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

// SPDX-License-Identifier: GPL-2.0
/*
 * (C) 2001 Clemson University and The University of Chicago
 * Copyright 2018 Omnibond Systems, L.L.C.
 *
 * See COPYING in top-level directory.
 */

/*
 *  Linux VFS file operations.
 */

#include "protocol.h"
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"
#include <linux/fs.h>
#include <linux/filelock.h>
#include <linux/pagemap.h>

static int flush_racache(struct inode *inode)
{
	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	struct orangefs_kernel_op_s *new_op;
	int ret;

	gossip_debug(GOSSIP_UTILS_DEBUG,
	    "%s: %pU: Handle is %pU | fs_id %d\n", __func__,
	    get_khandle_from_ino(inode), &orangefs_inode->refn.khandle,
	    orangefs_inode->refn.fs_id);

	new_op = op_alloc(ORANGEFS_VFS_OP_RA_FLUSH);
	if (!new_op)
		return -ENOMEM;
	new_op->upcall.req.ra_cache_flush.refn = orangefs_inode->refn;

	ret = service_operation(new_op, "orangefs_flush_racache",
	    get_interruptible_flag(inode));

	gossip_debug(GOSSIP_UTILS_DEBUG, "%s: got return value of %d\n",
	    __func__, ret);

	op_release(new_op);
	return ret;
}

/*
 * Post and wait for the I/O upcall to finish
 */
ssize_t wait_for_direct_io(enum ORANGEFS_io_type type, struct inode *inode,
	loff_t *offset, struct iov_iter *iter, size_t total_size,
	loff_t readahead_size, struct orangefs_write_range *wr,
	int *index_return, struct file *file)
{
	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
	struct orangefs_kernel_op_s *new_op = NULL;
	int buffer_index;
	ssize_t ret;
	size_t copy_amount;
	bool open_for_read;
	bool open_for_write;

	new_op = op_alloc(ORANGEFS_VFS_OP_FILE_IO);
	if (!new_op)
		return -ENOMEM;

	/* synchronous I/O */
	new_op->upcall.req.io.readahead_size = readahead_size;
	new_op->upcall.req.io.io_type = type;
	new_op->upcall.req.io.refn = orangefs_inode->refn;

populate_shared_memory:
	/* get a shared buffer index */
	buffer_index = orangefs_bufmap_get();
	if (buffer_index < 0) {
		ret = buffer_index;
		gossip_debug(GOSSIP_FILE_DEBUG,
			     "%s: orangefs_bufmap_get failure (%zd)\n",
			     __func__, ret);
		goto out;
	}
	gossip_debug(GOSSIP_FILE_DEBUG,
		     "%s(%pU): GET op %p -> buffer_index %d\n",
		     __func__,
		     handle,
		     new_op,
		     buffer_index);

	new_op->uses_shared_memory = 1;
	new_op->upcall.req.io.buf_index = buffer_index;
	new_op->upcall.req.io.count = total_size;
	new_op->upcall.req.io.offset = *offset;
	if (type == ORANGEFS_IO_WRITE && wr) {
		new_op->upcall.uid = from_kuid(&init_user_ns, wr->uid);
		new_op->upcall.gid = from_kgid(&init_user_ns, wr->gid);
	}
	/*
	 * Orangefs has no open, and orangefs checks file permissions
	 * on each file access. Posix requires that file permissions
	 * be checked on open and nowhere else. Orangefs-through-the-kernel
	 * needs to seem posix compliant.
	 *
	 * The VFS opens files, even if the filesystem provides no
	 * method. We can see if a file was successfully opened for
	 * read and or for write by looking at file->f_mode.
	 *
	 * When writes are flowing from the page cache, file is no
	 * longer available. We can trust the VFS to have checked
	 * file->f_mode before writing to the page cache.
	 *
	 * The mode of a file might change between when it is opened
	 * and IO commences, or it might be created with an arbitrary mode.
	 *
	 * We'll make sure we don't hit EACCES during the IO stage by
	 * using UID 0. Some of the time we have access without changing
	 * to UID 0 - how to check?
	 */
	if (file) {
		open_for_write = file->f_mode & FMODE_WRITE;
		open_for_read = file->f_mode & FMODE_READ;
	} else {
		open_for_write = 1;
		open_for_read = 0; /* not relevant? */
	}
	if ((type == ORANGEFS_IO_WRITE) && open_for_write)
		new_op->upcall.uid = 0;
	if ((type == ORANGEFS_IO_READ) && open_for_read)
		new_op->upcall.uid = 0;

	gossip_debug(GOSSIP_FILE_DEBUG,
		     "%s(%pU): offset: %llu total_size: %zd\n",
		     __func__,
		     handle,
		     llu(*offset),
		     total_size);
	/*
	 * Stage 1: copy the buffers into client-core's address space
	 */
	if (type == ORANGEFS_IO_WRITE && total_size) {
		ret = orangefs_bufmap_copy_from_iovec(iter, buffer_index,
		    total_size);
		if (ret < 0) {
			gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
			    __func__, (long)ret);
			goto out;
		}
	}

	gossip_debug(GOSSIP_FILE_DEBUG,
		     "%s(%pU): Calling post_io_request with tag (%llu)\n",
		     __func__,
		     handle,
		     llu(new_op->tag));

	/* Stage 2: Service the I/O operation */
	ret = service_operation(new_op,
				type == ORANGEFS_IO_WRITE ?
					"file_write" :
					"file_read",
				get_interruptible_flag(inode));

	/*
	 * If service_operation() returns -EAGAIN #and# the operation was
	 * purged from orangefs_request_list or htable_ops_in_progress, then
	 * we know that the client was restarted, causing the shared memory
	 * area to be wiped clean.  To restart a  write operation in this
	 * case, we must re-copy the data from the user's iovec to a NEW
	 * shared memory location. To restart a read operation, we must get
	 * a new shared memory location.
	 */
	if (ret == -EAGAIN && op_state_purged(new_op)) {
		orangefs_bufmap_put(buffer_index);
		if (type == ORANGEFS_IO_WRITE)
			iov_iter_revert(iter, total_size);
		gossip_debug(GOSSIP_FILE_DEBUG,
			     "%s:going to repopulate_shared_memory.\n",
			     __func__);
		goto populate_shared_memory;
	}

	if (ret < 0) {
		if (ret == -EINTR) {
			/*
			 * We can't return EINTR if any data was written,
			 * it's not POSIX. It is minimally acceptable
			 * to give a partial write, the way NFS does.
			 *
			 * It would be optimal to return all or nothing,
			 * but if a userspace write is bigger than
			 * an IO buffer, and the interrupt occurs
			 * between buffer writes, that would not be
			 * possible.
			 */
			switch (new_op->op_state - OP_VFS_STATE_GIVEN_UP) {
			/*
			 * If the op was waiting when the interrupt
			 * occurred, then the client-core did not
			 * trigger the write.
			 */
			case OP_VFS_STATE_WAITING:
				if (*offset == 0)
					ret = -EINTR;
				else
					ret = 0;
				break;
			/*
			 * If the op was in progress when the interrupt
			 * occurred, then the client-core was able to
			 * trigger the write.
			 */
			case OP_VFS_STATE_INPROGR:
				if (type == ORANGEFS_IO_READ)
					ret = -EINTR;
				else
					ret = total_size;
				break;
			default:
				gossip_err("%s: unexpected op state :%d:.\n",
					   __func__,
					   new_op->op_state);
				ret = 0;
				break;
			}
			gossip_debug(GOSSIP_FILE_DEBUG,
				     "%s: got EINTR, state:%d: %p\n",
				     __func__,
				     new_op->op_state,
				     new_op);
		} else {
			gossip_err("%s: error in %s handle %pU, returning %zd\n",
				__func__,
				type == ORANGEFS_IO_READ ?
					"read from" : "write to",
				handle, ret);
		}
		if (orangefs_cancel_op_in_progress(new_op))
			return ret;

		goto out;
	}

	/*
	 * Stage 3: Post copy buffers from client-core's address space
	 */
	if (type == ORANGEFS_IO_READ && new_op->downcall.resp.io.amt_complete) {
		/*
		 * NOTE: the iovector can either contain addresses which
		 *       can futher be kernel-space or user-space addresses.
		 *       or it can pointers to struct page's
		 */

		copy_amount = new_op->downcall.resp.io.amt_complete;

		ret = orangefs_bufmap_copy_to_iovec(iter, buffer_index,
			copy_amount);
		if (ret < 0) {
			gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n",
			    __func__, (long)ret);
			goto out;
		}
	}
	gossip_debug(GOSSIP_FILE_DEBUG,
	    "%s(%pU): Amount %s, returned by the sys-io call:%d\n",
	    __func__,
	    handle,
	    type == ORANGEFS_IO_READ ?  "read" : "written",
	    (int)new_op->downcall.resp.io.amt_complete);

	ret = new_op->downcall.resp.io.amt_complete;

out:
	if (buffer_index >= 0) {
		orangefs_bufmap_put(buffer_index);
		gossip_debug(GOSSIP_FILE_DEBUG,
			"%s(%pU): PUT buffer_index %d\n",
			__func__, handle, buffer_index);
	}
	op_release(new_op);
	return ret;
}

int orangefs_revalidate_mapping(struct inode *inode)
{
	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	struct address_space *mapping = inode->i_mapping;
	unsigned long *bitlock = &orangefs_inode->bitlock;
	int ret;

	while (1) {
		ret = wait_on_bit(bitlock, 1, TASK_KILLABLE);
		if (ret)
			return ret;
		spin_lock(&inode->i_lock);
		if (test_bit(1, bitlock)) {
			spin_unlock(&inode->i_lock);
			continue;
		}
		if (!time_before(jiffies, orangefs_inode->mapping_time))
			break;
		spin_unlock(&inode->i_lock);
		return 0;
	}

	set_bit(1, bitlock);
	smp_wmb();
	spin_unlock(&inode->i_lock);

	unmap_mapping_range(mapping, 0, 0, 0);
	ret = filemap_write_and_wait(mapping);
	if (!ret)
		ret = invalidate_inode_pages2(mapping);

	orangefs_inode->mapping_time = jiffies +
	    orangefs_cache_timeout_msecs*HZ/1000;

	clear_bit(1, bitlock);
	smp_mb__after_atomic();
	wake_up_bit(bitlock, 1);

	return ret;
}

static ssize_t orangefs_file_read_iter(struct kiocb *iocb,
    struct iov_iter *iter)
{
	int ret;
	orangefs_stats.reads++;

	down_read(&file_inode(iocb->ki_filp)->i_rwsem);
	ret = orangefs_revalidate_mapping(file_inode(iocb->ki_filp));
	if (ret)
		goto out;

	ret = generic_file_read_iter(iocb, iter);
out:
	up_read(&file_inode(iocb->ki_filp)->i_rwsem);
	return ret;
}

static ssize_t orangefs_file_splice_read(struct file *in, loff_t *ppos,
					 struct pipe_inode_info *pipe,
					 size_t len, unsigned int flags)
{
	struct inode *inode = file_inode(in);
	ssize_t ret;

	orangefs_stats.reads++;

	down_read(&inode->i_rwsem);
	ret = orangefs_revalidate_mapping(inode);
	if (ret)
		goto out;

	ret = filemap_splice_read(in, ppos, pipe, len, flags);
out:
	up_read(&inode->i_rwsem);
	return ret;
}

static ssize_t orangefs_file_write_iter(struct kiocb *iocb,
    struct iov_iter *iter)
{
	int ret;
	orangefs_stats.writes++;

	if (iocb->ki_pos > i_size_read(file_inode(iocb->ki_filp))) {
		ret = orangefs_revalidate_mapping(file_inode(iocb->ki_filp));
		if (ret)
			return ret;
	}

	ret = generic_file_write_iter(iocb, iter);
	return ret;
}

static vm_fault_t orangefs_fault(struct vm_fault *vmf)
{
	struct file *file = vmf->vma->vm_file;
	int ret;
	ret = orangefs_inode_getattr(file->f_mapping->host,
	    ORANGEFS_GETATTR_SIZE);
	if (ret == -ESTALE)
		ret = -EIO;
	if (ret) {
		gossip_err("%s: orangefs_inode_getattr failed, "
		    "ret:%d:.\n", __func__, ret);
		return VM_FAULT_SIGBUS;
	}
	return filemap_fault(vmf);
}

static const struct vm_operations_struct orangefs_file_vm_ops = {
	.fault = orangefs_fault,
	.map_pages = filemap_map_pages,
	.page_mkwrite = orangefs_page_mkwrite,
};

/*
 * Memory map a region of a file.
 */
static int orangefs_file_mmap(struct file *file, struct vm_area_struct *vma)
{
	int ret;

	ret = orangefs_revalidate_mapping(file_inode(file));
	if (ret)
		return ret;

	gossip_debug(GOSSIP_FILE_DEBUG,
		     "orangefs_file_mmap: called on %pD\n", file);

	/* set the sequential readahead hint */
	vm_flags_mod(vma, VM_SEQ_READ, VM_RAND_READ);

	file_accessed(file);
	vma->vm_ops = &orangefs_file_vm_ops;
	return 0;
}

#define mapping_nrpages(idata) ((idata)->nrpages)

/*
 * Called to notify the module that there are no more references to
 * this file (i.e. no processes have it open).
 *
 * \note Not called when each file is closed.
 */
static int orangefs_file_release(struct inode *inode, struct file *file)
{
	gossip_debug(GOSSIP_FILE_DEBUG,
		     "orangefs_file_release: called on %pD\n",
		     file);

	/*
	 * remove all associated inode pages from the page cache and
	 * readahead cache (if any); this forces an expensive refresh of
	 * data for the next caller of mmap (or 'get_block' accesses)
	 */
	if (mapping_nrpages(file->f_mapping)) {
		if (orangefs_features & ORANGEFS_FEATURE_READAHEAD) {
			gossip_debug(GOSSIP_INODE_DEBUG,
			    "calling flush_racache on %pU\n",
			    get_khandle_from_ino(inode));
			flush_racache(inode);
			gossip_debug(GOSSIP_INODE_DEBUG,
			    "flush_racache finished\n");
		}

	}
	return 0;
}

/*
 * Push all data for a specific file onto permanent storage.
 */
static int orangefs_fsync(struct file *file,
		       loff_t start,
		       loff_t end,
		       int datasync)
{
	int ret;
	struct orangefs_inode_s *orangefs_inode =
		ORANGEFS_I(file_inode(file));
	struct orangefs_kernel_op_s *new_op = NULL;

	ret = filemap_write_and_wait_range(file_inode(file)->i_mapping,
	    start, end);
	if (ret < 0)
		return ret;

	new_op = op_alloc(ORANGEFS_VFS_OP_FSYNC);
	if (!new_op)
		return -ENOMEM;
	new_op->upcall.req.fsync.refn = orangefs_inode->refn;

	ret = service_operation(new_op,
			"orangefs_fsync",
			get_interruptible_flag(file_inode(file)));

	gossip_debug(GOSSIP_FILE_DEBUG,
		     "orangefs_fsync got return value of %d\n",
		     ret);

	op_release(new_op);
	return ret;
}

/*
 * Change the file pointer position for an instance of an open file.
 *
 * \note If .llseek is overriden, we must acquire lock as described in
 *       Documentation/filesystems/locking.rst.
 *
 * Future upgrade could support SEEK_DATA and SEEK_HOLE but would
 * require much changes to the FS
 */
static loff_t orangefs_file_llseek(struct file *file, loff_t offset, int origin)
{
	int ret = -EINVAL;
	struct inode *inode = file_inode(file);

	if (origin == SEEK_END) {
		/*
		 * revalidate the inode's file size.
		 * NOTE: We are only interested in file size here,
		 * so we set mask accordingly.
		 */
		ret = orangefs_inode_getattr(file->f_mapping->host,
		    ORANGEFS_GETATTR_SIZE);
		if (ret == -ESTALE)
			ret = -EIO;
		if (ret) {
			gossip_debug(GOSSIP_FILE_DEBUG,
				     "%s:%s:%d calling make bad inode\n",
				     __FILE__,
				     __func__,
				     __LINE__);
			return ret;
		}
	}

	gossip_debug(GOSSIP_FILE_DEBUG,
		     "orangefs_file_llseek: offset is %ld | origin is %d"
		     " | inode size is %lu\n",
		     (long)offset,
		     origin,
		     (unsigned long)i_size_read(inode));

	return generic_file_llseek(file, offset, origin);
}

/*
 * Support local locks (locks that only this kernel knows about)
 * if Orangefs was mounted -o local_lock.
 */
static int orangefs_lock(struct file *filp, int cmd, struct file_lock *fl)
{
	int rc = -EINVAL;

	if (ORANGEFS_SB(file_inode(filp)->i_sb)->flags & ORANGEFS_OPT_LOCAL_LOCK) {
		if (cmd == F_GETLK) {
			rc = 0;
			posix_test_lock(filp, fl);
		} else {
			rc = posix_lock_file(filp, fl, NULL);
		}
	}

	return rc;
}

static int orangefs_flush(struct file *file, fl_owner_t id)
{
	/*
	 * This is vfs_fsync_range(file, 0, LLONG_MAX, 0) without the
	 * service_operation in orangefs_fsync.
	 *
	 * Do not send fsync to OrangeFS server on a close.  Do send fsync
	 * on an explicit fsync call.  This duplicates historical OrangeFS
	 * behavior.
	 */
	int r;

	r = filemap_write_and_wait_range(file->f_mapping, 0, LLONG_MAX);
	if (r > 0)
		return 0;
	else
		return r;
}

/** ORANGEFS implementation of VFS file operations */
const struct file_operations orangefs_file_operations = {
	.llseek		= orangefs_file_llseek,
	.read_iter	= orangefs_file_read_iter,
	.write_iter	= orangefs_file_write_iter,
	.lock		= orangefs_lock,
	.mmap		= orangefs_file_mmap,
	.open		= generic_file_open,
	.splice_read    = orangefs_file_splice_read,
	.splice_write   = iter_file_splice_write,
	.flush		= orangefs_flush,
	.release	= orangefs_file_release,
	.fsync		= orangefs_fsync,
};
Commit	Line	Data
	1	// SPDX-License-Identifier: GPL-2.0
	2	/*
	3	* (C) 2001 Clemson University and The University of Chicago
	4	* Copyright 2018 Omnibond Systems, L.L.C.
	5	*
	6	* See COPYING in top-level directory.
	7	*/
	8
	9	/*
	10	* Linux VFS file operations.
	11	*/
	12
	13	#include "protocol.h"
	14	#include "orangefs-kernel.h"
	15	#include "orangefs-bufmap.h"
	16	#include <linux/fs.h>
	17	#include <linux/filelock.h>
	18	#include <linux/pagemap.h>
	19
	20	static int flush_racache(struct inode *inode)
	21	{
	22	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	23	struct orangefs_kernel_op_s *new_op;
	24	int ret;
	25
	26	gossip_debug(GOSSIP_UTILS_DEBUG,
	27	"%s: %pU: Handle is %pU \| fs_id %d\n", __func__,
	28	get_khandle_from_ino(inode), &orangefs_inode->refn.khandle,
	29	orangefs_inode->refn.fs_id);
	30
	31	new_op = op_alloc(ORANGEFS_VFS_OP_RA_FLUSH);
	32	if (!new_op)
	33	return -ENOMEM;
	34	new_op->upcall.req.ra_cache_flush.refn = orangefs_inode->refn;
	35
	36	ret = service_operation(new_op, "orangefs_flush_racache",
	37	get_interruptible_flag(inode));
	38
	39	gossip_debug(GOSSIP_UTILS_DEBUG, "%s: got return value of %d\n",
	40	__func__, ret);
	41
	42	op_release(new_op);
	43	return ret;
	44	}
	45
	46	/*
	47	* Post and wait for the I/O upcall to finish
	48	*/
	49	ssize_t wait_for_direct_io(enum ORANGEFS_io_type type, struct inode *inode,
	50	loff_t offset, struct iov_iter iter, size_t total_size,
	51	loff_t readahead_size, struct orangefs_write_range *wr,
	52	int index_return, struct file file)
	53	{
	54	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	55	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
	56	struct orangefs_kernel_op_s *new_op = NULL;
	57	int buffer_index;
	58	ssize_t ret;
	59	size_t copy_amount;
	60	bool open_for_read;
	61	bool open_for_write;
	62
	63	new_op = op_alloc(ORANGEFS_VFS_OP_FILE_IO);
	64	if (!new_op)
	65	return -ENOMEM;
	66
	67	/* synchronous I/O */
	68	new_op->upcall.req.io.readahead_size = readahead_size;
	69	new_op->upcall.req.io.io_type = type;
	70	new_op->upcall.req.io.refn = orangefs_inode->refn;
	71
	72	populate_shared_memory:
	73	/* get a shared buffer index */
	74	buffer_index = orangefs_bufmap_get();
	75	if (buffer_index < 0) {
	76	ret = buffer_index;
	77	gossip_debug(GOSSIP_FILE_DEBUG,
	78	"%s: orangefs_bufmap_get failure (%zd)\n",
	79	__func__, ret);
	80	goto out;
	81	}
	82	gossip_debug(GOSSIP_FILE_DEBUG,
	83	"%s(%pU): GET op %p -> buffer_index %d\n",
	84	__func__,
	85	handle,
	86	new_op,
	87	buffer_index);
	88
	89	new_op->uses_shared_memory = 1;
	90	new_op->upcall.req.io.buf_index = buffer_index;
	91	new_op->upcall.req.io.count = total_size;
	92	new_op->upcall.req.io.offset = *offset;
	93	if (type == ORANGEFS_IO_WRITE && wr) {
	94	new_op->upcall.uid = from_kuid(&init_user_ns, wr->uid);
	95	new_op->upcall.gid = from_kgid(&init_user_ns, wr->gid);
	96	}
	97	/*
	98	* Orangefs has no open, and orangefs checks file permissions
	99	* on each file access. Posix requires that file permissions
	100	* be checked on open and nowhere else. Orangefs-through-the-kernel
	101	* needs to seem posix compliant.
	102	*
	103	* The VFS opens files, even if the filesystem provides no
	104	* method. We can see if a file was successfully opened for
	105	* read and or for write by looking at file->f_mode.
	106	*
	107	* When writes are flowing from the page cache, file is no
	108	* longer available. We can trust the VFS to have checked
	109	* file->f_mode before writing to the page cache.
	110	*
	111	* The mode of a file might change between when it is opened
	112	* and IO commences, or it might be created with an arbitrary mode.
	113	*
	114	* We'll make sure we don't hit EACCES during the IO stage by
	115	* using UID 0. Some of the time we have access without changing
	116	* to UID 0 - how to check?
	117	*/
	118	if (file) {
	119	open_for_write = file->f_mode & FMODE_WRITE;
	120	open_for_read = file->f_mode & FMODE_READ;
	121	} else {
	122	open_for_write = 1;
	123	open_for_read = 0; /* not relevant? */
	124	}
	125	if ((type == ORANGEFS_IO_WRITE) && open_for_write)
	126	new_op->upcall.uid = 0;
	127	if ((type == ORANGEFS_IO_READ) && open_for_read)
	128	new_op->upcall.uid = 0;
	129
	130	gossip_debug(GOSSIP_FILE_DEBUG,
	131	"%s(%pU): offset: %llu total_size: %zd\n",
	132	__func__,
	133	handle,
	134	llu(*offset),
	135	total_size);
	136	/*
	137	* Stage 1: copy the buffers into client-core's address space
	138	*/
	139	if (type == ORANGEFS_IO_WRITE && total_size) {
	140	ret = orangefs_bufmap_copy_from_iovec(iter, buffer_index,
	141	total_size);
	142	if (ret < 0) {
	143	gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
	144	__func__, (long)ret);
	145	goto out;
	146	}
	147	}
	148
	149	gossip_debug(GOSSIP_FILE_DEBUG,
	150	"%s(%pU): Calling post_io_request with tag (%llu)\n",
	151	__func__,
	152	handle,
	153	llu(new_op->tag));
	154
	155	/* Stage 2: Service the I/O operation */
	156	ret = service_operation(new_op,
	157	type == ORANGEFS_IO_WRITE ?
	158	"file_write" :
	159	"file_read",
	160	get_interruptible_flag(inode));
	161
	162	/*
	163	* If service_operation() returns -EAGAIN #and# the operation was
	164	* purged from orangefs_request_list or htable_ops_in_progress, then
	165	* we know that the client was restarted, causing the shared memory
	166	* area to be wiped clean. To restart a write operation in this
	167	* case, we must re-copy the data from the user's iovec to a NEW
	168	* shared memory location. To restart a read operation, we must get
	169	* a new shared memory location.
	170	*/
	171	if (ret == -EAGAIN && op_state_purged(new_op)) {
	172	orangefs_bufmap_put(buffer_index);
	173	if (type == ORANGEFS_IO_WRITE)
	174	iov_iter_revert(iter, total_size);
	175	gossip_debug(GOSSIP_FILE_DEBUG,
	176	"%s:going to repopulate_shared_memory.\n",
	177	__func__);
	178	goto populate_shared_memory;
	179	}
	180
	181	if (ret < 0) {
	182	if (ret == -EINTR) {
	183	/*
	184	* We can't return EINTR if any data was written,
	185	* it's not POSIX. It is minimally acceptable
	186	* to give a partial write, the way NFS does.
	187	*
	188	* It would be optimal to return all or nothing,
	189	* but if a userspace write is bigger than
	190	* an IO buffer, and the interrupt occurs
	191	* between buffer writes, that would not be
	192	* possible.
	193	*/
	194	switch (new_op->op_state - OP_VFS_STATE_GIVEN_UP) {
	195	/*
	196	* If the op was waiting when the interrupt
	197	* occurred, then the client-core did not
	198	* trigger the write.
	199	*/
	200	case OP_VFS_STATE_WAITING:
	201	if (*offset == 0)
	202	ret = -EINTR;
	203	else
	204	ret = 0;
	205	break;
	206	/*
	207	* If the op was in progress when the interrupt
	208	* occurred, then the client-core was able to
	209	* trigger the write.
	210	*/
	211	case OP_VFS_STATE_INPROGR:
	212	if (type == ORANGEFS_IO_READ)
	213	ret = -EINTR;
	214	else
	215	ret = total_size;
	216	break;
	217	default:
	218	gossip_err("%s: unexpected op state :%d:.\n",
	219	__func__,
	220	new_op->op_state);
	221	ret = 0;
	222	break;
	223	}
	224	gossip_debug(GOSSIP_FILE_DEBUG,
	225	"%s: got EINTR, state:%d: %p\n",
	226	__func__,
	227	new_op->op_state,
	228	new_op);
	229	} else {
	230	gossip_err("%s: error in %s handle %pU, returning %zd\n",
	231	__func__,
	232	type == ORANGEFS_IO_READ ?
	233	"read from" : "write to",
	234	handle, ret);
	235	}
	236	if (orangefs_cancel_op_in_progress(new_op))
	237	return ret;
	238
	239	goto out;
	240	}
	241
	242	/*
	243	* Stage 3: Post copy buffers from client-core's address space
	244	*/
	245	if (type == ORANGEFS_IO_READ && new_op->downcall.resp.io.amt_complete) {
	246	/*
	247	* NOTE: the iovector can either contain addresses which
	248	* can futher be kernel-space or user-space addresses.
	249	* or it can pointers to struct page's
	250	*/
	251
	252	copy_amount = new_op->downcall.resp.io.amt_complete;
	253
	254	ret = orangefs_bufmap_copy_to_iovec(iter, buffer_index,
	255	copy_amount);
	256	if (ret < 0) {
	257	gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n",
	258	__func__, (long)ret);
	259	goto out;
	260	}
	261	}
	262	gossip_debug(GOSSIP_FILE_DEBUG,
	263	"%s(%pU): Amount %s, returned by the sys-io call:%d\n",
	264	__func__,
	265	handle,
	266	type == ORANGEFS_IO_READ ? "read" : "written",
	267	(int)new_op->downcall.resp.io.amt_complete);
	268
	269	ret = new_op->downcall.resp.io.amt_complete;
	270
	271	out:
	272	if (buffer_index >= 0) {
	273	orangefs_bufmap_put(buffer_index);
	274	gossip_debug(GOSSIP_FILE_DEBUG,
	275	"%s(%pU): PUT buffer_index %d\n",
	276	__func__, handle, buffer_index);
	277	}
	278	op_release(new_op);
	279	return ret;
	280	}
	281
	282	int orangefs_revalidate_mapping(struct inode *inode)
	283	{
	284	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
	285	struct address_space *mapping = inode->i_mapping;
	286	unsigned long *bitlock = &orangefs_inode->bitlock;
	287	int ret;
	288
	289	while (1) {
	290	ret = wait_on_bit(bitlock, 1, TASK_KILLABLE);
	291	if (ret)
	292	return ret;
	293	spin_lock(&inode->i_lock);
	294	if (test_bit(1, bitlock)) {
	295	spin_unlock(&inode->i_lock);
	296	continue;
	297	}
	298	if (!time_before(jiffies, orangefs_inode->mapping_time))
	299	break;
	300	spin_unlock(&inode->i_lock);
	301	return 0;
	302	}
	303
	304	set_bit(1, bitlock);
	305	smp_wmb();
	306	spin_unlock(&inode->i_lock);
	307
	308	unmap_mapping_range(mapping, 0, 0, 0);
	309	ret = filemap_write_and_wait(mapping);
	310	if (!ret)
	311	ret = invalidate_inode_pages2(mapping);
	312
	313	orangefs_inode->mapping_time = jiffies +
	314	orangefs_cache_timeout_msecs*HZ/1000;
	315
	316	clear_bit(1, bitlock);
	317	smp_mb__after_atomic();
	318	wake_up_bit(bitlock, 1);
	319
	320	return ret;
	321	}
	322
	323	static ssize_t orangefs_file_read_iter(struct kiocb *iocb,
	324	struct iov_iter *iter)
	325	{
	326	int ret;
	327	orangefs_stats.reads++;
	328
	329	down_read(&file_inode(iocb->ki_filp)->i_rwsem);
	330	ret = orangefs_revalidate_mapping(file_inode(iocb->ki_filp));
	331	if (ret)
	332	goto out;
	333
	334	ret = generic_file_read_iter(iocb, iter);
	335	out:
	336	up_read(&file_inode(iocb->ki_filp)->i_rwsem);
	337	return ret;
	338	}
	339
	340	static ssize_t orangefs_file_splice_read(struct file in, loff_t ppos,
	341	struct pipe_inode_info *pipe,
	342	size_t len, unsigned int flags)
	343	{
	344	struct inode *inode = file_inode(in);
	345	ssize_t ret;
	346
	347	orangefs_stats.reads++;
	348
	349	down_read(&inode->i_rwsem);
	350	ret = orangefs_revalidate_mapping(inode);
	351	if (ret)
	352	goto out;
	353
	354	ret = filemap_splice_read(in, ppos, pipe, len, flags);
	355	out:
	356	up_read(&inode->i_rwsem);
	357	return ret;
	358	}
	359
	360	static ssize_t orangefs_file_write_iter(struct kiocb *iocb,
	361	struct iov_iter *iter)
	362	{
	363	int ret;
	364	orangefs_stats.writes++;
	365
	366	if (iocb->ki_pos > i_size_read(file_inode(iocb->ki_filp))) {
	367	ret = orangefs_revalidate_mapping(file_inode(iocb->ki_filp));
	368	if (ret)
	369	return ret;
	370	}
	371
	372	ret = generic_file_write_iter(iocb, iter);
	373	return ret;
	374	}
	375
	376	static vm_fault_t orangefs_fault(struct vm_fault *vmf)
	377	{
	378	struct file *file = vmf->vma->vm_file;
	379	int ret;
	380	ret = orangefs_inode_getattr(file->f_mapping->host,
	381	ORANGEFS_GETATTR_SIZE);
	382	if (ret == -ESTALE)
	383	ret = -EIO;
	384	if (ret) {
	385	gossip_err("%s: orangefs_inode_getattr failed, "
	386	"ret:%d:.\n", __func__, ret);
	387	return VM_FAULT_SIGBUS;
	388	}
	389	return filemap_fault(vmf);
	390	}
	391
	392	static const struct vm_operations_struct orangefs_file_vm_ops = {
	393	.fault = orangefs_fault,
	394	.map_pages = filemap_map_pages,
	395	.page_mkwrite = orangefs_page_mkwrite,
	396	};
	397
	398	/*
	399	* Memory map a region of a file.
	400	*/
	401	static int orangefs_file_mmap(struct file file, struct vm_area_struct vma)
	402	{
	403	int ret;
	404
	405	ret = orangefs_revalidate_mapping(file_inode(file));
	406	if (ret)
	407	return ret;
	408
	409	gossip_debug(GOSSIP_FILE_DEBUG,
	410	"orangefs_file_mmap: called on %pD\n", file);
	411
	412	/* set the sequential readahead hint */
	413	vm_flags_mod(vma, VM_SEQ_READ, VM_RAND_READ);
	414
	415	file_accessed(file);
	416	vma->vm_ops = &orangefs_file_vm_ops;
	417	return 0;
	418	}
	419
	420	#define mapping_nrpages(idata) ((idata)->nrpages)
	421
	422	/*
	423	* Called to notify the module that there are no more references to
	424	* this file (i.e. no processes have it open).
	425	*
	426	* \note Not called when each file is closed.
	427	*/
	428	static int orangefs_file_release(struct inode inode, struct file file)
	429	{
	430	gossip_debug(GOSSIP_FILE_DEBUG,
	431	"orangefs_file_release: called on %pD\n",
	432	file);
	433
	434	/*
	435	* remove all associated inode pages from the page cache and
	436	* readahead cache (if any); this forces an expensive refresh of
	437	* data for the next caller of mmap (or 'get_block' accesses)
	438	*/
	439	if (mapping_nrpages(file->f_mapping)) {
	440	if (orangefs_features & ORANGEFS_FEATURE_READAHEAD) {
	441	gossip_debug(GOSSIP_INODE_DEBUG,
	442	"calling flush_racache on %pU\n",
	443	get_khandle_from_ino(inode));
	444	flush_racache(inode);
	445	gossip_debug(GOSSIP_INODE_DEBUG,
	446	"flush_racache finished\n");
	447	}
	448
	449	}
	450	return 0;
	451	}
	452
	453	/*
	454	* Push all data for a specific file onto permanent storage.
	455	*/
	456	static int orangefs_fsync(struct file *file,
	457	loff_t start,
	458	loff_t end,
	459	int datasync)
	460	{
	461	int ret;
	462	struct orangefs_inode_s *orangefs_inode =
	463	ORANGEFS_I(file_inode(file));
	464	struct orangefs_kernel_op_s *new_op = NULL;
	465
	466	ret = filemap_write_and_wait_range(file_inode(file)->i_mapping,
	467	start, end);
	468	if (ret < 0)
	469	return ret;
	470
	471	new_op = op_alloc(ORANGEFS_VFS_OP_FSYNC);
	472	if (!new_op)
	473	return -ENOMEM;
	474	new_op->upcall.req.fsync.refn = orangefs_inode->refn;
	475
	476	ret = service_operation(new_op,
	477	"orangefs_fsync",
	478	get_interruptible_flag(file_inode(file)));
	479
	480	gossip_debug(GOSSIP_FILE_DEBUG,
	481	"orangefs_fsync got return value of %d\n",
	482	ret);
	483
	484	op_release(new_op);
	485	return ret;
	486	}
	487
	488	/*
	489	* Change the file pointer position for an instance of an open file.
	490	*
	491	* \note If .llseek is overriden, we must acquire lock as described in
	492	* Documentation/filesystems/locking.rst.
	493	*
	494	* Future upgrade could support SEEK_DATA and SEEK_HOLE but would
	495	* require much changes to the FS
	496	*/
	497	static loff_t orangefs_file_llseek(struct file *file, loff_t offset, int origin)
	498	{
	499	int ret = -EINVAL;
	500	struct inode *inode = file_inode(file);
	501
	502	if (origin == SEEK_END) {
	503	/*
	504	* revalidate the inode's file size.
	505	* NOTE: We are only interested in file size here,
	506	* so we set mask accordingly.
	507	*/
	508	ret = orangefs_inode_getattr(file->f_mapping->host,
	509	ORANGEFS_GETATTR_SIZE);
	510	if (ret == -ESTALE)
	511	ret = -EIO;
	512	if (ret) {
	513	gossip_debug(GOSSIP_FILE_DEBUG,
	514	"%s:%s:%d calling make bad inode\n",
	515	__FILE__,
	516	__func__,
	517	__LINE__);
	518	return ret;
	519	}
	520	}
	521
	522	gossip_debug(GOSSIP_FILE_DEBUG,
	523	"orangefs_file_llseek: offset is %ld \| origin is %d"
	524	" \| inode size is %lu\n",
	525	(long)offset,
	526	origin,
	527	(unsigned long)i_size_read(inode));
	528
	529	return generic_file_llseek(file, offset, origin);
	530	}
	531
	532	/*
	533	* Support local locks (locks that only this kernel knows about)
	534	* if Orangefs was mounted -o local_lock.
	535	*/
	536	static int orangefs_lock(struct file filp, int cmd, struct file_lock fl)
	537	{
	538	int rc = -EINVAL;
	539
	540	if (ORANGEFS_SB(file_inode(filp)->i_sb)->flags & ORANGEFS_OPT_LOCAL_LOCK) {
	541	if (cmd == F_GETLK) {
	542	rc = 0;
	543	posix_test_lock(filp, fl);
	544	} else {
	545	rc = posix_lock_file(filp, fl, NULL);
	546	}
	547	}
	548
	549	return rc;
	550	}
	551
	552	static int orangefs_flush(struct file *file, fl_owner_t id)
	553	{
	554	/*
	555	* This is vfs_fsync_range(file, 0, LLONG_MAX, 0) without the
	556	* service_operation in orangefs_fsync.
	557	*
	558	* Do not send fsync to OrangeFS server on a close. Do send fsync
	559	* on an explicit fsync call. This duplicates historical OrangeFS
	560	* behavior.
	561	*/
	562	int r;
	563
	564	r = filemap_write_and_wait_range(file->f_mapping, 0, LLONG_MAX);
	565	if (r > 0)
	566	return 0;
	567	else
	568	return r;
	569	}
	570
	571	/** ORANGEFS implementation of VFS file operations */
	572	const struct file_operations orangefs_file_operations = {
	573	.llseek = orangefs_file_llseek,
	574	.read_iter = orangefs_file_read_iter,
	575	.write_iter = orangefs_file_write_iter,
	576	.lock = orangefs_lock,
	577	.mmap = orangefs_file_mmap,
	578	.open = generic_file_open,
	579	.splice_read = orangefs_file_splice_read,
	580	.splice_write = iter_file_splice_write,
	581	.flush = orangefs_flush,
	582	.release = orangefs_file_release,
	583	.fsync = orangefs_fsync,
	584	};