[linux-block.git] / fs / orangefs / orangefs-bufmap.c

/*
 * (C) 2001 Clemson University and The University of Chicago
 *
 * See COPYING in top-level directory.
 */
#include "protocol.h"
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"

DECLARE_WAIT_QUEUE_HEAD(orangefs_bufmap_init_waitq);

/* used to describe mapped buffers */
struct orangefs_bufmap_desc {
	void *uaddr;			/* user space address pointer */
	struct page **page_array;	/* array of mapped pages */
	int array_count;		/* size of above arrays */
	struct list_head list_link;
};

static struct orangefs_bufmap {
	atomic_t refcnt;

	int desc_size;
	int desc_shift;
	int desc_count;
	int total_size;
	int page_count;

	struct page **page_array;
	struct orangefs_bufmap_desc *desc_array;

	/* array to track usage of buffer descriptors */
	int *buffer_index_array;
	spinlock_t buffer_index_lock;

	/* array to track usage of buffer descriptors for readdir */
	int readdir_index_array[ORANGEFS_READDIR_DEFAULT_DESC_COUNT];
	spinlock_t readdir_index_lock;
} *__orangefs_bufmap;

static DEFINE_SPINLOCK(orangefs_bufmap_lock);

static void
orangefs_bufmap_unmap(struct orangefs_bufmap *bufmap)
{
	int i;

	for (i = 0; i < bufmap->page_count; i++)
		page_cache_release(bufmap->page_array[i]);
}

static void
orangefs_bufmap_free(struct orangefs_bufmap *bufmap)
{
	kfree(bufmap->page_array);
	kfree(bufmap->desc_array);
	kfree(bufmap->buffer_index_array);
	kfree(bufmap);
}

static struct orangefs_bufmap *orangefs_bufmap_ref(void)
{
	struct orangefs_bufmap *bufmap = NULL;

	spin_lock(&orangefs_bufmap_lock);
	if (__orangefs_bufmap) {
		bufmap = __orangefs_bufmap;
		atomic_inc(&bufmap->refcnt);
	}
	spin_unlock(&orangefs_bufmap_lock);
	return bufmap;
}

static void orangefs_bufmap_unref(struct orangefs_bufmap *bufmap)
{
	if (atomic_dec_and_lock(&bufmap->refcnt, &orangefs_bufmap_lock)) {
		__orangefs_bufmap = NULL;
		spin_unlock(&orangefs_bufmap_lock);

		orangefs_bufmap_unmap(bufmap);
		orangefs_bufmap_free(bufmap);
	}
}

/*
 * XXX: Can the size and shift change while the caller gives up the 
 * XXX: lock between calling this and doing something useful?
 */

int orangefs_bufmap_size_query(void)
{
	struct orangefs_bufmap *bufmap;
	int size = 0;
	bufmap = orangefs_bufmap_ref();
	if (bufmap) {
		size = bufmap->desc_size;
		orangefs_bufmap_unref(bufmap);
	}
	return size;
}

int orangefs_bufmap_shift_query(void)
{
	struct orangefs_bufmap *bufmap;
	int shift = 0;
	bufmap = orangefs_bufmap_ref();
	if (bufmap) {
		shift = bufmap->desc_shift;
		orangefs_bufmap_unref(bufmap);
	}
	return shift;
}

static DECLARE_WAIT_QUEUE_HEAD(bufmap_waitq);
static DECLARE_WAIT_QUEUE_HEAD(readdir_waitq);

/*
 * get_bufmap_init
 *
 * If bufmap_init is 1, then the shared memory system, including the
 * buffer_index_array, is available.  Otherwise, it is not.
 *
 * returns the value of bufmap_init
 */
int get_bufmap_init(void)
{
	return __orangefs_bufmap ? 1 : 0;
}


static struct orangefs_bufmap *
orangefs_bufmap_alloc(struct ORANGEFS_dev_map_desc *user_desc)
{
	struct orangefs_bufmap *bufmap;

	bufmap = kzalloc(sizeof(*bufmap), GFP_KERNEL);
	if (!bufmap)
		goto out;

	atomic_set(&bufmap->refcnt, 1);
	bufmap->total_size = user_desc->total_size;
	bufmap->desc_count = user_desc->count;
	bufmap->desc_size = user_desc->size;
	bufmap->desc_shift = ilog2(bufmap->desc_size);

	spin_lock_init(&bufmap->buffer_index_lock);
	bufmap->buffer_index_array =
		kcalloc(bufmap->desc_count, sizeof(int), GFP_KERNEL);
	if (!bufmap->buffer_index_array) {
		gossip_err("orangefs: could not allocate %d buffer indices\n",
				bufmap->desc_count);
		goto out_free_bufmap;
	}
	spin_lock_init(&bufmap->readdir_index_lock);

	bufmap->desc_array =
		kcalloc(bufmap->desc_count, sizeof(struct orangefs_bufmap_desc),
			GFP_KERNEL);
	if (!bufmap->desc_array) {
		gossip_err("orangefs: could not allocate %d descriptors\n",
				bufmap->desc_count);
		goto out_free_index_array;
	}

	bufmap->page_count = bufmap->total_size / PAGE_SIZE;

	/* allocate storage to track our page mappings */
	bufmap->page_array =
		kcalloc(bufmap->page_count, sizeof(struct page *), GFP_KERNEL);
	if (!bufmap->page_array)
		goto out_free_desc_array;

	return bufmap;

out_free_desc_array:
	kfree(bufmap->desc_array);
out_free_index_array:
	kfree(bufmap->buffer_index_array);
out_free_bufmap:
	kfree(bufmap);
out:
	return NULL;
}

static int
orangefs_bufmap_map(struct orangefs_bufmap *bufmap,
		struct ORANGEFS_dev_map_desc *user_desc)
{
	int pages_per_desc = bufmap->desc_size / PAGE_SIZE;
	int offset = 0, ret, i;

	/* map the pages */
	ret = get_user_pages_fast((unsigned long)user_desc->ptr,
			     bufmap->page_count, 1, bufmap->page_array);

	if (ret < 0)
		return ret;

	if (ret != bufmap->page_count) {
		gossip_err("orangefs error: asked for %d pages, only got %d.\n",
				bufmap->page_count, ret);

		for (i = 0; i < ret; i++) {
			SetPageError(bufmap->page_array[i]);
			page_cache_release(bufmap->page_array[i]);
		}
		return -ENOMEM;
	}

	/*
	 * ideally we want to get kernel space pointers for each page, but
	 * we can't kmap that many pages at once if highmem is being used.
	 * so instead, we just kmap/kunmap the page address each time the
	 * kaddr is needed.
	 */
	for (i = 0; i < bufmap->page_count; i++)
		flush_dcache_page(bufmap->page_array[i]);

	/* build a list of available descriptors */
	for (offset = 0, i = 0; i < bufmap->desc_count; i++) {
		bufmap->desc_array[i].page_array = &bufmap->page_array[offset];
		bufmap->desc_array[i].array_count = pages_per_desc;
		bufmap->desc_array[i].uaddr =
		    (user_desc->ptr + (i * pages_per_desc * PAGE_SIZE));
		offset += pages_per_desc;
	}

	return 0;
}

/*
 * orangefs_bufmap_initialize()
 *
 * initializes the mapped buffer interface
 *
 * returns 0 on success, -errno on failure
 */
int orangefs_bufmap_initialize(struct ORANGEFS_dev_map_desc *user_desc)
{
	struct orangefs_bufmap *bufmap;
	int ret = -EINVAL;

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "orangefs_bufmap_initialize: called (ptr ("
		     "%p) sz (%d) cnt(%d).\n",
		     user_desc->ptr,
		     user_desc->size,
		     user_desc->count);

	/*
	 * sanity check alignment and size of buffer that caller wants to
	 * work with
	 */
	if (PAGE_ALIGN((unsigned long)user_desc->ptr) !=
	    (unsigned long)user_desc->ptr) {
		gossip_err("orangefs error: memory alignment (front). %p\n",
			   user_desc->ptr);
		goto out;
	}

	if (PAGE_ALIGN(((unsigned long)user_desc->ptr + user_desc->total_size))
	    != (unsigned long)(user_desc->ptr + user_desc->total_size)) {
		gossip_err("orangefs error: memory alignment (back).(%p + %d)\n",
			   user_desc->ptr,
			   user_desc->total_size);
		goto out;
	}

	if (user_desc->total_size != (user_desc->size * user_desc->count)) {
		gossip_err("orangefs error: user provided an oddly sized buffer: (%d, %d, %d)\n",
			   user_desc->total_size,
			   user_desc->size,
			   user_desc->count);
		goto out;
	}

	if ((user_desc->size % PAGE_SIZE) != 0) {
		gossip_err("orangefs error: bufmap size not page size divisible (%d).\n",
			   user_desc->size);
		goto out;
	}

	ret = -ENOMEM;
	bufmap = orangefs_bufmap_alloc(user_desc);
	if (!bufmap)
		goto out;

	ret = orangefs_bufmap_map(bufmap, user_desc);
	if (ret)
		goto out_free_bufmap;


	spin_lock(&orangefs_bufmap_lock);
	if (__orangefs_bufmap) {
		spin_unlock(&orangefs_bufmap_lock);
		gossip_err("orangefs: error: bufmap already initialized.\n");
		ret = -EALREADY;
		goto out_unmap_bufmap;
	}
	__orangefs_bufmap = bufmap;
	spin_unlock(&orangefs_bufmap_lock);

	/*
	 * If there are operations in orangefs_bufmap_init_waitq, wake them up.
	 * This scenario occurs when the client-core is restarted and I/O
	 * requests in the in-progress or waiting tables are restarted.  I/O
	 * requests cannot be restarted until the shared memory system is
	 * completely re-initialized, so we put the I/O requests in this
	 * waitq until initialization has completed.  NOTE:  the I/O requests
	 * are also on a timer, so they don't wait forever just in case the
	 * client-core doesn't come back up.
	 */
	wake_up_interruptible(&orangefs_bufmap_init_waitq);

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "orangefs_bufmap_initialize: exiting normally\n");
	return 0;

out_unmap_bufmap:
	orangefs_bufmap_unmap(bufmap);
out_free_bufmap:
	orangefs_bufmap_free(bufmap);
out:
	return ret;
}

/*
 * orangefs_bufmap_finalize()
 *
 * shuts down the mapped buffer interface and releases any resources
 * associated with it
 *
 * no return value
 */
void orangefs_bufmap_finalize(void)
{
	gossip_debug(GOSSIP_BUFMAP_DEBUG, "orangefs_bufmap_finalize: called\n");
	BUG_ON(!__orangefs_bufmap);
	orangefs_bufmap_unref(__orangefs_bufmap);
	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "orangefs_bufmap_finalize: exiting normally\n");
}

struct slot_args {
	int slot_count;
	int *slot_array;
	spinlock_t *slot_lock;
	wait_queue_head_t *slot_wq;
};

static int wait_for_a_slot(struct slot_args *slargs, int *buffer_index)
{
	int ret = -1;
	int i = 0;
	DEFINE_WAIT(wait_entry);

	while (1) {
		/*
		 * check for available desc, slot_lock is the appropriate
		 * index_lock
		 */
		spin_lock(slargs->slot_lock);
		prepare_to_wait_exclusive(slargs->slot_wq,
					  &wait_entry,
					  TASK_INTERRUPTIBLE);
		for (i = 0; i < slargs->slot_count; i++)
			if (slargs->slot_array[i] == 0) {
				slargs->slot_array[i] = 1;
				*buffer_index = i;
				ret = 0;
				break;
			}
		spin_unlock(slargs->slot_lock);

		/* if we acquired a buffer, then break out of while */
		if (ret == 0)
			break;

		if (!signal_pending(current)) {
			int timeout =
			    MSECS_TO_JIFFIES(1000 * slot_timeout_secs);
			gossip_debug(GOSSIP_BUFMAP_DEBUG,
				     "[BUFMAP]: waiting %d "
				     "seconds for a slot\n",
				     slot_timeout_secs);
			if (!schedule_timeout(timeout)) {
				gossip_debug(GOSSIP_BUFMAP_DEBUG,
					     "*** wait_for_a_slot timed out\n");
				ret = -ETIMEDOUT;
				break;
			}
			gossip_debug(GOSSIP_BUFMAP_DEBUG,
			  "[BUFMAP]: woken up by a slot becoming available.\n");
			continue;
		}

		gossip_debug(GOSSIP_BUFMAP_DEBUG, "orangefs: %s interrupted.\n",
			     __func__);
		ret = -EINTR;
		break;
	}

	spin_lock(slargs->slot_lock);
	finish_wait(slargs->slot_wq, &wait_entry);
	spin_unlock(slargs->slot_lock);
	return ret;
}

static void put_back_slot(struct slot_args *slargs, int buffer_index)
{
	/* slot_lock is the appropriate index_lock */
	spin_lock(slargs->slot_lock);
	if (buffer_index < 0 || buffer_index >= slargs->slot_count) {
		spin_unlock(slargs->slot_lock);
		return;
	}

	/* put the desc back on the queue */
	slargs->slot_array[buffer_index] = 0;
	spin_unlock(slargs->slot_lock);

	/* wake up anyone who may be sleeping on the queue */
	wake_up_interruptible(slargs->slot_wq);
}

/*
 * orangefs_bufmap_get()
 *
 * gets a free mapped buffer descriptor, will sleep until one becomes
 * available if necessary
 *
 * returns 0 on success, -errno on failure
 */
int orangefs_bufmap_get(struct orangefs_bufmap **mapp, int *buffer_index)
{
	struct orangefs_bufmap *bufmap = orangefs_bufmap_ref();
	struct slot_args slargs;
	int ret;

	if (!bufmap) {
		gossip_err("orangefs: please confirm that pvfs2-client daemon is running.\n");
		return -EIO;
	}

	slargs.slot_count = bufmap->desc_count;
	slargs.slot_array = bufmap->buffer_index_array;
	slargs.slot_lock = &bufmap->buffer_index_lock;
	slargs.slot_wq = &bufmap_waitq;
	ret = wait_for_a_slot(&slargs, buffer_index);
	if (ret)
		orangefs_bufmap_unref(bufmap);
	*mapp = bufmap;
	return ret;
}

/*
 * orangefs_bufmap_put()
 *
 * returns a mapped buffer descriptor to the collection
 *
 * no return value
 */
void orangefs_bufmap_put(struct orangefs_bufmap *bufmap, int buffer_index)
{
	struct slot_args slargs;

	slargs.slot_count = bufmap->desc_count;
	slargs.slot_array = bufmap->buffer_index_array;
	slargs.slot_lock = &bufmap->buffer_index_lock;
	slargs.slot_wq = &bufmap_waitq;
	put_back_slot(&slargs, buffer_index);
	orangefs_bufmap_unref(bufmap);
}

/*
 * readdir_index_get()
 *
 * gets a free descriptor, will sleep until one becomes
 * available if necessary.
 * Although the readdir buffers are not mapped into kernel space
 * we could do that at a later point of time. Regardless, these
 * indices are used by the client-core.
 *
 * returns 0 on success, -errno on failure
 */
int readdir_index_get(struct orangefs_bufmap **mapp, int *buffer_index)
{
	struct orangefs_bufmap *bufmap = orangefs_bufmap_ref();
	struct slot_args slargs;
	int ret;

	if (!bufmap) {
		gossip_err("orangefs: please confirm that pvfs2-client daemon is running.\n");
		return -EIO;
	}

	slargs.slot_count = ORANGEFS_READDIR_DEFAULT_DESC_COUNT;
	slargs.slot_array = bufmap->readdir_index_array;
	slargs.slot_lock = &bufmap->readdir_index_lock;
	slargs.slot_wq = &readdir_waitq;
	ret = wait_for_a_slot(&slargs, buffer_index);
	if (ret)
		orangefs_bufmap_unref(bufmap);
	*mapp = bufmap;
	return ret;
}

void readdir_index_put(struct orangefs_bufmap *bufmap, int buffer_index)
{
	struct slot_args slargs;

	slargs.slot_count = ORANGEFS_READDIR_DEFAULT_DESC_COUNT;
	slargs.slot_array = bufmap->readdir_index_array;
	slargs.slot_lock = &bufmap->readdir_index_lock;
	slargs.slot_wq = &readdir_waitq;
	put_back_slot(&slargs, buffer_index);
	orangefs_bufmap_unref(bufmap);
}

/*
 * we've been handed an iovec, we need to copy it to 
 * the shared memory descriptor at "buffer_index".
 */
int orangefs_bufmap_copy_from_iovec(struct orangefs_bufmap *bufmap,
				struct iov_iter *iter,
				int buffer_index,
				size_t size)
{
	struct orangefs_bufmap_desc *to = &bufmap->desc_array[buffer_index];
	int i;

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "%s: buffer_index:%d: size:%zu:\n",
		     __func__, buffer_index, size);


	for (i = 0; size; i++) {
		struct page *page = to->page_array[i];
		size_t n = size;
		if (n > PAGE_SIZE)
			n = PAGE_SIZE;
		n = copy_page_from_iter(page, 0, n, iter);
		if (!n)
			return -EFAULT;
		size -= n;
	}
	return 0;

}

/*
 * we've been handed an iovec, we need to fill it from
 * the shared memory descriptor at "buffer_index".
 */
int orangefs_bufmap_copy_to_iovec(struct orangefs_bufmap *bufmap,
				    struct iov_iter *iter,
				    int buffer_index,
				    size_t size)
{
	struct orangefs_bufmap_desc *from = &bufmap->desc_array[buffer_index];
	int i;

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "%s: buffer_index:%d: size:%zu:\n",
		     __func__, buffer_index, size);


	for (i = 0; size; i++) {
		struct page *page = from->page_array[i];
		size_t n = size;
		if (n > PAGE_SIZE)
			n = PAGE_SIZE;
		n = copy_page_to_iter(page, 0, n, iter);
		if (!n)
			return -EFAULT;
		size -= n;
	}
	return 0;
}
Commit	Line	Data
274dcf55 MM	1	/*
	2	* (C) 2001 Clemson University and The University of Chicago
	3	*
	4	* See COPYING in top-level directory.
	5	*/
	6	#include "protocol.h"
575e9461 MM	7	#include "orangefs-kernel.h"
575e9461 MM	8	#include "orangefs-bufmap.h"
274dcf55	9
8bb8aefd	10	DECLARE_WAIT_QUEUE_HEAD(orangefs_bufmap_init_waitq);
274dcf55	11
bf89f584 MB	12	/* used to describe mapped buffers */
	13	struct orangefs_bufmap_desc {
	14	void uaddr; / user space address pointer */
	15	struct page *page_array; / array of mapped pages */
	16	int array_count; /* size of above arrays */
	17	struct list_head list_link;
	18	};
	19
8bb8aefd	20	static struct orangefs_bufmap {
274dcf55 MM	21	atomic_t refcnt;
	22
	23	int desc_size;
	24	int desc_shift;
	25	int desc_count;
	26	int total_size;
	27	int page_count;
	28
	29	struct page **page_array;
8bb8aefd	30	struct orangefs_bufmap_desc *desc_array;
274dcf55 MM	31
	32	/* array to track usage of buffer descriptors */
	33	int *buffer_index_array;
	34	spinlock_t buffer_index_lock;
	35
	36	/* array to track usage of buffer descriptors for readdir */
8bb8aefd	37	int readdir_index_array[ORANGEFS_READDIR_DEFAULT_DESC_COUNT];
274dcf55	38	spinlock_t readdir_index_lock;
8bb8aefd	39	} *__orangefs_bufmap;
274dcf55	40
8bb8aefd	41	static DEFINE_SPINLOCK(orangefs_bufmap_lock);
274dcf55 MM	42
274dcf55 MM	43	static void
8bb8aefd	44	orangefs_bufmap_unmap(struct orangefs_bufmap *bufmap)
274dcf55 MM	45	{
	46	int i;
	47
	48	for (i = 0; i < bufmap->page_count; i++)
	49	page_cache_release(bufmap->page_array[i]);
	50	}
	51
	52	static void
8bb8aefd	53	orangefs_bufmap_free(struct orangefs_bufmap *bufmap)
274dcf55 MM	54	{
	55	kfree(bufmap->page_array);
	56	kfree(bufmap->desc_array);
	57	kfree(bufmap->buffer_index_array);
	58	kfree(bufmap);
	59	}
	60
bf89f584	61	static struct orangefs_bufmap *orangefs_bufmap_ref(void)
274dcf55	62	{
8bb8aefd	63	struct orangefs_bufmap *bufmap = NULL;
274dcf55	64
8bb8aefd YL	65	spin_lock(&orangefs_bufmap_lock);
	66	if (__orangefs_bufmap) {
	67	bufmap = __orangefs_bufmap;
274dcf55 MM	68	atomic_inc(&bufmap->refcnt);
274dcf55 MM	69	}
8bb8aefd	70	spin_unlock(&orangefs_bufmap_lock);
274dcf55 MM	71	return bufmap;
	72	}
	73
bf89f584	74	static void orangefs_bufmap_unref(struct orangefs_bufmap *bufmap)
274dcf55	75	{
8bb8aefd YL	76	if (atomic_dec_and_lock(&bufmap->refcnt, &orangefs_bufmap_lock)) {
	77	__orangefs_bufmap = NULL;
	78	spin_unlock(&orangefs_bufmap_lock);
274dcf55	79
8bb8aefd YL	80	orangefs_bufmap_unmap(bufmap);
8bb8aefd YL	81	orangefs_bufmap_free(bufmap);
274dcf55 MM	82	}
	83	}
	84
b09d10df MB	85	/*
	86	* XXX: Can the size and shift change while the caller gives up the
	87	* XXX: lock between calling this and doing something useful?
	88	*/
	89
765a75b3	90	int orangefs_bufmap_size_query(void)
274dcf55	91	{
b09d10df MB	92	struct orangefs_bufmap *bufmap;
	93	int size = 0;
	94	bufmap = orangefs_bufmap_ref();
	95	if (bufmap) {
	96	size = bufmap->desc_size;
	97	orangefs_bufmap_unref(bufmap);
	98	}
274dcf55 MM	99	return size;
	100	}
	101
765a75b3	102	int orangefs_bufmap_shift_query(void)
274dcf55	103	{
b09d10df MB	104	struct orangefs_bufmap *bufmap;
	105	int shift = 0;
	106	bufmap = orangefs_bufmap_ref();
	107	if (bufmap) {
	108	shift = bufmap->desc_shift;
	109	orangefs_bufmap_unref(bufmap);
	110	}
274dcf55 MM	111	return shift;
	112	}
	113
	114	static DECLARE_WAIT_QUEUE_HEAD(bufmap_waitq);
	115	static DECLARE_WAIT_QUEUE_HEAD(readdir_waitq);
	116
	117	/*
	118	* get_bufmap_init
	119	*
	120	* If bufmap_init is 1, then the shared memory system, including the
	121	* buffer_index_array, is available. Otherwise, it is not.
	122	*
	123	* returns the value of bufmap_init
	124	*/
	125	int get_bufmap_init(void)
	126	{
8bb8aefd	127	return __orangefs_bufmap ? 1 : 0;
274dcf55 MM	128	}
	129
	130
8bb8aefd YL	131	static struct orangefs_bufmap *
8bb8aefd YL	132	orangefs_bufmap_alloc(struct ORANGEFS_dev_map_desc *user_desc)
274dcf55	133	{
8bb8aefd	134	struct orangefs_bufmap *bufmap;
274dcf55 MM	135
	136	bufmap = kzalloc(sizeof(*bufmap), GFP_KERNEL);
	137	if (!bufmap)
	138	goto out;
	139
	140	atomic_set(&bufmap->refcnt, 1);
	141	bufmap->total_size = user_desc->total_size;
	142	bufmap->desc_count = user_desc->count;
	143	bufmap->desc_size = user_desc->size;
	144	bufmap->desc_shift = ilog2(bufmap->desc_size);
	145
	146	spin_lock_init(&bufmap->buffer_index_lock);
	147	bufmap->buffer_index_array =
	148	kcalloc(bufmap->desc_count, sizeof(int), GFP_KERNEL);
	149	if (!bufmap->buffer_index_array) {
8bb8aefd	150	gossip_err("orangefs: could not allocate %d buffer indices\n",
274dcf55 MM	151	bufmap->desc_count);
	152	goto out_free_bufmap;
	153	}
	154	spin_lock_init(&bufmap->readdir_index_lock);
	155
	156	bufmap->desc_array =
8bb8aefd	157	kcalloc(bufmap->desc_count, sizeof(struct orangefs_bufmap_desc),
274dcf55 MM	158	GFP_KERNEL);
274dcf55 MM	159	if (!bufmap->desc_array) {
8bb8aefd	160	gossip_err("orangefs: could not allocate %d descriptors\n",
274dcf55 MM	161	bufmap->desc_count);
	162	goto out_free_index_array;
	163	}
	164
	165	bufmap->page_count = bufmap->total_size / PAGE_SIZE;
	166
	167	/* allocate storage to track our page mappings */
	168	bufmap->page_array =
	169	kcalloc(bufmap->page_count, sizeof(struct page *), GFP_KERNEL);
	170	if (!bufmap->page_array)
	171	goto out_free_desc_array;
	172
	173	return bufmap;
	174
	175	out_free_desc_array:
	176	kfree(bufmap->desc_array);
	177	out_free_index_array:
	178	kfree(bufmap->buffer_index_array);
	179	out_free_bufmap:
	180	kfree(bufmap);
	181	out:
	182	return NULL;
	183	}
	184
	185	static int
8bb8aefd YL	186	orangefs_bufmap_map(struct orangefs_bufmap *bufmap,
8bb8aefd YL	187	struct ORANGEFS_dev_map_desc *user_desc)
274dcf55 MM	188	{
	189	int pages_per_desc = bufmap->desc_size / PAGE_SIZE;
	190	int offset = 0, ret, i;
	191
	192	/* map the pages */
16742f2d AV	193	ret = get_user_pages_fast((unsigned long)user_desc->ptr,
16742f2d AV	194	bufmap->page_count, 1, bufmap->page_array);
274dcf55 MM	195
	196	if (ret < 0)
	197	return ret;
	198
	199	if (ret != bufmap->page_count) {
8bb8aefd	200	gossip_err("orangefs error: asked for %d pages, only got %d.\n",
274dcf55 MM	201	bufmap->page_count, ret);
	202
	203	for (i = 0; i < ret; i++) {
	204	SetPageError(bufmap->page_array[i]);
	205	page_cache_release(bufmap->page_array[i]);
	206	}
	207	return -ENOMEM;
	208	}
	209
	210	/*
	211	* ideally we want to get kernel space pointers for each page, but
	212	* we can't kmap that many pages at once if highmem is being used.
	213	* so instead, we just kmap/kunmap the page address each time the
	214	* kaddr is needed.
	215	*/
	216	for (i = 0; i < bufmap->page_count; i++)
	217	flush_dcache_page(bufmap->page_array[i]);
	218
	219	/* build a list of available descriptors */
	220	for (offset = 0, i = 0; i < bufmap->desc_count; i++) {
	221	bufmap->desc_array[i].page_array = &bufmap->page_array[offset];
	222	bufmap->desc_array[i].array_count = pages_per_desc;
	223	bufmap->desc_array[i].uaddr =
	224	(user_desc->ptr + (i * pages_per_desc * PAGE_SIZE));
	225	offset += pages_per_desc;
	226	}
	227
	228	return 0;
	229	}
	230
	231	/*
8bb8aefd	232	* orangefs_bufmap_initialize()
274dcf55 MM	233	*
	234	* initializes the mapped buffer interface
	235	*
	236	* returns 0 on success, -errno on failure
	237	*/
8bb8aefd	238	int orangefs_bufmap_initialize(struct ORANGEFS_dev_map_desc *user_desc)
274dcf55	239	{
8bb8aefd	240	struct orangefs_bufmap *bufmap;
274dcf55 MM	241	int ret = -EINVAL;
	242
	243	gossip_debug(GOSSIP_BUFMAP_DEBUG,
8bb8aefd	244	"orangefs_bufmap_initialize: called (ptr ("
274dcf55 MM	245	"%p) sz (%d) cnt(%d).\n",
	246	user_desc->ptr,
	247	user_desc->size,
	248	user_desc->count);
	249
	250	/*
	251	* sanity check alignment and size of buffer that caller wants to
	252	* work with
	253	*/
	254	if (PAGE_ALIGN((unsigned long)user_desc->ptr) !=
	255	(unsigned long)user_desc->ptr) {
8bb8aefd	256	gossip_err("orangefs error: memory alignment (front). %p\n",
274dcf55 MM	257	user_desc->ptr);
	258	goto out;
	259	}
	260
	261	if (PAGE_ALIGN(((unsigned long)user_desc->ptr + user_desc->total_size))
	262	!= (unsigned long)(user_desc->ptr + user_desc->total_size)) {
8bb8aefd	263	gossip_err("orangefs error: memory alignment (back).(%p + %d)\n",
274dcf55 MM	264	user_desc->ptr,
	265	user_desc->total_size);
	266	goto out;
	267	}
	268
	269	if (user_desc->total_size != (user_desc->size * user_desc->count)) {
8bb8aefd	270	gossip_err("orangefs error: user provided an oddly sized buffer: (%d, %d, %d)\n",
274dcf55 MM	271	user_desc->total_size,
	272	user_desc->size,
	273	user_desc->count);
	274	goto out;
	275	}
	276
	277	if ((user_desc->size % PAGE_SIZE) != 0) {
8bb8aefd	278	gossip_err("orangefs error: bufmap size not page size divisible (%d).\n",
274dcf55 MM	279	user_desc->size);
	280	goto out;
	281	}
	282
	283	ret = -ENOMEM;
8bb8aefd	284	bufmap = orangefs_bufmap_alloc(user_desc);
274dcf55 MM	285	if (!bufmap)
	286	goto out;
	287
8bb8aefd	288	ret = orangefs_bufmap_map(bufmap, user_desc);
274dcf55 MM	289	if (ret)
	290	goto out_free_bufmap;
	291
	292
8bb8aefd YL	293	spin_lock(&orangefs_bufmap_lock);
	294	if (__orangefs_bufmap) {
	295	spin_unlock(&orangefs_bufmap_lock);
	296	gossip_err("orangefs: error: bufmap already initialized.\n");
274dcf55 MM	297	ret = -EALREADY;
	298	goto out_unmap_bufmap;
	299	}
8bb8aefd YL	300	__orangefs_bufmap = bufmap;
8bb8aefd YL	301	spin_unlock(&orangefs_bufmap_lock);
274dcf55 MM	302
274dcf55 MM	303	/*
8bb8aefd	304	* If there are operations in orangefs_bufmap_init_waitq, wake them up.
274dcf55 MM	305	* This scenario occurs when the client-core is restarted and I/O
	306	* requests in the in-progress or waiting tables are restarted. I/O
	307	* requests cannot be restarted until the shared memory system is
	308	* completely re-initialized, so we put the I/O requests in this
	309	* waitq until initialization has completed. NOTE: the I/O requests
	310	* are also on a timer, so they don't wait forever just in case the
	311	* client-core doesn't come back up.
	312	*/
8bb8aefd	313	wake_up_interruptible(&orangefs_bufmap_init_waitq);
274dcf55 MM	314
274dcf55 MM	315	gossip_debug(GOSSIP_BUFMAP_DEBUG,
8bb8aefd	316	"orangefs_bufmap_initialize: exiting normally\n");
274dcf55 MM	317	return 0;
	318
	319	out_unmap_bufmap:
8bb8aefd	320	orangefs_bufmap_unmap(bufmap);
274dcf55	321	out_free_bufmap:
8bb8aefd	322	orangefs_bufmap_free(bufmap);
274dcf55 MM	323	out:
	324	return ret;
	325	}
	326
	327	/*
8bb8aefd	328	* orangefs_bufmap_finalize()
274dcf55 MM	329	*
	330	* shuts down the mapped buffer interface and releases any resources
	331	* associated with it
	332	*
	333	* no return value
	334	*/
8bb8aefd	335	void orangefs_bufmap_finalize(void)
274dcf55	336	{
8bb8aefd YL	337	gossip_debug(GOSSIP_BUFMAP_DEBUG, "orangefs_bufmap_finalize: called\n");
	338	BUG_ON(!__orangefs_bufmap);
	339	orangefs_bufmap_unref(__orangefs_bufmap);
274dcf55	340	gossip_debug(GOSSIP_BUFMAP_DEBUG,
8bb8aefd	341	"orangefs_bufmap_finalize: exiting normally\n");
274dcf55 MM	342	}
	343
	344	struct slot_args {
	345	int slot_count;
	346	int *slot_array;
	347	spinlock_t *slot_lock;
	348	wait_queue_head_t *slot_wq;
	349	};
	350
	351	static int wait_for_a_slot(struct slot_args slargs, int buffer_index)
	352	{
	353	int ret = -1;
	354	int i = 0;
ce6c414e	355	DEFINE_WAIT(wait_entry);
274dcf55 MM	356
274dcf55 MM	357	while (1) {
274dcf55 MM	358	/*
	359	* check for available desc, slot_lock is the appropriate
	360	* index_lock
	361	*/
	362	spin_lock(slargs->slot_lock);
ce6c414e MM	363	prepare_to_wait_exclusive(slargs->slot_wq,
	364	&wait_entry,
	365	TASK_INTERRUPTIBLE);
274dcf55 MM	366	for (i = 0; i < slargs->slot_count; i++)
	367	if (slargs->slot_array[i] == 0) {
	368	slargs->slot_array[i] = 1;
	369	*buffer_index = i;
	370	ret = 0;
	371	break;
	372	}
	373	spin_unlock(slargs->slot_lock);
	374
	375	/* if we acquired a buffer, then break out of while */
	376	if (ret == 0)
	377	break;
	378
	379	if (!signal_pending(current)) {
	380	int timeout =
	381	MSECS_TO_JIFFIES(1000 * slot_timeout_secs);
	382	gossip_debug(GOSSIP_BUFMAP_DEBUG,
	383	"[BUFMAP]: waiting %d "
	384	"seconds for a slot\n",
	385	slot_timeout_secs);
	386	if (!schedule_timeout(timeout)) {
	387	gossip_debug(GOSSIP_BUFMAP_DEBUG,
	388	"*** wait_for_a_slot timed out\n");
	389	ret = -ETIMEDOUT;
	390	break;
	391	}
	392	gossip_debug(GOSSIP_BUFMAP_DEBUG,
	393	"[BUFMAP]: woken up by a slot becoming available.\n");
	394	continue;
	395	}
	396
8bb8aefd	397	gossip_debug(GOSSIP_BUFMAP_DEBUG, "orangefs: %s interrupted.\n",
274dcf55 MM	398	__func__);
	399	ret = -EINTR;
	400	break;
	401	}
	402
ce6c414e MM	403	spin_lock(slargs->slot_lock);
	404	finish_wait(slargs->slot_wq, &wait_entry);
	405	spin_unlock(slargs->slot_lock);
274dcf55 MM	406	return ret;
	407	}
	408
	409	static void put_back_slot(struct slot_args *slargs, int buffer_index)
	410	{
	411	/* slot_lock is the appropriate index_lock */
	412	spin_lock(slargs->slot_lock);
	413	if (buffer_index < 0 \|\| buffer_index >= slargs->slot_count) {
	414	spin_unlock(slargs->slot_lock);
	415	return;
	416	}
	417
	418	/* put the desc back on the queue */
	419	slargs->slot_array[buffer_index] = 0;
	420	spin_unlock(slargs->slot_lock);
	421
	422	/* wake up anyone who may be sleeping on the queue */
	423	wake_up_interruptible(slargs->slot_wq);
	424	}
	425
	426	/*
8bb8aefd	427	* orangefs_bufmap_get()
274dcf55 MM	428	*
	429	* gets a free mapped buffer descriptor, will sleep until one becomes
	430	* available if necessary
	431	*
	432	* returns 0 on success, -errno on failure
	433	*/
8bb8aefd	434	int orangefs_bufmap_get(struct orangefs_bufmap *mapp, int buffer_index)
274dcf55	435	{
8bb8aefd	436	struct orangefs_bufmap *bufmap = orangefs_bufmap_ref();
274dcf55 MM	437	struct slot_args slargs;
	438	int ret;
	439
	440	if (!bufmap) {
8bb8aefd	441	gossip_err("orangefs: please confirm that pvfs2-client daemon is running.\n");
274dcf55 MM	442	return -EIO;
	443	}
	444
	445	slargs.slot_count = bufmap->desc_count;
	446	slargs.slot_array = bufmap->buffer_index_array;
	447	slargs.slot_lock = &bufmap->buffer_index_lock;
	448	slargs.slot_wq = &bufmap_waitq;
	449	ret = wait_for_a_slot(&slargs, buffer_index);
	450	if (ret)
8bb8aefd	451	orangefs_bufmap_unref(bufmap);
274dcf55 MM	452	*mapp = bufmap;
	453	return ret;
	454	}
	455
	456	/*
8bb8aefd	457	* orangefs_bufmap_put()
274dcf55 MM	458	*
	459	* returns a mapped buffer descriptor to the collection
	460	*
	461	* no return value
	462	*/
8bb8aefd	463	void orangefs_bufmap_put(struct orangefs_bufmap *bufmap, int buffer_index)
274dcf55 MM	464	{
	465	struct slot_args slargs;
	466
	467	slargs.slot_count = bufmap->desc_count;
	468	slargs.slot_array = bufmap->buffer_index_array;
	469	slargs.slot_lock = &bufmap->buffer_index_lock;
	470	slargs.slot_wq = &bufmap_waitq;
	471	put_back_slot(&slargs, buffer_index);
8bb8aefd	472	orangefs_bufmap_unref(bufmap);
274dcf55 MM	473	}
	474
	475	/*
	476	* readdir_index_get()
	477	*
	478	* gets a free descriptor, will sleep until one becomes
	479	* available if necessary.
	480	* Although the readdir buffers are not mapped into kernel space
	481	* we could do that at a later point of time. Regardless, these
	482	* indices are used by the client-core.
	483	*
	484	* returns 0 on success, -errno on failure
	485	*/
8bb8aefd	486	int readdir_index_get(struct orangefs_bufmap *mapp, int buffer_index)
274dcf55	487	{
8bb8aefd	488	struct orangefs_bufmap *bufmap = orangefs_bufmap_ref();
274dcf55 MM	489	struct slot_args slargs;
	490	int ret;
	491
	492	if (!bufmap) {
8bb8aefd	493	gossip_err("orangefs: please confirm that pvfs2-client daemon is running.\n");
274dcf55 MM	494	return -EIO;
	495	}
	496
8bb8aefd	497	slargs.slot_count = ORANGEFS_READDIR_DEFAULT_DESC_COUNT;
274dcf55 MM	498	slargs.slot_array = bufmap->readdir_index_array;
	499	slargs.slot_lock = &bufmap->readdir_index_lock;
	500	slargs.slot_wq = &readdir_waitq;
	501	ret = wait_for_a_slot(&slargs, buffer_index);
	502	if (ret)
8bb8aefd	503	orangefs_bufmap_unref(bufmap);
274dcf55 MM	504	*mapp = bufmap;
	505	return ret;
	506	}
	507
8bb8aefd	508	void readdir_index_put(struct orangefs_bufmap *bufmap, int buffer_index)
274dcf55 MM	509	{
	510	struct slot_args slargs;
	511
8bb8aefd	512	slargs.slot_count = ORANGEFS_READDIR_DEFAULT_DESC_COUNT;
274dcf55 MM	513	slargs.slot_array = bufmap->readdir_index_array;
	514	slargs.slot_lock = &bufmap->readdir_index_lock;
	515	slargs.slot_wq = &readdir_waitq;
	516	put_back_slot(&slargs, buffer_index);
8bb8aefd	517	orangefs_bufmap_unref(bufmap);
274dcf55 MM	518	}
274dcf55 MM	519
b5e376ea MM	520	/*
	521	* we've been handed an iovec, we need to copy it to
	522	* the shared memory descriptor at "buffer_index".
	523	*/
8bb8aefd	524	int orangefs_bufmap_copy_from_iovec(struct orangefs_bufmap *bufmap,
54804949 MM	525	struct iov_iter *iter,
	526	int buffer_index,
	527	size_t size)
274dcf55	528	{
8bb8aefd	529	struct orangefs_bufmap_desc *to = &bufmap->desc_array[buffer_index];
4d1c4404	530	int i;
274dcf55 MM	531
274dcf55 MM	532	gossip_debug(GOSSIP_BUFMAP_DEBUG,
34204fde	533	"%s: buffer_index:%d: size:%zu:\n",
4d1c4404	534	__func__, buffer_index, size);
274dcf55	535
274dcf55	536
4d1c4404	537	for (i = 0; size; i++) {
34204fde AV	538	struct page *page = to->page_array[i];
	539	size_t n = size;
	540	if (n > PAGE_SIZE)
	541	n = PAGE_SIZE;
	542	n = copy_page_from_iter(page, 0, n, iter);
	543	if (!n)
	544	return -EFAULT;
	545	size -= n;
274dcf55	546	}
34204fde	547	return 0;
274dcf55	548
274dcf55 MM	549	}
	550
	551	/*
b5e376ea MM	552	* we've been handed an iovec, we need to fill it from
b5e376ea MM	553	* the shared memory descriptor at "buffer_index".
274dcf55	554	*/
8bb8aefd	555	int orangefs_bufmap_copy_to_iovec(struct orangefs_bufmap *bufmap,
4d1c4404	556	struct iov_iter *iter,
5c278228 AV	557	int buffer_index,
5c278228 AV	558	size_t size)
274dcf55	559	{
8bb8aefd	560	struct orangefs_bufmap_desc *from = &bufmap->desc_array[buffer_index];
4d1c4404	561	int i;
274dcf55 MM	562
274dcf55 MM	563	gossip_debug(GOSSIP_BUFMAP_DEBUG,
5c278228 AV	564	"%s: buffer_index:%d: size:%zu:\n",
5c278228 AV	565	__func__, buffer_index, size);
274dcf55	566
274dcf55	567
5c278228 AV	568	for (i = 0; size; i++) {
	569	struct page *page = from->page_array[i];
	570	size_t n = size;
	571	if (n > PAGE_SIZE)
	572	n = PAGE_SIZE;
	573	n = copy_page_to_iter(page, 0, n, iter);
	574	if (!n)
	575	return -EFAULT;
	576	size -= n;
274dcf55	577	}
5c278228	578	return 0;
274dcf55	579	}