[linux-block.git] / fs / orangefs / pvfs2-bufmap.c

/*
 * (C) 2001 Clemson University and The University of Chicago
 *
 * See COPYING in top-level directory.
 */
#include "protocol.h"
#include "pvfs2-kernel.h"
#include "pvfs2-bufmap.h"

DECLARE_WAIT_QUEUE_HEAD(pvfs2_bufmap_init_waitq);

static struct pvfs2_bufmap {
	atomic_t refcnt;

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

	struct page **page_array;
	struct pvfs_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[PVFS2_READDIR_DEFAULT_DESC_COUNT];
	spinlock_t readdir_index_lock;
} *__pvfs2_bufmap;

static DEFINE_SPINLOCK(pvfs2_bufmap_lock);

static void
pvfs2_bufmap_unmap(struct pvfs2_bufmap *bufmap)
{
	int i;

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

static void
pvfs2_bufmap_free(struct pvfs2_bufmap *bufmap)
{
	kfree(bufmap->page_array);
	kfree(bufmap->desc_array);
	kfree(bufmap->buffer_index_array);
	kfree(bufmap);
}

struct pvfs2_bufmap *pvfs2_bufmap_ref(void)
{
	struct pvfs2_bufmap *bufmap = NULL;

	spin_lock(&pvfs2_bufmap_lock);
	if (__pvfs2_bufmap) {
		bufmap = __pvfs2_bufmap;
		atomic_inc(&bufmap->refcnt);
	}
	spin_unlock(&pvfs2_bufmap_lock);
	return bufmap;
}

void pvfs2_bufmap_unref(struct pvfs2_bufmap *bufmap)
{
	if (atomic_dec_and_lock(&bufmap->refcnt, &pvfs2_bufmap_lock)) {
		__pvfs2_bufmap = NULL;
		spin_unlock(&pvfs2_bufmap_lock);

		pvfs2_bufmap_unmap(bufmap);
		pvfs2_bufmap_free(bufmap);
	}
}

inline int pvfs_bufmap_size_query(void)
{
	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
	int size = bufmap ? bufmap->desc_size : 0;

	pvfs2_bufmap_unref(bufmap);
	return size;
}

inline int pvfs_bufmap_shift_query(void)
{
	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
	int shift = bufmap ? bufmap->desc_shift : 0;

	pvfs2_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 __pvfs2_bufmap ? 1 : 0;
}


static struct pvfs2_bufmap *
pvfs2_bufmap_alloc(struct PVFS_dev_map_desc *user_desc)
{
	struct pvfs2_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("pvfs2: 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 pvfs_bufmap_desc),
			GFP_KERNEL);
	if (!bufmap->desc_array) {
		gossip_err("pvfs2: 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
pvfs2_bufmap_map(struct pvfs2_bufmap *bufmap,
		struct PVFS_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("pvfs2 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;
}

/*
 * pvfs_bufmap_initialize()
 *
 * initializes the mapped buffer interface
 *
 * returns 0 on success, -errno on failure
 */
int pvfs_bufmap_initialize(struct PVFS_dev_map_desc *user_desc)
{
	struct pvfs2_bufmap *bufmap;
	int ret = -EINVAL;

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "pvfs_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("pvfs2 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("pvfs2 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("pvfs2 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("pvfs2 error: bufmap size not page size divisible (%d).\n",
			   user_desc->size);
		goto out;
	}

	ret = -ENOMEM;
	bufmap = pvfs2_bufmap_alloc(user_desc);
	if (!bufmap)
		goto out;

	ret = pvfs2_bufmap_map(bufmap, user_desc);
	if (ret)
		goto out_free_bufmap;


	spin_lock(&pvfs2_bufmap_lock);
	if (__pvfs2_bufmap) {
		spin_unlock(&pvfs2_bufmap_lock);
		gossip_err("pvfs2: error: bufmap already initialized.\n");
		ret = -EALREADY;
		goto out_unmap_bufmap;
	}
	__pvfs2_bufmap = bufmap;
	spin_unlock(&pvfs2_bufmap_lock);

	/*
	 * If there are operations in pvfs2_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(&pvfs2_bufmap_init_waitq);

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

out_unmap_bufmap:
	pvfs2_bufmap_unmap(bufmap);
out_free_bufmap:
	pvfs2_bufmap_free(bufmap);
out:
	return ret;
}

/*
 * pvfs_bufmap_finalize()
 *
 * shuts down the mapped buffer interface and releases any resources
 * associated with it
 *
 * no return value
 */
void pvfs_bufmap_finalize(void)
{
	gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2_bufmap_finalize: called\n");
	BUG_ON(!__pvfs2_bufmap);
	pvfs2_bufmap_unref(__pvfs2_bufmap);
	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "pvfs2_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;
	DECLARE_WAITQUEUE(my_wait, current);


	add_wait_queue_exclusive(slargs->slot_wq, &my_wait);

	while (1) {
		set_current_state(TASK_INTERRUPTIBLE);

		/*
		 * check for available desc, slot_lock is the appropriate
		 * index_lock
		 */
		spin_lock(slargs->slot_lock);
		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, "pvfs2: %s interrupted.\n",
			     __func__);
		ret = -EINTR;
		break;
	}

	set_current_state(TASK_RUNNING);
	remove_wait_queue(slargs->slot_wq, &my_wait);
	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);
}

/*
 * pvfs_bufmap_get()
 *
 * gets a free mapped buffer descriptor, will sleep until one becomes
 * available if necessary
 *
 * returns 0 on success, -errno on failure
 */
int pvfs_bufmap_get(struct pvfs2_bufmap **mapp, int *buffer_index)
{
	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
	struct slot_args slargs;
	int ret;

	if (!bufmap) {
		gossip_err("pvfs2: 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)
		pvfs2_bufmap_unref(bufmap);
	*mapp = bufmap;
	return ret;
}

/*
 * pvfs_bufmap_put()
 *
 * returns a mapped buffer descriptor to the collection
 *
 * no return value
 */
void pvfs_bufmap_put(struct pvfs2_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);
	pvfs2_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 pvfs2_bufmap **mapp, int *buffer_index)
{
	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
	struct slot_args slargs;
	int ret;

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

	slargs.slot_count = PVFS2_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)
		pvfs2_bufmap_unref(bufmap);
	*mapp = bufmap;
	return ret;
}

void readdir_index_put(struct pvfs2_bufmap *bufmap, int buffer_index)
{
	struct slot_args slargs;

	slargs.slot_count = PVFS2_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);
	pvfs2_bufmap_unref(bufmap);
}

int pvfs_bufmap_copy_from_iovec(struct pvfs2_bufmap *bufmap,
				struct iov_iter *iter,
				int buffer_index,
				size_t size)
{
	struct pvfs_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;

}

/*
 * Iterate through the array of pages containing the bytes from
 * a file being read.
 *
 */
int pvfs_bufmap_copy_to_iovec(struct pvfs2_bufmap *bufmap,
				    struct iov_iter *iter,
				    int buffer_index,
				    size_t size)
{
	struct pvfs_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"
	7	#include "pvfs2-kernel.h"
	8	#include "pvfs2-bufmap.h"
	9
	10	DECLARE_WAIT_QUEUE_HEAD(pvfs2_bufmap_init_waitq);
	11
84d02150	12	static struct pvfs2_bufmap {
274dcf55 MM	13	atomic_t refcnt;
	14
	15	int desc_size;
	16	int desc_shift;
	17	int desc_count;
	18	int total_size;
	19	int page_count;
	20
	21	struct page **page_array;
	22	struct pvfs_bufmap_desc *desc_array;
	23
	24	/* array to track usage of buffer descriptors */
	25	int *buffer_index_array;
	26	spinlock_t buffer_index_lock;
	27
	28	/* array to track usage of buffer descriptors for readdir */
	29	int readdir_index_array[PVFS2_READDIR_DEFAULT_DESC_COUNT];
	30	spinlock_t readdir_index_lock;
	31	} *__pvfs2_bufmap;
	32
	33	static DEFINE_SPINLOCK(pvfs2_bufmap_lock);
	34
	35	static void
	36	pvfs2_bufmap_unmap(struct pvfs2_bufmap *bufmap)
	37	{
	38	int i;
	39
	40	for (i = 0; i < bufmap->page_count; i++)
	41	page_cache_release(bufmap->page_array[i]);
	42	}
	43
	44	static void
	45	pvfs2_bufmap_free(struct pvfs2_bufmap *bufmap)
	46	{
	47	kfree(bufmap->page_array);
	48	kfree(bufmap->desc_array);
	49	kfree(bufmap->buffer_index_array);
	50	kfree(bufmap);
	51	}
	52
	53	struct pvfs2_bufmap *pvfs2_bufmap_ref(void)
	54	{
	55	struct pvfs2_bufmap *bufmap = NULL;
	56
	57	spin_lock(&pvfs2_bufmap_lock);
	58	if (__pvfs2_bufmap) {
	59	bufmap = __pvfs2_bufmap;
	60	atomic_inc(&bufmap->refcnt);
	61	}
	62	spin_unlock(&pvfs2_bufmap_lock);
	63	return bufmap;
	64	}
	65
	66	void pvfs2_bufmap_unref(struct pvfs2_bufmap *bufmap)
	67	{
	68	if (atomic_dec_and_lock(&bufmap->refcnt, &pvfs2_bufmap_lock)) {
	69	__pvfs2_bufmap = NULL;
	70	spin_unlock(&pvfs2_bufmap_lock);
	71
	72	pvfs2_bufmap_unmap(bufmap);
	73	pvfs2_bufmap_free(bufmap);
	74	}
	75	}
	76
77	inline int pvfs_bufmap_size_query(void)
78	{
79	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
80	int size = bufmap ? bufmap->desc_size : 0;
81
82	pvfs2_bufmap_unref(bufmap);
83	return size;
84	}
85
86	inline int pvfs_bufmap_shift_query(void)
87	{
88	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
89	int shift = bufmap ? bufmap->desc_shift : 0;
90
91	pvfs2_bufmap_unref(bufmap);
92	return shift;
93	}
94
95	static DECLARE_WAIT_QUEUE_HEAD(bufmap_waitq);
96	static DECLARE_WAIT_QUEUE_HEAD(readdir_waitq);
97
98	/*
99	* get_bufmap_init
100	*
101	* If bufmap_init is 1, then the shared memory system, including the
102	* buffer_index_array, is available. Otherwise, it is not.
103	*
104	* returns the value of bufmap_init
105	*/
106	int get_bufmap_init(void)
107	{
108	return __pvfs2_bufmap ? 1 : 0;
109	}
110
111
112	static struct pvfs2_bufmap *
113	pvfs2_bufmap_alloc(struct PVFS_dev_map_desc *user_desc)
114	{
115	struct pvfs2_bufmap *bufmap;
116
117	bufmap = kzalloc(sizeof(*bufmap), GFP_KERNEL);
118	if (!bufmap)
119	goto out;
120
121	atomic_set(&bufmap->refcnt, 1);
122	bufmap->total_size = user_desc->total_size;
123	bufmap->desc_count = user_desc->count;
124	bufmap->desc_size = user_desc->size;
125	bufmap->desc_shift = ilog2(bufmap->desc_size);
126
127	spin_lock_init(&bufmap->buffer_index_lock);
128	bufmap->buffer_index_array =
129	kcalloc(bufmap->desc_count, sizeof(int), GFP_KERNEL);
130	if (!bufmap->buffer_index_array) {
131	gossip_err("pvfs2: could not allocate %d buffer indices\n",
132	bufmap->desc_count);
133	goto out_free_bufmap;
134	}
135	spin_lock_init(&bufmap->readdir_index_lock);
136
137	bufmap->desc_array =
138	kcalloc(bufmap->desc_count, sizeof(struct pvfs_bufmap_desc),
139	GFP_KERNEL);
140	if (!bufmap->desc_array) {
141	gossip_err("pvfs2: could not allocate %d descriptors\n",
142	bufmap->desc_count);
143	goto out_free_index_array;
144	}
145
146	bufmap->page_count = bufmap->total_size / PAGE_SIZE;
147
148	/* allocate storage to track our page mappings */
149	bufmap->page_array =
150	kcalloc(bufmap->page_count, sizeof(struct page *), GFP_KERNEL);
151	if (!bufmap->page_array)
152	goto out_free_desc_array;
153
154	return bufmap;
155
156	out_free_desc_array:
157	kfree(bufmap->desc_array);
158	out_free_index_array:
159	kfree(bufmap->buffer_index_array);
160	out_free_bufmap:
161	kfree(bufmap);
162	out:
163	return NULL;
164	}
165
166	static int
167	pvfs2_bufmap_map(struct pvfs2_bufmap *bufmap,
168	struct PVFS_dev_map_desc *user_desc)
169	{
170	int pages_per_desc = bufmap->desc_size / PAGE_SIZE;
171	int offset = 0, ret, i;
172
173	/* map the pages */
16742f2d AV	174	ret = get_user_pages_fast((unsigned long)user_desc->ptr,
16742f2d AV	175	bufmap->page_count, 1, bufmap->page_array);
274dcf55 MM	176
	177	if (ret < 0)
	178	return ret;
	179
	180	if (ret != bufmap->page_count) {
	181	gossip_err("pvfs2 error: asked for %d pages, only got %d.\n",
	182	bufmap->page_count, ret);
	183
	184	for (i = 0; i < ret; i++) {
	185	SetPageError(bufmap->page_array[i]);
	186	page_cache_release(bufmap->page_array[i]);
	187	}
	188	return -ENOMEM;
	189	}
	190
	191	/*
	192	* ideally we want to get kernel space pointers for each page, but
	193	* we can't kmap that many pages at once if highmem is being used.
	194	* so instead, we just kmap/kunmap the page address each time the
	195	* kaddr is needed.
	196	*/
	197	for (i = 0; i < bufmap->page_count; i++)
	198	flush_dcache_page(bufmap->page_array[i]);
	199
	200	/* build a list of available descriptors */
	201	for (offset = 0, i = 0; i < bufmap->desc_count; i++) {
	202	bufmap->desc_array[i].page_array = &bufmap->page_array[offset];
	203	bufmap->desc_array[i].array_count = pages_per_desc;
	204	bufmap->desc_array[i].uaddr =
	205	(user_desc->ptr + (i * pages_per_desc * PAGE_SIZE));
	206	offset += pages_per_desc;
	207	}
	208
	209	return 0;
	210	}
	211
	212	/*
	213	* pvfs_bufmap_initialize()
	214	*
	215	* initializes the mapped buffer interface
	216	*
	217	* returns 0 on success, -errno on failure
	218	*/
	219	int pvfs_bufmap_initialize(struct PVFS_dev_map_desc *user_desc)
	220	{
	221	struct pvfs2_bufmap *bufmap;
	222	int ret = -EINVAL;
	223
	224	gossip_debug(GOSSIP_BUFMAP_DEBUG,
	225	"pvfs_bufmap_initialize: called (ptr ("
	226	"%p) sz (%d) cnt(%d).\n",
	227	user_desc->ptr,
	228	user_desc->size,
	229	user_desc->count);
	230
	231	/*
	232	* sanity check alignment and size of buffer that caller wants to
	233	* work with
	234	*/
	235	if (PAGE_ALIGN((unsigned long)user_desc->ptr) !=
	236	(unsigned long)user_desc->ptr) {
	237	gossip_err("pvfs2 error: memory alignment (front). %p\n",
	238	user_desc->ptr);
	239	goto out;
240	}
241
242	if (PAGE_ALIGN(((unsigned long)user_desc->ptr + user_desc->total_size))
243	!= (unsigned long)(user_desc->ptr + user_desc->total_size)) {
244	gossip_err("pvfs2 error: memory alignment (back).(%p + %d)\n",
245	user_desc->ptr,
246	user_desc->total_size);
247	goto out;
248	}
249
250	if (user_desc->total_size != (user_desc->size * user_desc->count)) {
251	gossip_err("pvfs2 error: user provided an oddly sized buffer: (%d, %d, %d)\n",
252	user_desc->total_size,
253	user_desc->size,
254	user_desc->count);
255	goto out;
256	}
257
258	if ((user_desc->size % PAGE_SIZE) != 0) {
259	gossip_err("pvfs2 error: bufmap size not page size divisible (%d).\n",
260	user_desc->size);
261	goto out;
262	}
263
264	ret = -ENOMEM;
265	bufmap = pvfs2_bufmap_alloc(user_desc);
266	if (!bufmap)
267	goto out;
268
269	ret = pvfs2_bufmap_map(bufmap, user_desc);
270	if (ret)
271	goto out_free_bufmap;
272
273
274	spin_lock(&pvfs2_bufmap_lock);
275	if (__pvfs2_bufmap) {
276	spin_unlock(&pvfs2_bufmap_lock);
277	gossip_err("pvfs2: error: bufmap already initialized.\n");
278	ret = -EALREADY;
279	goto out_unmap_bufmap;
280	}
281	__pvfs2_bufmap = bufmap;
282	spin_unlock(&pvfs2_bufmap_lock);
283
284	/*
285	* If there are operations in pvfs2_bufmap_init_waitq, wake them up.
286	* This scenario occurs when the client-core is restarted and I/O
287	* requests in the in-progress or waiting tables are restarted. I/O
288	* requests cannot be restarted until the shared memory system is
289	* completely re-initialized, so we put the I/O requests in this
290	* waitq until initialization has completed. NOTE: the I/O requests
291	* are also on a timer, so they don't wait forever just in case the
292	* client-core doesn't come back up.
293	*/
294	wake_up_interruptible(&pvfs2_bufmap_init_waitq);
295
296	gossip_debug(GOSSIP_BUFMAP_DEBUG,
297	"pvfs_bufmap_initialize: exiting normally\n");
298	return 0;
299
300	out_unmap_bufmap:
301	pvfs2_bufmap_unmap(bufmap);
302	out_free_bufmap:
303	pvfs2_bufmap_free(bufmap);
304	out:
305	return ret;
306	}
307
308	/*
309	* pvfs_bufmap_finalize()
310	*
311	* shuts down the mapped buffer interface and releases any resources
312	* associated with it
313	*
314	* no return value
315	*/
316	void pvfs_bufmap_finalize(void)
317	{
318	gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2_bufmap_finalize: called\n");
319	BUG_ON(!__pvfs2_bufmap);
320	pvfs2_bufmap_unref(__pvfs2_bufmap);
321	gossip_debug(GOSSIP_BUFMAP_DEBUG,
322	"pvfs2_bufmap_finalize: exiting normally\n");
323	}
324
325	struct slot_args {
326	int slot_count;
327	int *slot_array;
328	spinlock_t *slot_lock;
329	wait_queue_head_t *slot_wq;
330	};
331
332	static int wait_for_a_slot(struct slot_args slargs, int buffer_index)
333	{
334	int ret = -1;
335	int i = 0;
336	DECLARE_WAITQUEUE(my_wait, current);
337
338
339	add_wait_queue_exclusive(slargs->slot_wq, &my_wait);
340
341	while (1) {
342	set_current_state(TASK_INTERRUPTIBLE);
343
344	/*
345	* check for available desc, slot_lock is the appropriate
346	* index_lock
347	*/
348	spin_lock(slargs->slot_lock);
349	for (i = 0; i < slargs->slot_count; i++)
350	if (slargs->slot_array[i] == 0) {
351	slargs->slot_array[i] = 1;
352	*buffer_index = i;
353	ret = 0;
354	break;
355	}
356	spin_unlock(slargs->slot_lock);
357
358	/* if we acquired a buffer, then break out of while */
359	if (ret == 0)
360	break;
361
362	if (!signal_pending(current)) {
363	int timeout =
364	MSECS_TO_JIFFIES(1000 * slot_timeout_secs);
365	gossip_debug(GOSSIP_BUFMAP_DEBUG,
366	"[BUFMAP]: waiting %d "
367	"seconds for a slot\n",
368	slot_timeout_secs);
369	if (!schedule_timeout(timeout)) {
370	gossip_debug(GOSSIP_BUFMAP_DEBUG,
371	"*** wait_for_a_slot timed out\n");
372	ret = -ETIMEDOUT;
373	break;
374	}
375	gossip_debug(GOSSIP_BUFMAP_DEBUG,
376	"[BUFMAP]: woken up by a slot becoming available.\n");
377	continue;
378	}
379
380	gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2: %s interrupted.\n",
381	__func__);
382	ret = -EINTR;
383	break;
384	}
385
386	set_current_state(TASK_RUNNING);
387	remove_wait_queue(slargs->slot_wq, &my_wait);
388	return ret;
389	}
390
391	static void put_back_slot(struct slot_args *slargs, int buffer_index)
392	{
393	/* slot_lock is the appropriate index_lock */
394	spin_lock(slargs->slot_lock);
395	if (buffer_index < 0 \|\| buffer_index >= slargs->slot_count) {
396	spin_unlock(slargs->slot_lock);
397	return;
398	}
399
400	/* put the desc back on the queue */
401	slargs->slot_array[buffer_index] = 0;
402	spin_unlock(slargs->slot_lock);
403
404	/* wake up anyone who may be sleeping on the queue */
405	wake_up_interruptible(slargs->slot_wq);
406	}
407
408	/*
409	* pvfs_bufmap_get()
410	*
411	* gets a free mapped buffer descriptor, will sleep until one becomes
412	* available if necessary
413	*
414	* returns 0 on success, -errno on failure
415	*/
416	int pvfs_bufmap_get(struct pvfs2_bufmap *mapp, int buffer_index)
417	{
418	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
419	struct slot_args slargs;
420	int ret;
421
422	if (!bufmap) {
423	gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n");
424	return -EIO;
425	}
426
427	slargs.slot_count = bufmap->desc_count;
428	slargs.slot_array = bufmap->buffer_index_array;
429	slargs.slot_lock = &bufmap->buffer_index_lock;
430	slargs.slot_wq = &bufmap_waitq;
431	ret = wait_for_a_slot(&slargs, buffer_index);
432	if (ret)
433	pvfs2_bufmap_unref(bufmap);
434	*mapp = bufmap;
435	return ret;
436	}
437
438	/*
439	* pvfs_bufmap_put()
440	*
441	* returns a mapped buffer descriptor to the collection
442	*
443	* no return value
444	*/
445	void pvfs_bufmap_put(struct pvfs2_bufmap *bufmap, int buffer_index)
446	{
447	struct slot_args slargs;
448
449	slargs.slot_count = bufmap->desc_count;
450	slargs.slot_array = bufmap->buffer_index_array;
451	slargs.slot_lock = &bufmap->buffer_index_lock;
452	slargs.slot_wq = &bufmap_waitq;
453	put_back_slot(&slargs, buffer_index);
454	pvfs2_bufmap_unref(bufmap);
455	}
456
457	/*
458	* readdir_index_get()
459	*
460	* gets a free descriptor, will sleep until one becomes
461	* available if necessary.
462	* Although the readdir buffers are not mapped into kernel space
463	* we could do that at a later point of time. Regardless, these
464	* indices are used by the client-core.
465	*
466	* returns 0 on success, -errno on failure
467	*/
468	int readdir_index_get(struct pvfs2_bufmap *mapp, int buffer_index)
469	{
470	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
471	struct slot_args slargs;
472	int ret;
473
474	if (!bufmap) {
475	gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n");
476	return -EIO;
477	}
478
479	slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT;
480	slargs.slot_array = bufmap->readdir_index_array;
481	slargs.slot_lock = &bufmap->readdir_index_lock;
482	slargs.slot_wq = &readdir_waitq;
483	ret = wait_for_a_slot(&slargs, buffer_index);
484	if (ret)
485	pvfs2_bufmap_unref(bufmap);
486	*mapp = bufmap;
487	return ret;
488	}
489
490	void readdir_index_put(struct pvfs2_bufmap *bufmap, int buffer_index)
491	{
492	struct slot_args slargs;
493
494	slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT;
495	slargs.slot_array = bufmap->readdir_index_array;
496	slargs.slot_lock = &bufmap->readdir_index_lock;
497	slargs.slot_wq = &readdir_waitq;
498	put_back_slot(&slargs, buffer_index);
499	pvfs2_bufmap_unref(bufmap);
500	}
501
4d1c4404	502	int pvfs_bufmap_copy_from_iovec(struct pvfs2_bufmap *bufmap,
54804949 MM	503	struct iov_iter *iter,
	504	int buffer_index,
	505	size_t size)
274dcf55	506	{
34204fde	507	struct pvfs_bufmap_desc *to = &bufmap->desc_array[buffer_index];
4d1c4404	508	int i;
274dcf55 MM	509
274dcf55 MM	510	gossip_debug(GOSSIP_BUFMAP_DEBUG,
34204fde	511	"%s: buffer_index:%d: size:%zu:\n",
4d1c4404	512	__func__, buffer_index, size);
274dcf55	513
274dcf55	514
4d1c4404	515	for (i = 0; size; i++) {
34204fde AV	516	struct page *page = to->page_array[i];
	517	size_t n = size;
	518	if (n > PAGE_SIZE)
	519	n = PAGE_SIZE;
	520	n = copy_page_from_iter(page, 0, n, iter);
	521	if (!n)
	522	return -EFAULT;
	523	size -= n;
274dcf55	524	}
34204fde	525	return 0;
274dcf55	526
274dcf55 MM	527	}
	528
	529	/*
4d1c4404 MM	530	* Iterate through the array of pages containing the bytes from
4d1c4404 MM	531	* a file being read.
274dcf55	532	*
274dcf55	533	*/
4d1c4404 MM	534	int pvfs_bufmap_copy_to_iovec(struct pvfs2_bufmap *bufmap,
4d1c4404 MM	535	struct iov_iter *iter,
5c278228 AV	536	int buffer_index,
5c278228 AV	537	size_t size)
274dcf55	538	{
5c278228	539	struct pvfs_bufmap_desc *from = &bufmap->desc_array[buffer_index];
4d1c4404	540	int i;
274dcf55 MM	541
274dcf55 MM	542	gossip_debug(GOSSIP_BUFMAP_DEBUG,
5c278228 AV	543	"%s: buffer_index:%d: size:%zu:\n",
5c278228 AV	544	__func__, buffer_index, size);
274dcf55	545
274dcf55	546
5c278228 AV	547	for (i = 0; size; i++) {
	548	struct page *page = from->page_array[i];
	549	size_t n = size;
	550	if (n > PAGE_SIZE)
	551	n = PAGE_SIZE;
	552	n = copy_page_to_iter(page, 0, n, iter);
	553	if (!n)
	554	return -EFAULT;
	555	size -= n;
274dcf55	556	}
5c278228	557	return 0;
274dcf55	558	}