WRITE)
+               push(&kc->complete_jobs, job);
+
+       else {
+               job->rw = WRITE;
+               push(&kc->io_jobs, job);
+       }
+
+       wake(kc);
+}
+
+/*
+ * Request io on as many buffer heads as we can currently get for
+ * a particular job.
+ */
+static int run_io_job(struct kcopyd_job *job)
+{
+       int r;
+       struct dm_io_request io_req = {
+               .bi_rw = job->rw,
+               .mem.type = DM_IO_PAGE_LIST,
+               .mem.ptr.pl = job->pages,
+               .mem.offset = job->offset,
+               .notify.fn = complete_io,
+               .notify.context = job,
+               .client = job->kc->io_client,
+       };
+
+       if (job->rw == READ)
+               r = dm_io(&io_req, 1, &job->source, NULL);
+       else
+               r = dm_io(&io_req, job->num_dests, job->dests, NULL);
+
+       return r;
+}
+
+static int run_pages_job(struct kcopyd_job *job)
+{
+       int r;
+
+       job->nr_pages = dm_div_up(job->dests[0].count + job->offset,
+                                 PAGE_SIZE >> 9);
+       r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages);
+       if (!r) {
+               /* this job is ready for io */
+               push(&job->kc->io_jobs, job);
+               return 0;
+       }
+
+       if (r == -ENOMEM)
+               /* can't complete now */
+               return 1;
+
+       return r;
+}
+
+/*
+ * Run through a list for as long as possible.  Returns the count
+ * of successful jobs.
+ */
+static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
+                       int (*fn) (struct kcopyd_job *))
+{
+       struct kcopyd_job *job;
+       int r, count = 0;
+
+       while ((job = pop(jobs, kc))) {
+
+               r = fn(job);
+
+               if (r < 0) {
+                       /* error this rogue job */
+                       if (job->rw == WRITE)
+                               job->write_err = (unsigned long) -1L;
+                       else
+                               job->read_err = 1;
+                       push(&kc->complete_jobs, job);
+                       break;
+               }
+
+               if (r > 0) {
+                       /*
+                        * We couldn't service this job ATM, so
+                        * push this job back onto the list.
+                        */
+                       push(jobs, job);
+                       break;
+               }
+
+               count++;
+       }
+
+       return count;
+}
+
+/*
+ * kcopyd does this every time it's woken up.
+ */
+static void do_work(struct work_struct *work)
+{
+       struct dm_kcopyd_client *kc = container_of(work,
+                                       struct dm_kcopyd_client, kcopyd_work);
+
+       /*
+        * The order that these are called is *very* important.
+        * complete jobs can free some pages for pages jobs.
+        * Pages jobs when successful will jump onto the io jobs
+        * list.  io jobs call wake when they complete and it all
+        * starts again.
+        */
+       process_jobs(&kc->complete_jobs, kc, run_complete_job);
+       process_jobs(&kc->pages_jobs, kc, run_pages_job);
+       process_jobs(&kc->io_jobs, kc, run_io_job);
+}
+
+/*
+ * If we are copying a small region we just dispatch a single job
+ * to do the copy, otherwise the io has to be split up into many
+ * jobs.
+ */
+static void dispatch_job(struct kcopyd_job *job)
+{
+       struct dm_kcopyd_client *kc = job->kc;
+       atomic_inc(&kc->nr_jobs);
+       push(&kc->pages_jobs, job);
+       wake(kc);
+}
+
+#define SUB_JOB_SIZE 128
+static void segment_complete(int read_err, unsigned long write_err,
+                            void *context)
+{
+       /* FIXME: tidy this function */
+       sector_t progress = 0;
+       sector_t count = 0;
+       struct kcopyd_job *job = (struct kcopyd_job *) context;
+
+       mutex_lock(&job->lock);
+
+       /* update the error */
+       if (read_err)
+               job->read_err = 1;
+
+       if (write_err)
+               job->write_err |= write_err;
+
+       /*
+        * Only dispatch more work if there hasn't been an error.
+        */
+       if ((!job->read_err && !job->write_err) ||
+           test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
+               /* get the next chunk of work */
+               progress = job->progress;
+               count = job->source.count - progress;
+               if (count) {
+                       if (count > SUB_JOB_SIZE)
+                               count = SUB_JOB_SIZE;
+
+                       job->progress += count;
+               }
+       }
+       mutex_unlock(&job->lock);
+
+       if (count) {
+               int i;
+               struct kcopyd_job *sub_job = mempool_alloc(job->kc->job_pool,
+                                                          GFP_NOIO);
+
+               *sub_job = *job;
+               sub_job->source.sector += progress;
+               sub_job->source.count = count;
+
+               for (i = 0; i < job->num_dests; i++) {
+                       sub_job->dests[i].sector += progress;
+                       sub_job->dests[i].count = count;
+               }
+
+               sub_job->fn = segment_complete;
+               sub_job->context = job;
+               dispatch_job(sub_job);
+
+       } else if (atomic_dec_and_test(&job->sub_jobs)) {
+
+               /*
+                * To avoid a race we must keep the job around
+                * until after the notify function has completed.
+                * Otherwise the client may try and stop the job
+                * after we've completed.
+                */
+               job->fn(read_err, write_err, job->context);
+               mempool_free(job, job->kc->job_pool);
+       }
+}
+
+/*
+ * Create some little jobs that will do the move between
+ * them.
+ */
+#define SPLIT_COUNT 8
+static void split_job(struct kcopyd_job *job)
+{
+       int i;
+
+       atomic_set(&job->sub_jobs, SPLIT_COUNT);
+       for (i = 0; i < SPLIT_COUNT; i++)
+               segment_complete(0, 0u, job);
+}
+
+int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
+                  unsigned int num_dests, struct dm_io_region *dests,
+                  unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
+{
+       struct kcopyd_job *job;
+
+       /*
+        * Allocate a new job.
+        */
+       job = mempool_alloc(kc->job_pool, GFP_NOIO);
+
+       /*
+        * set up for the read.
+        */
+       job->kc = kc;
+       job->flags = flags;
+       job->read_err = 0;
+       job->write_err = 0;
+       job->rw = READ;
+
+       job->source = *from;
+
+       job->num_dests = num_dests;
+       memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
+
+       job->offset = 0;
+       job->nr_pages = 0;
+       job->pages = NULL;
+
+       job->fn = fn;
+       job->context = context;
+
+       if (job->source.count < SUB_JOB_SIZE)
+               dispatch_job(job);
+
+       else {
+               mutex_init(&job->lock);
+               job->progress = 0;
+               split_job(job);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(dm_kcopyd_copy);
+
+/*
+ * Cancels a kcopyd job, eg. someone might be deactivating a
+ * mirror.
+ */
+#if 0
+int kcopyd_cancel(struct kcopyd_job *job, int block)
+{
+       /* FIXME: finish */
+       return -1;
+}
+#endif  /*  0  */
+
+/*-----------------------------------------------------------------
+ * Client setup
+ *---------------------------------------------------------------*/
+int dm_kcopyd_client_create(unsigned int nr_pages,
+                           struct dm_kcopyd_client **result)
+{
+       int r = -ENOMEM;
+       struct dm_kcopyd_client *kc;
+
+       kc = kmalloc(sizeof(*kc), GFP_KERNEL);
+       if (!kc)
+               return -ENOMEM;
+
+       spin_lock_init(&kc->lock);
+       spin_lock_init(&kc->job_lock);
+       INIT_LIST_HEAD(&kc->complete_jobs);
+       INIT_LIST_HEAD(&kc->io_jobs);
+       INIT_LIST_HEAD(&kc->pages_jobs);
+
+       kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
+       if (!kc->job_pool)
+               goto bad_slab;
+
+       INIT_WORK(&kc->kcopyd_work, do_work);
+       kc->kcopyd_wq = create_singlethread_workqueue("kcopyd");
+       if (!kc->kcopyd_wq)
+               goto bad_workqueue;
+
+       kc->pages = NULL;
+       kc->nr_pages = kc->nr_free_pages = 0;
+       r = client_alloc_pages(kc, nr_pages);
+       if (r)
+               goto bad_client_pages;
+
+       kc->io_client = dm_io_client_create(nr_pages);
+       if (IS_ERR(kc->io_client)) {
+               r = PTR_ERR(kc->io_client);
+               goto bad_io_client;
+       }
+
+       init_waitqueue_head(&kc->destroyq);
+       atomic_set(&kc->nr_jobs, 0);
+
+       *result = kc;
+       return 0;
+
+bad_io_client:
+       client_free_pages(kc);
+bad_client_pages:
+       destroy_workqueue(kc->kcopyd_wq);
+bad_workqueue:
+       mempool_destroy(kc->job_pool);
+bad_slab:
+       kfree(kc);
+
+       return r;
+}
+EXPORT_SYMBOL(dm_kcopyd_client_create);
+
+void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
+{
+       /* Wait for completion of all jobs submitted by this client. */
+       wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
+
+       BUG_ON(!list_empty(&kc->complete_jobs));
+       BUG_ON(!list_empty(&kc->io_jobs));
+       BUG_ON(!list_empty(&kc->pages_jobs));
+       destroy_workqueue(kc->kcopyd_wq);
+       dm_io_client_destroy(kc->io_client);
+       client_free_pages(kc);
+       mempool_destroy(kc->job_pool);
+       kfree(kc);
+}
+EXPORT_SYMBOL(dm_kcopyd_client_destroy);
diff --git a/drivers/md/dm-kcopyd.h b/drivers/md/dm-kcopyd.h
new file mode 100644 (file)
index 0000000..d305784
--- /dev/null
@@ -0,0 +1,47 @@
+/*
+ * Copyright (C) 2001 - 2003 Sistina Software
+ * Copyright (C) 2004 - 2008 Red Hat, Inc. All rights reserved.
+ *
+ * kcopyd provides a simple interface for copying an area of one
+ * block-device to one or more other block-devices, either synchronous
+ * or with an asynchronous completion notification.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _LINUX_DM_KCOPYD_H
+#define _LINUX_DM_KCOPYD_H
+
+#ifdef __KERNEL__
+
+#include "dm-io.h"
+
+/* FIXME: make this configurable */
+#define DM_KCOPYD_MAX_REGIONS 8
+
+#define DM_KCOPYD_IGNORE_ERROR 1
+
+/*
+ * To use kcopyd you must first create a dm_kcopyd_client object.
+ */
+struct dm_kcopyd_client;
+int dm_kcopyd_client_create(unsigned num_pages,
+                           struct dm_kcopyd_client **result);
+void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc);
+
+/*
+ * Submit a copy job to kcopyd.  This is built on top of the
+ * previous three fns.
+ *
+ * read_err is a boolean,
+ * write_err is a bitset, with 1 bit for each destination region
+ */
+typedef void (*dm_kcopyd_notify_fn)(int read_err, unsigned long write_err,
+                                   void *context);
+
+int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
+                  unsigned num_dests, struct dm_io_region *dests,
+                  unsigned flags, dm_kcopyd_notify_fn fn, void *context);
+
+#endif /* __KERNEL__ */
+#endif /* _LINUX_DM_KCOPYD_H */
diff --git a/drivers/md/kcopyd.c b/drivers/md/kcopyd.c
deleted file mode 100644 (file)
index 1734584..0000000
+++ /dev/null
@@ -1,654 +0,0 @@
-/*
- * Copyright (C) 2002 Sistina Software (UK) Limited.
- * Copyright (C) 2006 Red Hat GmbH
- *
- * This file is released under the GPL.
- *
- * Kcopyd provides a simple interface for copying an area of one
- * block-device to one or more other block-devices, with an asynchronous
- * completion notification.
- */
-
-#include <linux/types.h>
-#include <asm/atomic.h>
-#include <linux/blkdev.h>
-#include <linux/fs.h>
-#include <linux/init.h>
-#include <linux/list.h>
-#include <linux/mempool.h>
-#include <linux/module.h>
-#include <linux/pagemap.h>
-#include <linux/slab.h>
-#include <linux/vmalloc.h>
-#include <linux/workqueue.h>
-#include <linux/mutex.h>
-
-#include "kcopyd.h"
-#include "dm.h"
-
-/*-----------------------------------------------------------------
- * Each kcopyd client has its own little pool of preallocated
- * pages for kcopyd io.
- *---------------------------------------------------------------*/
-struct dm_kcopyd_client {
-       spinlock_t lock;
-       struct page_list *pages;
-       unsigned int nr_pages;
-       unsigned int nr_free_pages;
-
-       struct dm_io_client *io_client;
-
-       wait_queue_head_t destroyq;
-       atomic_t nr_jobs;
-
-       mempool_t *job_pool;
-
-       struct workqueue_struct *kcopyd_wq;
-       struct work_struct kcopyd_work;
-
-/*
- * We maintain three lists of jobs:
- *
- * i)   jobs waiting for pages
- * ii)  jobs that have pages, and are waiting for the io to be issued.
- * iii) jobs that have completed.
- *
- * All three of these are protected by job_lock.
- */
-       spinlock_t job_lock;
-       struct list_head complete_jobs;
-       struct list_head io_jobs;
-       struct list_head pages_jobs;
-};
-
-static void wake(struct dm_kcopyd_client *kc)
-{
-       queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
-}
-
-static struct page_list *alloc_pl(void)
-{
-       struct page_list *pl;
-
-       pl = kmalloc(sizeof(*pl), GFP_KERNEL);
-       if (!pl)
-               return NULL;
-
-       pl->page = alloc_page(GFP_KERNEL);
-       if (!pl->page) {
-               kfree(pl);
-               return NULL;
-       }
-
-       return pl;
-}
-
-static void free_pl(struct page_list *pl)
-{
-       __free_page(pl->page);
-       kfree(pl);
-}
-
-static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
-                           unsigned int nr, struct page_list **pages)
-{
-       struct page_list *pl;
-
-       spin_lock(&kc->lock);
-       if (kc->nr_free_pages < nr) {
-               spin_unlock(&kc->lock);
-               return -ENOMEM;
-       }
-
-       kc->nr_free_pages -= nr;
-       for (*pages = pl = kc->pages; --nr; pl = pl->next)
-               ;
-
-       kc->pages = pl->next;
-       pl->next = NULL;
-
-       spin_unlock(&kc->lock);
-
-       return 0;
-}
-
-static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
-{
-       struct page_list *cursor;
-
-       spin_lock(&kc->lock);
-       for (cursor = pl; cursor->next; cursor = cursor->next)
-               kc->nr_free_pages++;
-
-       kc->nr_free_pages++;
-       cursor->next = kc->pages;
-       kc->pages = pl;
-       spin_unlock(&kc->lock);
-}
-
-/*
- * These three functions resize the page pool.
- */
-static void drop_pages(struct page_list *pl)
-{
-       struct page_list *next;
-
-       while (pl) {
-               next = pl->next;
-               free_pl(pl);
-               pl = next;
-       }
-}
-
-static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
-{
-       unsigned int i;
-       struct page_list *pl = NULL, *next;
-
-       for (i = 0; i < nr; i++) {
-               next = alloc_pl();
-               if (!next) {
-                       if (pl)
-                               drop_pages(pl);
-                       return -ENOMEM;
-               }
-               next->next = pl;
-               pl = next;
-       }
-
-       kcopyd_put_pages(kc, pl);
-       kc->nr_pages += nr;
-       return 0;
-}
-
-static void client_free_pages(struct dm_kcopyd_client *kc)
-{
-       BUG_ON(kc->nr_free_pages != kc->nr_pages);
-       drop_pages(kc->pages);
-       kc->pages = NULL;
-       kc->nr_free_pages = kc->nr_pages = 0;
-}
-
-/*-----------------------------------------------------------------
- * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
- * for this reason we use a mempool to prevent the client from
- * ever having to do io (which could cause a deadlock).
- *---------------------------------------------------------------*/
-struct kcopyd_job {
-       struct dm_kcopyd_client *kc;
-       struct list_head list;
-       unsigned long flags;
-
-       /*
-        * Error state of the job.
-        */
-       int read_err;
-       unsigned long write_err;
-
-       /*
-        * Either READ or WRITE
-        */
-       int rw;
-       struct dm_io_region source;
-
-       /*
-        * The destinations for the transfer.
-        */
-       unsigned int num_dests;
-       struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
-
-       sector_t offset;
-       unsigned int nr_pages;
-       struct page_list *pages;
-
-       /*
-        * Set this to ensure you are notified when the job has
-        * completed.  'context' is for callback to use.
-        */
-       dm_kcopyd_notify_fn fn;
-       void *context;
-
-       /*
-        * These fields are only used if the job has been split
-        * into more manageable parts.
-        */
-       struct mutex lock;
-       atomic_t sub_jobs;
-       sector_t progress;
-};
-
-/* FIXME: this should scale with the number of pages */
-#define MIN_JOBS 512
-
-static struct kmem_cache *_job_cache;
-
-int __init dm_kcopyd_init(void)
-{
-       _job_cache = KMEM_CACHE(kcopyd_job, 0);
-       if (!_job_cache)
-               return -ENOMEM;
-
-       return 0;
-}
-
-void dm_kcopyd_exit(void)
-{
-       kmem_cache_destroy(_job_cache);
-       _job_cache = NULL;
-}
-
-/*
- * Functions to push and pop a job onto the head of a given job
- * list.
- */
-static struct kcopyd_job *pop(struct list_head *jobs,
-                             struct dm_kcopyd_client *kc)
-{
-       struct kcopyd_job *job = NULL;
-       unsigned long flags;
-
-       spin_lock_irqsave(&kc->job_lock, flags);
-
-       if (!list_empty(jobs)) {
-               job = list_entry(jobs->next, struct kcopyd_job, list);
-               list_del(&job->list);
-       }
-       spin_unlock_irqrestore(&kc->job_lock, flags);
-
-       return job;
-}
-
-static void push(struct list_head *jobs, struct kcopyd_job *job)
-{
-       unsigned long flags;
-       struct dm_kcopyd_client *kc = job->kc;
-
-       spin_lock_irqsave(&kc->job_lock, flags);
-       list_add_tail(&job->list, jobs);
-       spin_unlock_irqrestore(&kc->job_lock, flags);
-}
-
-/*
- * These three functions process 1 item from the corresponding
- * job list.
- *
- * They return:
- * < 0: error
- *   0: success
- * > 0: can't process yet.
- */
-static int run_complete_job(struct kcopyd_job *job)
-{
-       void *context = job->context;
-       int read_err = job->read_err;
-       unsigned long write_err = job->write_err;
-       dm_kcopyd_notify_fn fn = job->fn;
-       struct dm_kcopyd_client *kc = job->kc;
-
-       kcopyd_put_pages(kc, job->pages);
-       mempool_free(job, kc->job_pool);
-       fn(read_err, write_err, context);
-
-       if (atomic_dec_and_test(&kc->nr_jobs))
-               wake_up(&kc->destroyq);
-
-       return 0;
-}
-
-static void complete_io(unsigned long error, void *context)
-{
-       struct kcopyd_job *job = (struct kcopyd_job *) context;
-       struct dm_kcopyd_client *kc = job->kc;
-
-       if (error) {
-               if (job->rw == WRITE)
-                       job->write_err |= error;
-               else
-                       job->read_err = 1;
-
-               if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
-                       push(&kc->complete_jobs, job);
-                       wake(kc);
-                       return;
-               }
-       }
-
-       if (job->rw == WRITE)
-               push(&kc->complete_jobs, job);
-
-       else {
-               job->rw = WRITE;
-               push(&kc->io_jobs, job);
-       }
-
-       wake(kc);
-}
-
-/*
- * Request io on as many buffer heads as we can currently get for
- * a particular job.
- */
-static int run_io_job(struct kcopyd_job *job)
-{
-       int r;
-       struct dm_io_request io_req = {
-               .bi_rw = job->rw,
-               .mem.type = DM_IO_PAGE_LIST,
-               .mem.ptr.pl = job->pages,
-               .mem.offset = job->offset,
-               .notify.fn = complete_io,
-               .notify.context = job,
-               .client = job->kc->io_client,
-       };
-
-       if (job->rw == READ)
-               r = dm_io(&io_req, 1, &job->source, NULL);
-       else
-               r = dm_io(&io_req, job->num_dests, job->dests, NULL);
-
-       return r;
-}
-
-static int run_pages_job(struct kcopyd_job *job)
-{
-       int r;
-
-       job->nr_pages = dm_div_up(job->dests[0].count + job->offset,
-                                 PAGE_SIZE >> 9);
-       r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages);
-       if (!r) {
-               /* this job is ready for io */
-               push(&job->kc->io_jobs, job);
-               return 0;
-       }
-
-       if (r == -ENOMEM)
-               /* can't complete now */
-               return 1;
-
-       return r;
-}
-
-/*
- * Run through a list for as long as possible.  Returns the count
- * of successful jobs.
- */
-static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
-                       int (*fn) (struct kcopyd_job *))
-{
-       struct kcopyd_job *job;
-       int r, count = 0;
-
-       while ((job = pop(jobs, kc))) {
-
-               r = fn(job);
-
-               if (r < 0) {
-                       /* error this rogue job */
-                       if (job->rw == WRITE)
-                               job->write_err = (unsigned long) -1L;
-                       else
-                               job->read_err = 1;
-                       push(&kc->complete_jobs, job);
-                       break;
-               }
-
-               if (r > 0) {
-                       /*
-                        * We couldn't service this job ATM, so
-                        * push this job back onto the list.
-                        */
-                       push(jobs, job);
-                       break;
-               }
-
-               count++;
-       }
-
-       return count;
-}
-
-/*
- * kcopyd does this every time it's woken up.
- */
-static void do_work(struct work_struct *work)
-{
-       struct dm_kcopyd_client *kc = container_of(work,
-                                       struct dm_kcopyd_client, kcopyd_work);
-
-       /*
-        * The order that these are called is *very* important.
-        * complete jobs can free some pages for pages jobs.
-        * Pages jobs when successful will jump onto the io jobs
-        * list.  io jobs call wake when they complete and it all
-        * starts again.
-        */
-       process_jobs(&kc->complete_jobs, kc, run_complete_job);
-       process_jobs(&kc->pages_jobs, kc, run_pages_job);
-       process_jobs(&kc->io_jobs, kc, run_io_job);
-}
-
-/*
- * If we are copying a small region we just dispatch a single job
- * to do the copy, otherwise the io has to be split up into many
- * jobs.
- */
-static void dispatch_job(struct kcopyd_job *job)
-{
-       struct dm_kcopyd_client *kc = job->kc;
-       atomic_inc(&kc->nr_jobs);
-       push(&kc->pages_jobs, job);
-       wake(kc);
-}
-
-#define SUB_JOB_SIZE 128
-static void segment_complete(int read_err, unsigned long write_err,
-                            void *context)
-{
-       /* FIXME: tidy this function */
-       sector_t progress = 0;
-       sector_t count = 0;
-       struct kcopyd_job *job = (struct kcopyd_job *) context;
-
-       mutex_lock(&job->lock);
-
-       /* update the error */
-       if (read_err)
-               job->read_err = 1;
-
-       if (write_err)
-               job->write_err |= write_err;
-
-       /*
-        * Only dispatch more work if there hasn't been an error.
-        */
-       if ((!job->read_err && !job->write_err) ||
-           test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
-               /* get the next chunk of work */
-               progress = job->progress;
-               count = job->source.count - progress;
-               if (count) {
-                       if (count > SUB_JOB_SIZE)
-                               count = SUB_JOB_SIZE;
-
-                       job->progress += count;
-               }
-       }
-       mutex_unlock(&job->lock);
-
-       if (count) {
-               int i;
-               struct kcopyd_job *sub_job = mempool_alloc(job->kc->job_pool,
-                                                          GFP_NOIO);
-
-               *sub_job = *job;
-               sub_job->source.sector += progress;
-               sub_job->source.count = count;
-
-               for (i = 0; i < job->num_dests; i++) {
-                       sub_job->dests[i].sector += progress;
-                       sub_job->dests[i].count = count;
-               }
-
-               sub_job->fn = segment_complete;
-               sub_job->context = job;
-               dispatch_job(sub_job);
-
-       } else if (atomic_dec_and_test(&job->sub_jobs)) {
-
-               /*
-                * To avoid a race we must keep the job around
-                * until after the notify function has completed.
-                * Otherwise the client may try and stop the job
-                * after we've completed.
-                */
-               job->fn(read_err, write_err, job->context);
-               mempool_free(job, job->kc->job_pool);
-       }
-}
-
-/*
- * Create some little jobs that will do the move between
- * them.
- */
-#define SPLIT_COUNT 8
-static void split_job(struct kcopyd_job *job)
-{
-       int i;
-
-       atomic_set(&job->sub_jobs, SPLIT_COUNT);
-       for (i = 0; i < SPLIT_COUNT; i++)
-               segment_complete(0, 0u, job);
-}
-
-int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
-                  unsigned int num_dests, struct dm_io_region *dests,
-                  unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
-{
-       struct kcopyd_job *job;
-
-       /*
-        * Allocate a new job.
-        */
-       job = mempool_alloc(kc->job_pool, GFP_NOIO);
-
-       /*
-        * set up for the read.
-        */
-       job->kc = kc;
-       job->flags = flags;
-       job->read_err = 0;
-       job->write_err = 0;
-       job->rw = READ;
-
-       job->source = *from;
-
-       job->num_dests = num_dests;
-       memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
-
-       job->offset = 0;
-       job->nr_pages = 0;
-       job->pages = NULL;
-
-       job->fn = fn;
-       job->context = context;
-
-       if (job->source.count < SUB_JOB_SIZE)
-               dispatch_job(job);
-
-       else {
-               mutex_init(&job->lock);
-               job->progress = 0;
-               split_job(job);
-       }
-
-       return 0;
-}
-EXPORT_SYMBOL(dm_kcopyd_copy);
-
-/*
- * Cancels a kcopyd job, eg. someone might be deactivating a
- * mirror.
- */
-#if 0
-int kcopyd_cancel(struct kcopyd_job *job, int block)
-{
-       /* FIXME: finish */
-       return -1;
-}
-#endif  /*  0  */
-
-/*-----------------------------------------------------------------
- * Client setup
- *---------------------------------------------------------------*/
-int dm_kcopyd_client_create(unsigned int nr_pages,
-                           struct dm_kcopyd_client **result)
-{
-       int r = -ENOMEM;
-       struct dm_kcopyd_client *kc;
-
-       kc = kmalloc(sizeof(*kc), GFP_KERNEL);
-       if (!kc)
-               return -ENOMEM;
-
-       spin_lock_init(&kc->lock);
-       spin_lock_init(&kc->job_lock);
-       INIT_LIST_HEAD(&kc->complete_jobs);
-       INIT_LIST_HEAD(&kc->io_jobs);
-       INIT_LIST_HEAD(&kc->pages_jobs);
-
-       kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
-       if (!kc->job_pool)
-               goto bad_slab;
-
-       INIT_WORK(&kc->kcopyd_work, do_work);
-       kc->kcopyd_wq = create_singlethread_workqueue("kcopyd");
-       if (!kc->kcopyd_wq)
-               goto bad_workqueue;
-
-       kc->pages = NULL;
-       kc->nr_pages = kc->nr_free_pages = 0;
-       r = client_alloc_pages(kc, nr_pages);
-       if (r)
-               goto bad_client_pages;
-
-       kc->io_client = dm_io_client_create(nr_pages);
-       if (IS_ERR(kc->io_client)) {
-               r = PTR_ERR(kc->io_client);
-               goto bad_io_client;
-       }
-
-       init_waitqueue_head(&kc->destroyq);
-       atomic_set(&kc->nr_jobs, 0);
-
-       *result = kc;
-       return 0;
-
-bad_io_client:
-       client_free_pages(kc);
-bad_client_pages:
-       destroy_workqueue(kc->kcopyd_wq);
-bad_workqueue:
-       mempool_destroy(kc->job_pool);
-bad_slab:
-       kfree(kc);
-
-       return r;
-}
-EXPORT_SYMBOL(dm_kcopyd_client_create);
-
-void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
-{
-       /* Wait for completion of all jobs submitted by this client. */
-       wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
-
-       BUG_ON(!list_empty(&kc->complete_jobs));
-       BUG_ON(!list_empty(&kc->io_jobs));
-       BUG_ON(!list_empty(&kc->pages_jobs));
-       destroy_workqueue(kc->kcopyd_wq);
-       dm_io_client_destroy(kc->io_client);
-       client_free_pages(kc);
-       mempool_destroy(kc->job_pool);
-       kfree(kc);
-}
-EXPORT_SYMBOL(dm_kcopyd_client_destroy);
diff --git a/drivers/md/kcopyd.h b/drivers/md/kcopyd.h
deleted file mode 100644 (file)
index d305784..0000000
+++ /dev/null
@@ -1,47 +0,0 @@
-/*
- * Copyright (C) 2001 - 2003 Sistina Software
- * Copyright (C) 2004 - 2008 Red Hat, Inc. All rights reserved.
- *
- * kcopyd provides a simple interface for copying an area of one
- * block-device to one or more other block-devices, either synchronous
- * or with an asynchronous completion notification.
- *
- * This file is released under the GPL.
- */
-
-#ifndef _LINUX_DM_KCOPYD_H
-#define _LINUX_DM_KCOPYD_H
-
-#ifdef __KERNEL__
-
-#include "dm-io.h"
-
-/* FIXME: make this configurable */
-#define DM_KCOPYD_MAX_REGIONS 8
-
-#define DM_KCOPYD_IGNORE_ERROR 1
-
-/*
- * To use kcopyd you must first create a dm_kcopyd_client object.
- */
-struct dm_kcopyd_client;
-int dm_kcopyd_client_create(unsigned num_pages,
-                           struct dm_kcopyd_client **result);
-void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc);
-
-/*
- * Submit a copy job to kcopyd.  This is built on top of the
- * previous three fns.
- *
- * read_err is a boolean,
- * write_err is a bitset, with 1 bit for each destination region
- */
-typedef void (*dm_kcopyd_notify_fn)(int read_err, unsigned long write_err,
-                                   void *context);
-
-int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
-                  unsigned num_dests, struct dm_io_region *dests,
-                  unsigned flags, dm_kcopyd_notify_fn fn, void *context);
-
-#endif /* __KERNEL__ */
-#endif /* _LINUX_DM_KCOPYD_H */