[linux.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 */
	down_write(&current->mm->mmap_sem);
	ret = get_user_pages(current,
			     current->mm,
			     (unsigned long)user_desc->ptr,
			     bufmap->page_count,
			     1,
			     0,
			     bufmap->page_array,
			     NULL);
	up_write(&current->mm->mmap_sem);

	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;
	struct page *page;
	size_t copied;
	int i;

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

	to = &bufmap->desc_array[buffer_index];

	for (i = 0; size; i++) {
		page = to->page_array[i];
		copied = copy_page_from_iter(page, 0, PAGE_SIZE, iter);
		size -= copied;
		if ((copied == 0) && (size))
			break;
	}

	return size ? -EFAULT : 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)
{
	struct pvfs_bufmap_desc *from;
	struct page *page;
	int i;
	size_t written;

	gossip_debug(GOSSIP_BUFMAP_DEBUG,
		     "%s: buffer_index:%d: iov_iter_count(iter):%lu:\n",
		     __func__, buffer_index, iov_iter_count(iter));

	from = &bufmap->desc_array[buffer_index];

	for (i = 0; iov_iter_count(iter); i++) {
		page = from->page_array[i];
		written = copy_page_to_iter(page, 0, PAGE_SIZE, iter);
		if ((written == 0) && (iov_iter_count(iter)))
			break;
	}

	return iov_iter_count(iter) ? -EFAULT : 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 */
174	down_write(&current->mm->mmap_sem);
175	ret = get_user_pages(current,
176	current->mm,
177	(unsigned long)user_desc->ptr,
178	bufmap->page_count,
179	1,
180	0,
181	bufmap->page_array,
182	NULL);
183	up_write(&current->mm->mmap_sem);
184
185	if (ret < 0)
186	return ret;
187
188	if (ret != bufmap->page_count) {
189	gossip_err("pvfs2 error: asked for %d pages, only got %d.\n",
190	bufmap->page_count, ret);
191
192	for (i = 0; i < ret; i++) {
193	SetPageError(bufmap->page_array[i]);
194	page_cache_release(bufmap->page_array[i]);
195	}
196	return -ENOMEM;
197	}
198
199	/*
200	* ideally we want to get kernel space pointers for each page, but
201	* we can't kmap that many pages at once if highmem is being used.
202	* so instead, we just kmap/kunmap the page address each time the
203	* kaddr is needed.
204	*/
205	for (i = 0; i < bufmap->page_count; i++)
206	flush_dcache_page(bufmap->page_array[i]);
207
208	/* build a list of available descriptors */
209	for (offset = 0, i = 0; i < bufmap->desc_count; i++) {
210	bufmap->desc_array[i].page_array = &bufmap->page_array[offset];
211	bufmap->desc_array[i].array_count = pages_per_desc;
212	bufmap->desc_array[i].uaddr =
213	(user_desc->ptr + (i * pages_per_desc * PAGE_SIZE));
214	offset += pages_per_desc;
215	}
216
217	return 0;
218	}
219
220	/*
221	* pvfs_bufmap_initialize()
222	*
223	* initializes the mapped buffer interface
224	*
225	* returns 0 on success, -errno on failure
226	*/
227	int pvfs_bufmap_initialize(struct PVFS_dev_map_desc *user_desc)
228	{
229	struct pvfs2_bufmap *bufmap;
230	int ret = -EINVAL;
231
232	gossip_debug(GOSSIP_BUFMAP_DEBUG,
233	"pvfs_bufmap_initialize: called (ptr ("
234	"%p) sz (%d) cnt(%d).\n",
235	user_desc->ptr,
236	user_desc->size,
237	user_desc->count);
238
239	/*
240	* sanity check alignment and size of buffer that caller wants to
241	* work with
242	*/
243	if (PAGE_ALIGN((unsigned long)user_desc->ptr) !=
244	(unsigned long)user_desc->ptr) {
245	gossip_err("pvfs2 error: memory alignment (front). %p\n",
246	user_desc->ptr);
247	goto out;
248	}
249
250	if (PAGE_ALIGN(((unsigned long)user_desc->ptr + user_desc->total_size))
251	!= (unsigned long)(user_desc->ptr + user_desc->total_size)) {
252	gossip_err("pvfs2 error: memory alignment (back).(%p + %d)\n",
253	user_desc->ptr,
254	user_desc->total_size);
255	goto out;
256	}
257
258	if (user_desc->total_size != (user_desc->size * user_desc->count)) {
259	gossip_err("pvfs2 error: user provided an oddly sized buffer: (%d, %d, %d)\n",
260	user_desc->total_size,
261	user_desc->size,
262	user_desc->count);
263	goto out;
264	}
265
266	if ((user_desc->size % PAGE_SIZE) != 0) {
267	gossip_err("pvfs2 error: bufmap size not page size divisible (%d).\n",
268	user_desc->size);
269	goto out;
270	}
271
272	ret = -ENOMEM;
273	bufmap = pvfs2_bufmap_alloc(user_desc);
274	if (!bufmap)
275	goto out;
276
277	ret = pvfs2_bufmap_map(bufmap, user_desc);
278	if (ret)
279	goto out_free_bufmap;
280
281
282	spin_lock(&pvfs2_bufmap_lock);
283	if (__pvfs2_bufmap) {
284	spin_unlock(&pvfs2_bufmap_lock);
285	gossip_err("pvfs2: error: bufmap already initialized.\n");
286	ret = -EALREADY;
287	goto out_unmap_bufmap;
288	}
289	__pvfs2_bufmap = bufmap;
290	spin_unlock(&pvfs2_bufmap_lock);
291
292	/*
293	* If there are operations in pvfs2_bufmap_init_waitq, wake them up.
294	* This scenario occurs when the client-core is restarted and I/O
295	* requests in the in-progress or waiting tables are restarted. I/O
296	* requests cannot be restarted until the shared memory system is
297	* completely re-initialized, so we put the I/O requests in this
298	* waitq until initialization has completed. NOTE: the I/O requests
299	* are also on a timer, so they don't wait forever just in case the
300	* client-core doesn't come back up.
301	*/
302	wake_up_interruptible(&pvfs2_bufmap_init_waitq);
303
304	gossip_debug(GOSSIP_BUFMAP_DEBUG,
305	"pvfs_bufmap_initialize: exiting normally\n");
306	return 0;
307
308	out_unmap_bufmap:
309	pvfs2_bufmap_unmap(bufmap);
310	out_free_bufmap:
311	pvfs2_bufmap_free(bufmap);
312	out:
313	return ret;
314	}
315
316	/*
317	* pvfs_bufmap_finalize()
318	*
319	* shuts down the mapped buffer interface and releases any resources
320	* associated with it
321	*
322	* no return value
323	*/
324	void pvfs_bufmap_finalize(void)
325	{
326	gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2_bufmap_finalize: called\n");
327	BUG_ON(!__pvfs2_bufmap);
328	pvfs2_bufmap_unref(__pvfs2_bufmap);
329	gossip_debug(GOSSIP_BUFMAP_DEBUG,
330	"pvfs2_bufmap_finalize: exiting normally\n");
331	}
332
333	struct slot_args {
334	int slot_count;
335	int *slot_array;
336	spinlock_t *slot_lock;
337	wait_queue_head_t *slot_wq;
338	};
339
340	static int wait_for_a_slot(struct slot_args slargs, int buffer_index)
341	{
342	int ret = -1;
343	int i = 0;
344	DECLARE_WAITQUEUE(my_wait, current);
345
346
347	add_wait_queue_exclusive(slargs->slot_wq, &my_wait);
348
349	while (1) {
350	set_current_state(TASK_INTERRUPTIBLE);
351
352	/*
353	* check for available desc, slot_lock is the appropriate
354	* index_lock
355	*/
356	spin_lock(slargs->slot_lock);
357	for (i = 0; i < slargs->slot_count; i++)
358	if (slargs->slot_array[i] == 0) {
359	slargs->slot_array[i] = 1;
360	*buffer_index = i;
361	ret = 0;
362	break;
363	}
364	spin_unlock(slargs->slot_lock);
365
366	/* if we acquired a buffer, then break out of while */
367	if (ret == 0)
368	break;
369
370	if (!signal_pending(current)) {
371	int timeout =
372	MSECS_TO_JIFFIES(1000 * slot_timeout_secs);
373	gossip_debug(GOSSIP_BUFMAP_DEBUG,
374	"[BUFMAP]: waiting %d "
375	"seconds for a slot\n",
376	slot_timeout_secs);
377	if (!schedule_timeout(timeout)) {
378	gossip_debug(GOSSIP_BUFMAP_DEBUG,
379	"*** wait_for_a_slot timed out\n");
380	ret = -ETIMEDOUT;
381	break;
382	}
383	gossip_debug(GOSSIP_BUFMAP_DEBUG,
384	"[BUFMAP]: woken up by a slot becoming available.\n");
385	continue;
386	}
387
388	gossip_debug(GOSSIP_BUFMAP_DEBUG, "pvfs2: %s interrupted.\n",
389	__func__);
390	ret = -EINTR;
391	break;
392	}
393
394	set_current_state(TASK_RUNNING);
395	remove_wait_queue(slargs->slot_wq, &my_wait);
396	return ret;
397	}
398
399	static void put_back_slot(struct slot_args *slargs, int buffer_index)
400	{
401	/* slot_lock is the appropriate index_lock */
402	spin_lock(slargs->slot_lock);
403	if (buffer_index < 0 \|\| buffer_index >= slargs->slot_count) {
404	spin_unlock(slargs->slot_lock);
405	return;
406	}
407
408	/* put the desc back on the queue */
409	slargs->slot_array[buffer_index] = 0;
410	spin_unlock(slargs->slot_lock);
411
412	/* wake up anyone who may be sleeping on the queue */
413	wake_up_interruptible(slargs->slot_wq);
414	}
415
416	/*
417	* pvfs_bufmap_get()
418	*
419	* gets a free mapped buffer descriptor, will sleep until one becomes
420	* available if necessary
421	*
422	* returns 0 on success, -errno on failure
423	*/
424	int pvfs_bufmap_get(struct pvfs2_bufmap *mapp, int buffer_index)
425	{
426	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
427	struct slot_args slargs;
428	int ret;
429
430	if (!bufmap) {
431	gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n");
432	return -EIO;
433	}
434
435	slargs.slot_count = bufmap->desc_count;
436	slargs.slot_array = bufmap->buffer_index_array;
437	slargs.slot_lock = &bufmap->buffer_index_lock;
438	slargs.slot_wq = &bufmap_waitq;
439	ret = wait_for_a_slot(&slargs, buffer_index);
440	if (ret)
441	pvfs2_bufmap_unref(bufmap);
442	*mapp = bufmap;
443	return ret;
444	}
445
446	/*
447	* pvfs_bufmap_put()
448	*
449	* returns a mapped buffer descriptor to the collection
450	*
451	* no return value
452	*/
453	void pvfs_bufmap_put(struct pvfs2_bufmap *bufmap, int buffer_index)
454	{
455	struct slot_args slargs;
456
457	slargs.slot_count = bufmap->desc_count;
458	slargs.slot_array = bufmap->buffer_index_array;
459	slargs.slot_lock = &bufmap->buffer_index_lock;
460	slargs.slot_wq = &bufmap_waitq;
461	put_back_slot(&slargs, buffer_index);
462	pvfs2_bufmap_unref(bufmap);
463	}
464
465	/*
466	* readdir_index_get()
467	*
468	* gets a free descriptor, will sleep until one becomes
469	* available if necessary.
470	* Although the readdir buffers are not mapped into kernel space
471	* we could do that at a later point of time. Regardless, these
472	* indices are used by the client-core.
473	*
474	* returns 0 on success, -errno on failure
475	*/
476	int readdir_index_get(struct pvfs2_bufmap *mapp, int buffer_index)
477	{
478	struct pvfs2_bufmap *bufmap = pvfs2_bufmap_ref();
479	struct slot_args slargs;
480	int ret;
481
482	if (!bufmap) {
483	gossip_err("pvfs2: please confirm that pvfs2-client daemon is running.\n");
484	return -EIO;
485	}
486
487	slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT;
488	slargs.slot_array = bufmap->readdir_index_array;
489	slargs.slot_lock = &bufmap->readdir_index_lock;
490	slargs.slot_wq = &readdir_waitq;
491	ret = wait_for_a_slot(&slargs, buffer_index);
492	if (ret)
493	pvfs2_bufmap_unref(bufmap);
494	*mapp = bufmap;
495	return ret;
496	}
497
498	void readdir_index_put(struct pvfs2_bufmap *bufmap, int buffer_index)
499	{
500	struct slot_args slargs;
501
502	slargs.slot_count = PVFS2_READDIR_DEFAULT_DESC_COUNT;
503	slargs.slot_array = bufmap->readdir_index_array;
504	slargs.slot_lock = &bufmap->readdir_index_lock;
505	slargs.slot_wq = &readdir_waitq;
506	put_back_slot(&slargs, buffer_index);
507	pvfs2_bufmap_unref(bufmap);
508	}
509
4d1c4404	510	int pvfs_bufmap_copy_from_iovec(struct pvfs2_bufmap *bufmap,
54804949 MM	511	struct iov_iter *iter,
	512	int buffer_index,
	513	size_t size)
274dcf55	514	{
274dcf55	515	struct pvfs_bufmap_desc *to;
4d1c4404 MM	516	struct page *page;
	517	size_t copied;
	518	int i;
274dcf55 MM	519
274dcf55 MM	520	gossip_debug(GOSSIP_BUFMAP_DEBUG,
4d1c4404 MM	521	"%s: buffer_index:%d: size:%lu:\n",
4d1c4404 MM	522	__func__, buffer_index, size);
274dcf55 MM	523
274dcf55 MM	524	to = &bufmap->desc_array[buffer_index];
274dcf55	525
4d1c4404 MM	526	for (i = 0; size; i++) {
	527	page = to->page_array[i];
	528	copied = copy_page_from_iter(page, 0, PAGE_SIZE, iter);
	529	size -= copied;
	530	if ((copied == 0) && (size))
	531	break;
274dcf55 MM	532	}
274dcf55 MM	533
4d1c4404	534	return size ? -EFAULT : 0;
274dcf55	535
274dcf55 MM	536	}
	537
	538	/*
4d1c4404 MM	539	* Iterate through the array of pages containing the bytes from
4d1c4404 MM	540	* a file being read.
274dcf55	541	*
274dcf55	542	*/
4d1c4404 MM	543	int pvfs_bufmap_copy_to_iovec(struct pvfs2_bufmap *bufmap,
	544	struct iov_iter *iter,
	545	int buffer_index)
274dcf55	546	{
274dcf55	547	struct pvfs_bufmap_desc *from;
4d1c4404 MM	548	struct page *page;
	549	int i;
	550	size_t written;
274dcf55 MM	551
274dcf55 MM	552	gossip_debug(GOSSIP_BUFMAP_DEBUG,
4d1c4404 MM	553	"%s: buffer_index:%d: iov_iter_count(iter):%lu:\n",
4d1c4404 MM	554	__func__, buffer_index, iov_iter_count(iter));
274dcf55	555
54804949	556	from = &bufmap->desc_array[buffer_index];
274dcf55	557
4d1c4404 MM	558	for (i = 0; iov_iter_count(iter); i++) {
	559	page = from->page_array[i];
	560	written = copy_page_to_iter(page, 0, PAGE_SIZE, iter);
	561	if ((written == 0) && (iov_iter_count(iter)))
	562	break;
274dcf55 MM	563	}
274dcf55 MM	564
54804949	565	return iov_iter_count(iter) ? -EFAULT : 0;
274dcf55	566	}