Hi all,
Attached patch cleans up a long overdue (my bad) need for consolidating
the I/O paths for read, write and readv, writev. The aio read and write
parts also need consolidation into a common routine which is not done in
this patch. I figured I might as well do this now since readx, writex
system calls also require similar consolidation.
Note: This does not (yet) consolidate readv, writev paths to make use of
the read, write path.
Comments welcome,
Thanks,
Murali
Index: src/apps/admin/pvfs2-ping.c
===================================================================
RCS file: /anoncvs/pvfs2/src/apps/admin/pvfs2-ping.c,v
retrieving revision 1.47
diff -u -r1.47 pvfs2-ping.c
--- src/apps/admin/pvfs2-ping.c 25 May 2006 18:14:14 -0000 1.47
+++ src/apps/admin/pvfs2-ping.c 26 May 2006 02:59:13 -0000
@@ -223,9 +223,12 @@
PVFS_error_details_free(error_details);
- /* if we hit this point, then everything is ok */
- printf(" Ok; root handle is owned by exactly one server.\n");
- printf("\n");
+ if (!err)
+ {
+ /* if we hit this point, then everything is ok */
+ printf(" Ok; root handle is owned by exactly one server.\n");
+ printf("\n");
+ }
PVFS_sys_finalize();
Index: src/kernel/linux-2.6/file.c
===================================================================
RCS file: /anoncvs/pvfs2/src/kernel/linux-2.6/file.c,v
retrieving revision 1.114
diff -u -r1.114 file.c
--- src/kernel/linux-2.6/file.c 4 Apr 2006 15:04:28 -0000 1.114
+++ src/kernel/linux-2.6/file.c 26 May 2006 02:59:20 -0000
@@ -17,6 +17,13 @@
#include <linux/fs.h>
#include <linux/pagemap.h>
+enum {
+ IO_READ = 0,
+ IO_WRITE = 1,
+ IO_READV = 0,
+ IO_WRITEV = 1,
+};
+
extern struct list_head pvfs2_request_list;
extern spinlock_t pvfs2_request_list_lock;
extern wait_queue_head_t pvfs2_request_list_waitq;
@@ -101,53 +108,127 @@
return ret;
}
-/** Read data from a specified offset in a file (referenced by inode).
- * Data may be placed either in a user or kernel buffer.
- */
-ssize_t pvfs2_inode_read(
- struct inode *inode,
- char __user *buf,
- size_t count,
- loff_t *offset,
- int copy_to_user,
- loff_t readahead_size)
+struct rw_options {
+ int type;
+ /* sigh.. we will never pass sparse type checks.. */
+ char *buf;
+ size_t count;
+ loff_t *offset;
+ union {
+ struct {
+ struct inode *inode;
+ int copy_to_user;
+ loff_t readahead_size;
+ } read;
+ struct {
+ struct file *file;
+ } write;
+ } io;
+};
+
+static ssize_t do_read_write(struct rw_options *rw)
{
- int ret = -1;
- size_t each_count = 0, amt_complete = 0;
- size_t total_count = 0;
pvfs2_kernel_op_t *new_op = NULL;
int buffer_index = -1;
- char *current_buf = buf;
- loff_t original_offset = *offset;
- pvfs2_inode_t *pvfs2_inode = PVFS2_I(inode);
+ struct inode *inode;
+ pvfs2_inode_t *pvfs2_inode = NULL;
+ char *current_buf = NULL;
+ size_t count;
+ loff_t *offset;
+ ssize_t ret;
+ ssize_t total_count;
+ char *fnstr = NULL;
+ size_t readahead_size;
+ int copy_to_user;
+ struct file *file;
- if (copy_to_user && (!access_ok(VERIFY_WRITE, buf, count)))
- return -EFAULT;
+ total_count = 0;
+ ret = -EINVAL;
+ file = NULL;
+ inode = NULL;
+ if (!rw)
+ goto out;
+ count = rw->count;
+ current_buf = (char *) rw->buf;
+ if (!current_buf)
+ goto out;
+ offset = rw->offset;
+ if (!offset)
+ goto out;
+ if (rw->type == IO_READ)
+ {
+ inode = rw->io.read.inode;
+ if (!inode)
+ goto out;
+ file = NULL;
+ copy_to_user = rw->io.read.copy_to_user;
+ ret = -EFAULT;
+ if (copy_to_user &&
+ !access_ok(VERIFY_WRITE, (char __user *) current_buf, count))
+ goto out;
+ fnstr = "pvfs2_file_read";
+ readahead_size = rw->io.read.readahead_size;
+ }
+ else
+ {
+ file = rw->io.write.file;
+ copy_to_user = 1;
+ readahead_size = 0;
+ if (!file)
+ goto out;
+ inode = file->f_dentry->d_inode;
+ if (!inode)
+ goto out;
+ fnstr = "pvfs2_file_write";
+ ret = -EFAULT;
+ if (!access_ok(VERIFY_READ, (char __user *) current_buf, count))
+ goto out;
+ if(file->f_pos > i_size_read(inode))
+ {
+ i_size_write(inode, file->f_pos);
+ }
+ /* perform generic linux kernel tests for sanity of write arguments */
+ /* NOTE: this is particularly helpful in handling fsize rlimit
properly */
+#ifdef PVFS2_LINUX_KERNEL_2_4
+ ret = pvfs2_precheck_file_write(file, inode, &count, offset);
+#else
+ ret = generic_write_checks(file, offset, &count,
S_ISBLK(inode->i_mode));
+#endif
+ if (ret != 0 || count == 0)
+ {
+ pvfs2_print("pvfs2_file_write: failed generic argument checks.\n");
+ goto out;
+ }
+ pvfs2_print("%s: proceeding with offset : %ld, size %ld\n",
+ fnstr, (unsigned long) *offset, (unsigned long) count);
+ }
+ pvfs2_inode = PVFS2_I(inode);
while(total_count < count)
{
+ size_t each_count, amt_complete;
+
new_op = op_alloc();
if (!new_op)
{
- return -ENOMEM;
+ ret = -ENOMEM;
+ goto out;
}
new_op->upcall.type = PVFS2_VFS_OP_FILE_IO;
new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO; /* synchronous
I/O */
new_op->upcall.req.io.readahead_size = readahead_size;
- new_op->upcall.req.io.io_type = PVFS_IO_READ;
+ new_op->upcall.req.io.io_type =
+ (rw->type == IO_READ) ? PVFS_IO_READ : PVFS_IO_WRITE;
new_op->upcall.req.io.refn = pvfs2_inode->refn;
ret = pvfs_bufmap_get(&buffer_index);
if (ret < 0)
{
- pvfs2_error("pvfs2_inode_read: pvfs_bufmap_get() "
- "failure (%d)\n", ret);
- op_release(new_op);
- *offset = original_offset;
- return ret;
+ pvfs2_error("do_read_write: pvfs_bufmap_get() "
+ "failure (%ld)\n", (long) ret);
+ goto out;
}
-
/* how much to transfer in this loop iteration */
each_count = (((count - total_count) > pvfs_bufmap_size_query()) ?
pvfs_bufmap_size_query() : (count - total_count));
@@ -155,9 +236,18 @@
new_op->upcall.req.io.buf_index = buffer_index;
new_op->upcall.req.io.count = each_count;
new_op->upcall.req.io.offset = *offset;
-
+ if (rw->type == IO_WRITE)
+ {
+ /* copy data from application */
+ ret = pvfs_bufmap_copy_from_user(buffer_index, current_buf,
each_count);
+ if(ret < 0)
+ {
+ pvfs2_print("%s: Failed to copy user buffer.\n", fnstr);
+ goto out;
+ }
+ }
ret = service_operation(
- new_op, "pvfs2_inode_read", PVFS2_OP_RETRY_COUNT,
+ new_op, fnstr, PVFS2_OP_RETRY_COUNT,
get_interruptible_flag(inode));
if (ret < 0)
@@ -170,53 +260,54 @@
termination unless we've got debugging turned on, as
this can happen regularly (i.e. ctrl-c)
*/
- if(ret == -EINTR)
+ if (ret == -EINTR)
{
- pvfs2_print("pvfs2_inode_read: returning error %d\n", ret);
+ pvfs2_print("%s: returning error %ld\n", fnstr, (long) ret);
}
else
{
pvfs2_error(
- "pvfs2_inode_read: error reading from handle %llu, "
- "\n -- returning %d \n",
- llu(pvfs2_ino_to_handle(inode->i_ino)), ret);
+ "%s: error writing to handle %llu, "
+ "-- returning %ld\n",
+ fnstr,
+ llu(pvfs2_ino_to_handle(inode->i_ino)),
+ (long) ret);
}
- return ret;
+ goto out;
}
-
- /* copy data out to destination */
- if (new_op->downcall.resp.io.amt_complete)
+ if (rw->type == IO_READ)
{
- if (copy_to_user)
- {
- ret = pvfs_bufmap_copy_to_user(
- current_buf, buffer_index,
- new_op->downcall.resp.io.amt_complete);
- }
- else
+ /* copy data out to destination */
+ if (new_op->downcall.resp.io.amt_complete)
{
- ret = pvfs_bufmap_copy_to_kernel(
- current_buf, buffer_index,
- new_op->downcall.resp.io.amt_complete);
- }
-
- if (ret)
- {
- pvfs2_print("Failed to copy user buffer.\n");
- /* put error code in downcall so that handle_io_error()
- * preserves properly
- */
- new_op->downcall.status = ret;
- handle_io_error();
- return(ret);
+ if (copy_to_user)
+ {
+ ret = pvfs_bufmap_copy_to_user(
+ current_buf, buffer_index,
+ new_op->downcall.resp.io.amt_complete);
+ }
+ else
+ {
+ ret = pvfs_bufmap_copy_to_kernel(
+ current_buf, buffer_index,
+ new_op->downcall.resp.io.amt_complete);
+ }
+ if (ret)
+ {
+ pvfs2_print("Failed to copy user buffer.\n");
+ /* put error code in downcall so that handle_io_error()
+ * preserves properly
+ */
+ new_op->downcall.status = ret;
+ handle_io_error();
+ goto out;
+ }
}
}
-
current_buf += new_op->downcall.resp.io.amt_complete;
*offset += new_op->downcall.resp.io.amt_complete;
total_count += new_op->downcall.resp.io.amt_complete;
amt_complete = new_op->downcall.resp.io.amt_complete;
-
/*
tell the device file owner waiting on I/O that this read has
completed and it can return now. in this exact case, on
@@ -224,9 +315,10 @@
after this.
*/
wake_up_device_for_return(new_op);
+ new_op = NULL;
pvfs_bufmap_put(buffer_index);
-
- /* if we got a short read, fall out and return what we
+ buffer_index = -1;
+ /* if we got a short read/write, fall out and return what we
* got so far
*/
if (amt_complete < each_count)
@@ -234,12 +326,45 @@
break;
}
}
+ if (total_count > 0) {
+ ret = total_count;
+ }
+out:
+ if (new_op)
+ op_release(new_op);
+ if (buffer_index >= 0)
+ pvfs_bufmap_put(buffer_index);
+ if (ret > 0 && file != NULL && inode != NULL)
+ {
+#ifdef HAVE_TOUCH_ATIME
+ touch_atime(file->f_vfsmnt, file->f_dentry);
+#else
+ update_atime(inode);
+#endif
+ }
+ return ret;
+}
- /*
- NOTE: for this special case, op is freed by devreq_writev and
- *not* here.
- */
- return(total_count);
+/** Read data from a specified offset in a file (referenced by inode).
+ * Data may be placed either in a user or kernel buffer.
+ */
+ssize_t pvfs2_inode_read(
+ struct inode *inode,
+ char __user *buf,
+ size_t count,
+ loff_t *offset,
+ int copy_to_user,
+ loff_t readahead_size)
+{
+ struct rw_options rw;
+ rw.type = IO_READ;
+ rw.buf = buf;
+ rw.count = count;
+ rw.offset = offset;
+ rw.io.read.inode = inode;
+ rw.io.read.copy_to_user = copy_to_user;
+ rw.io.read.readahead_size = readahead_size;
+ return do_read_write(&rw);
}
/** Read data from a specified offset in a file into a user buffer.
@@ -268,160 +393,13 @@
size_t count,
loff_t *offset)
{
- int ret = -1;
- pvfs2_kernel_op_t *new_op = NULL;
- char __user *current_buf = (char __user *)buf;
- loff_t original_offset = *offset;
- int buffer_index = -1;
- size_t each_count = 0, total_count = 0;
- struct inode *inode = file->f_dentry->d_inode;
- pvfs2_inode_t *pvfs2_inode = PVFS2_I(inode);
- size_t amt_complete = 0;
-
- pvfs2_print("pvfs2_file_write: called on %s [f_pos %ld off %ld size
%ld]\n",
- (file && file->f_dentry && file->f_dentry->d_name.name ?
- (char *)file->f_dentry->d_name.name : "UNKNOWN"),
- (unsigned long) file->f_pos,
- (unsigned long) *offset, (unsigned long) count);
-
- if (!access_ok(VERIFY_READ, buf, count))
- return -EFAULT;
-
- if(file->f_pos > i_size_read(inode))
- {
- i_size_write(inode, file->f_pos);
- }
-
- /* perform generic linux kernel tests for sanity of write arguments */
- /* NOTE: this is particularly helpful in handling fsize rlimit properly */
-#ifdef PVFS2_LINUX_KERNEL_2_4
- ret = pvfs2_precheck_file_write(file, inode, &count, offset);
-#else
- ret = generic_write_checks(file, offset, &count, S_ISBLK(inode->i_mode));
-#endif
- if (ret != 0 || count == 0)
- {
- pvfs2_print("pvfs2_file_write: failed generic argument checks.\n");
- return(ret);
- }
-
- pvfs2_print("pvfs2_file_write: proceeding with offset : %ld, size %ld\n",
- (unsigned long) *offset, (unsigned long) count);
-
- while(total_count < count)
- {
- new_op = op_alloc();
- if (!new_op)
- {
- *offset = original_offset;
- return -ENOMEM;
- }
-
- new_op->upcall.type = PVFS2_VFS_OP_FILE_IO;
- new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO; /* synchronous
I/O */
- new_op->upcall.req.io.io_type = PVFS_IO_WRITE;
- new_op->upcall.req.io.refn = pvfs2_inode->refn;
-
- pvfs2_print("pvfs2_file_write: writing %d bytes at offset %lu (%lu)\n",
- (int)count, (unsigned long)file->f_pos, (unsigned
long)*offset);
-
- ret = pvfs_bufmap_get(&buffer_index);
- if (ret < 0)
- {
- pvfs2_error("pvfs2_file_write: pvfs_bufmap_get() "
- "failure (%d)\n", ret);
- op_release(new_op);
- *offset = original_offset;
- return ret;
- }
-
- /* how much to transfer in this loop iteration */
- each_count = (((count - total_count) > pvfs_bufmap_size_query()) ?
- pvfs_bufmap_size_query() : (count - total_count));
-
- new_op->upcall.req.io.buf_index = buffer_index;
- new_op->upcall.req.io.count = each_count;
- new_op->upcall.req.io.offset = *offset;
-
- /* copy data from application */
- ret = pvfs_bufmap_copy_from_user(
- buffer_index, current_buf, each_count);
- if(ret < 0)
- {
- pvfs2_print("Failed to copy user buffer.\n");
- op_release(new_op);
- pvfs_bufmap_put(buffer_index);
- *offset = original_offset;
- return ret;
- }
-
- ret = service_operation(
- new_op, "pvfs2_file_write", PVFS2_OP_RETRY_COUNT,
- get_interruptible_flag(inode));
-
- if (ret < 0)
- {
- /* this macro is defined in pvfs2-kernel.h */
- handle_io_error();
-
- /*
- don't write an error to syslog on signaled operation
- termination unless we've got debugging turned on, as
- this can happen regularly (i.e. ctrl-c)
- */
- if (ret == -EINTR)
- {
- pvfs2_print("pvfs2_file_write: returning error %d\n", ret);
- }
- else
- {
- pvfs2_error(
- "pvfs2_file_write: error writing to handle %llu, "
- "FILE: %s\n -- returning %d\n",
- llu(pvfs2_ino_to_handle(inode->i_ino)),
- (file && file->f_dentry && file->f_dentry->d_name.name ?
- (char *)file->f_dentry->d_name.name : "UNKNOWN"),
- ret);
- }
- *offset = original_offset;
- return ret;
- }
-
- current_buf += new_op->downcall.resp.io.amt_complete;
- pvfs2_print("amt_complete = %ld\n", (unsigned long)
new_op->downcall.resp.io.amt_complete);
- *offset += new_op->downcall.resp.io.amt_complete;
- total_count += new_op->downcall.resp.io.amt_complete;
- amt_complete = new_op->downcall.resp.io.amt_complete;
-
- /*
- tell the device file owner waiting on I/O that this read has
- completed and it can return now. in this exact case, on
- wakeup the device will free the op, so we *cannot* touch it
- after this.
- */
- wake_up_device_for_return(new_op);
- pvfs_bufmap_put(buffer_index);
-
- /* if we got a short write, fall out and return what we got so
- * far TODO: define semantics here- kind of depends on pvfs2
- * semantics that don't really exist yet
- */
- if (amt_complete < each_count)
- {
- break;
- }
- }
- pvfs2_print("pvfs2_file_write: pos at the end was %lu(%lu)\n",
- (unsigned long) *offset, (unsigned long) file->f_pos);
- if (total_count)
- {
-#ifdef HAVE_TOUCH_ATIME
- touch_atime(file->f_vfsmnt, file->f_dentry);
-#else
- update_atime(inode);
-#endif
- }
- return total_count;
+ struct rw_options rw;
+ rw.type = IO_WRITE;
+ rw.buf = (char *) buf;
+ rw.count = count;
+ rw.offset = offset;
+ rw.io.write.file = file;
+ return do_read_write(&rw);
}
/*
@@ -563,37 +541,38 @@
}
-/** Reads data to several contiguous user buffers (an iovec) from a file at a
- * specified offset.
- */
-static ssize_t pvfs2_file_readv(
- struct file *file,
- const struct iovec *iov,
- unsigned long nr_segs,
- loff_t *offset)
+static ssize_t do_readv_writev(int type, struct file *file,
+ const struct iovec *iov, unsigned long nr_segs, loff_t *offset)
{
- int ret = -1;
- pvfs2_kernel_op_t *new_op = NULL;
- struct iovec *iovecptr = NULL, *ptr = NULL;
- loff_t original_offset = *offset;
- int buffer_index = -1;
+ ssize_t ret;
+ unsigned int to_free;
+ unsigned long seg;
+ ssize_t total_count, count;
+ size_t each_count;
struct inode *inode = file->f_dentry->d_inode;
pvfs2_inode_t *pvfs2_inode = PVFS2_I(inode);
- size_t amt_complete = 0;
- size_t total_count = 0, count = 0, each_count = 0;
- unsigned int seg, to_free = 0;
unsigned long new_nr_segs = 0, max_new_nr_segs = 0;
- unsigned int seg_count, *seg_array = NULL;
-
+ unsigned int seg_count = 0, *seg_array = NULL;
+ struct iovec *iovecptr = NULL, *ptr = NULL;
+ pvfs2_kernel_op_t *new_op = NULL;
+ int buffer_index = -1;
+ size_t amt_complete = 0;
+ char *fnstr = (type == IO_READV) ? "pvfs2_file_readv" :
"pvfs2_file_writev";
- /* Calculate the total length to read by adding up the length of each io
- * segment */
+ ret = -EINVAL;
+ total_count = 0;
+ count = 0;
+ to_free = 0;
+ /*
+ * Calculate the total length to read/write by adding up the
+ * lengths of each io segment
+ */
for (seg = 0; seg < nr_segs; seg++)
{
const struct iovec *iv = &iov[seg];
count += iv->iov_len;
if (unlikely((ssize_t)(count|iv->iov_len) < 0))
- return -EINVAL;
+ goto out;
if (total_count + iv->iov_len < pvfs_bufmap_size_query())
{
total_count += iv->iov_len;
@@ -602,7 +581,22 @@
else {
total_count = (total_count + iv->iov_len -
pvfs_bufmap_size_query());
- max_new_nr_segs+=2;
+ max_new_nr_segs += 2;
+ }
+ }
+ if (type == IO_WRITEV)
+ {
+ /* perform generic linux kernel tests for sanity of write arguments */
+ /* NOTE: this is particularly helpful in handling fsize rlimit
properly */
+#ifdef PVFS2_LINUX_KERNEL_2_4
+ ret = pvfs2_precheck_file_write(file, inode, &count, offset);
+#else
+ ret = generic_write_checks(file, offset, &count,
S_ISBLK(inode->i_mode));
+#endif
+ if (ret != 0 || count == 0)
+ {
+ pvfs2_print("%s: failed generic argument checks.\n", fnstr);
+ goto out;
}
}
total_count = 0;
@@ -625,12 +619,12 @@
&new_nr_segs, &iovecptr, /* OUT */
&seg_count, &seg_array) /* OUT */ ) < 0)
{
- pvfs2_error("Failed to split iovecs to satisfy larger "
- " than blocksize readv request %d\n", ret);
- return ret;
+ pvfs2_error("%s: Failed to split iovecs to satisfy larger "
+ " than blocksize readv/writev request %d\n", fnstr, ret);
+ goto out;
}
- pvfs2_print("pvfs_file_readv: Splitting iovecs from %lu to %lu
[max_new %lu]\n",
- nr_segs, new_nr_segs, max_new_nr_segs);
+ pvfs2_print("%s: Splitting iovecs from %lu to %lu [max_new %lu]\n",
+ fnstr, nr_segs, new_nr_segs, max_new_nr_segs);
/* We must free seg_array and iovecptr */
to_free = 1;
}
@@ -646,53 +640,48 @@
to_free = 0;
}
ptr = iovecptr;
- pvfs2_print("pvfs2_file_readv reading [EMAIL PROTECTED]", (int) count,
lld(*offset));
- pvfs2_print("pvfs2_file_readv: new_nr_segs: %lu, seg_count: %u\n",
- new_nr_segs, seg_count);
+
+ pvfs2_print("%s [EMAIL PROTECTED]", fnstr, (int) count, *offset);
+ pvfs2_print("%s: new_nr_segs: %lu, seg_count: %u\n",
+ fnstr, new_nr_segs, seg_count);
+#ifdef PVFS2_KERNEL_DEBUG
for (seg = 0; seg < new_nr_segs; seg++)
{
- pvfs2_print("pvfs2_file_readv: %d) %p to %p [%d bytes]\n",
+ pvfs2_print("%s: %d) %p to %p [%d bytes]\n",
+ fnstr,
seg + 1, iovecptr[seg].iov_base,
iovecptr[seg].iov_base + iovecptr[seg].iov_len,
(int) iovecptr[seg].iov_len);
}
for (seg = 0; seg < seg_count; seg++)
{
- pvfs2_print("pvfs2_file_readv: %d) %u\n", seg + 1, seg_array[seg]);
+ pvfs2_print("%s: %d) %u\n", fnstr, seg + 1, seg_array[seg]);
}
+#endif
seg = 0;
while (total_count < count)
{
new_op = op_alloc();
if (!new_op)
{
- *offset = original_offset;
- if (to_free) {
- kfree(iovecptr);
- kfree(seg_array);
- }
- return -ENOMEM;
+ ret = -ENOMEM;
+ goto out;
}
-
new_op->upcall.type = PVFS2_VFS_OP_FILE_IO;
new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO; /* synchronous
I/O */
/* disable read-ahead */
new_op->upcall.req.io.readahead_size = 0;
- new_op->upcall.req.io.io_type = PVFS_IO_READ;
+ new_op->upcall.req.io.io_type =
+ (type == IO_READV) ? PVFS_IO_READ : PVFS_IO_WRITE;
new_op->upcall.req.io.refn = pvfs2_inode->refn;
+ /* get a shared buffer index */
ret = pvfs_bufmap_get(&buffer_index);
if (ret < 0)
{
- pvfs2_error("pvfs2_file_readv: pvfs_bufmap_get() "
- "failure (%d)\n", ret);
- op_release(new_op);
- *offset = original_offset;
- if (to_free) {
- kfree(iovecptr);
- kfree(seg_array);
- }
- return ret;
+ pvfs2_error("%s: pvfs_bufmap_get() "
+ "failure (%d)\n", fnstr, ret);
+ goto out;
}
/* how much to transfer in this loop iteration */
@@ -702,10 +691,28 @@
new_op->upcall.req.io.buf_index = buffer_index;
new_op->upcall.req.io.count = each_count;
new_op->upcall.req.io.offset = *offset;
-
- ret = service_operation(
- new_op, "pvfs2_file_readv", PVFS2_OP_RETRY_COUNT,
- get_interruptible_flag(inode));
+ if (type == IO_WRITEV)
+ {
+ /*
+ * copy data from application by pulling it out of the iovec.
+ * Number of segments to copy so that we don't overflow the
block-size
+ * is set in seg_array[], and ptr points to the appropriate
+ * beginning of the iovec from where data needs to be copied out,
+ * and each_count indicates the size in bytes that needs to be
pulled
+ * out. */
+ pvfs2_print("%s nr_segs %u, offset: %llu each_count: %d\n",
+ fnstr, seg_array[seg], *offset, (int) each_count);
+ ret = pvfs_bufmap_copy_iovec_from_user(
+ buffer_index, ptr, seg_array[seg], each_count);
+ if (ret < 0)
+ {
+ pvfs2_error("%s: Failed to copy user buffer. Please make sure
"
+ "that the pvfs2-client is running. %d\n", fnstr,
ret);
+ goto out;
+ }
+ }
+ ret = service_operation(new_op, fnstr,
+ PVFS2_OP_RETRY_COUNT, get_interruptible_flag(inode));
if (ret < 0)
{
@@ -719,50 +726,48 @@
*/
if (ret == -EINTR)
{
- pvfs2_print("pvfs2_file_readv: returning error %d\n", ret);
+ pvfs2_print("%s: returning error %d\n", fnstr, ret);
}
else
{
pvfs2_error(
- "pvfs2_file_readv: error writing to handle %llu, "
+ "%s: error on handle %llu, "
"FILE: %s\n -- returning %d\n",
- llu(pvfs2_ino_to_handle(inode->i_ino)),
+ fnstr, llu(pvfs2_ino_to_handle(inode->i_ino)),
(file && file->f_dentry && file->f_dentry->d_name.name
?
(char *)file->f_dentry->d_name.name : "UNKNOWN"),
ret);
}
- *offset = original_offset;
- if (to_free) {
- kfree(seg_array);
- kfree(iovecptr);
- }
- return ret;
+ goto out;
}
- pvfs2_print("pvfs2_file_readv nr_segs %u, offset: %llu
each_count:%d\n",
- (int) seg_array[seg], *offset, (int) each_count);
- /*
- * copy data to application by pushing it out to the iovec.
- * Number of segments to copy so that we don't
- * overflow the block-size is set in seg_array[], and
- * ptr points to the appropriate beginning of the
- * iovec from where data needs to be copied to, and
- * new_op->downcall.resp.io.amt_complete indicates
- * the size in bytes that needs to be pushed out
- */
- if (new_op->downcall.resp.io.amt_complete)
+
+ if (type == IO_READV)
{
- ret = pvfs_bufmap_copy_to_user_iovec(
- buffer_index, ptr, seg_array[seg],
- new_op->downcall.resp.io.amt_complete);
- if (ret < 0)
+ pvfs2_print("%s: nr_segs %u, offset: %llu each_count:%d\n",
+ fnstr, (int) seg_array[seg], *offset, (int) each_count);
+ /*
+ * copy data to application by pushing it out to the iovec.
+ * Number of segments to copy so that we don't
+ * overflow the block-size is set in seg_array[], and
+ * ptr points to the appropriate beginning of the
+ * iovec from where data needs to be copied to, and
+ * new_op->downcall.resp.io.amt_complete indicates
+ * the size in bytes that needs to be pushed out
+ */
+ if (new_op->downcall.resp.io.amt_complete)
{
- pvfs2_error("Failed to copy user buffer. Please make sure "
- "that the pvfs2-client is running.\n");
- /* put error codes in downcall so that handle_io_error()
- * preserves it properly */
- new_op->downcall.status = ret;
- handle_io_error();
- return(ret);
+ ret = pvfs_bufmap_copy_to_user_iovec(buffer_index, ptr,
seg_array[seg],
+ new_op->downcall.resp.io.amt_complete);
+ if (ret < 0)
+ {
+ pvfs2_error("Failed to copy user buffer. Please make sure
"
+ "that the pvfs2-client is running.\n");
+ /* put error codes in downcall so that handle_io_error()
+ * preserves it properly */
+ new_op->downcall.status = ret;
+ handle_io_error();
+ goto out;
+ }
}
}
/* advance the iovec pointer */
@@ -779,22 +784,54 @@
after this.
*/
wake_up_device_for_return(new_op);
+ new_op = NULL;
pvfs_bufmap_put(buffer_index);
+ buffer_index = -1;
- /* if we got a short write, fall out and return what we got so
- * far
+ /* if we got a short I/O operations,
+ * fall out and return what we got so far
*/
if (amt_complete < each_count)
{
break;
}
}
-
- if (to_free) {
+ if (total_count > 0)
+ {
+ ret = total_count;
+ }
+out:
+ if (new_op)
+ op_release(new_op);
+ if (buffer_index >= 0)
+ pvfs_bufmap_put(buffer_index);
+ if (to_free)
+ {
kfree(iovecptr);
kfree(seg_array);
}
- return total_count;
+ if (ret > 0 && file != NULL && inode != NULL)
+ {
+#ifdef HAVE_TOUCH_ATIME
+ touch_atime(file->f_vfsmnt, file->f_dentry);
+#else
+ update_atime(inode);
+#endif
+ }
+ return ret;
+}
+
+
+/** Reads data to several contiguous user buffers (an iovec) from a file at a
+ * specified offset.
+ */
+static ssize_t pvfs2_file_readv(
+ struct file *file,
+ const struct iovec *iov,
+ unsigned long nr_segs,
+ loff_t *offset)
+{
+ return do_readv_writev(IO_READV, file, iov, nr_segs, offset);
}
@@ -807,248 +844,7 @@
unsigned long nr_segs,
loff_t *offset)
{
- int ret = -1;
- pvfs2_kernel_op_t *new_op = NULL;
- struct iovec *iovecptr = NULL, *ptr = NULL;
- loff_t original_offset = *offset;
- int buffer_index = -1;
- struct inode *inode = file->f_dentry->d_inode;
- pvfs2_inode_t *pvfs2_inode = PVFS2_I(inode);
- size_t amt_complete = 0;
- size_t total_count = 0, count = 0, each_count = 0;
- unsigned int seg, to_free = 0;
- unsigned long new_nr_segs = 0, max_new_nr_segs = 0;
- unsigned int seg_count, *seg_array = NULL;
-
-
- /* Calculate the total length to write by adding up the length of each io
- * segment */
- for (seg = 0; seg < nr_segs; seg++)
- {
- const struct iovec *iv = &iov[seg];
- count += iv->iov_len;
- if (unlikely((ssize_t)(count|iv->iov_len) < 0))
- return -EINVAL;
- if (total_count + iv->iov_len < pvfs_bufmap_size_query())
- {
- total_count += iv->iov_len;
- max_new_nr_segs++;
- }
- else {
- total_count = (total_count + iv->iov_len -
pvfs_bufmap_size_query());
- max_new_nr_segs+=2;
- }
-
- }
- /* perform generic linux kernel tests for sanity of write arguments */
- /* NOTE: this is particularly helpful in handling fsize rlimit properly */
-#ifdef PVFS2_LINUX_KERNEL_2_4
- ret = pvfs2_precheck_file_write(file, inode, &count, offset);
-#else
- ret = generic_write_checks(file, offset, &count, S_ISBLK(inode->i_mode));
-#endif
- if (ret != 0 || count == 0)
- {
- pvfs2_print("pvfs2_file_writev: failed generic argument checks.\n");
- return(ret);
- }
-
- total_count = 0;
- /*
- * if the total size of data transfer requested is greater than
- * the kernel-set blocksize of PVFS2, then we split the iovecs
- * such that no iovec description straddles this block size
- * limitation.
- */
- if (count > pvfs_bufmap_size_query())
- {
- /*
- * Split up the given iovec description such that
- * no iovec descriptor straddles over the block-size limitation.
- * This makes us our job easier to stage the I/O.
- * In addition, this function will also compute an array with seg_count
- * entries that will store the number of segments that straddle the
- * block-size boundaries.
- */
- if ((ret = split_iovecs(max_new_nr_segs, nr_segs, iov, /* IN */
- &new_nr_segs, &iovecptr, /* OUT */
- &seg_count, &seg_array) /* OUT */ ) < 0)
- {
- pvfs2_error("Failed to split iovecs to satisfy larger than
blocksize writev request %d\n", ret);
- return ret;
- }
- pvfs2_print("pvfs_file_writev: Splitting iovecs from %lu to %lu
[max_new %lu]\n",
- nr_segs, new_nr_segs, max_new_nr_segs);
- /* We must free seg_array and iovecptr */
- to_free = 1;
- }
- else {
- /* Number of segments dont change! */
- new_nr_segs = nr_segs;
- /* use the given iovec description */
- iovecptr = (struct iovec *) iov;
- /* There is only 1 element in the seg_array */
- seg_count = 1;
- /* and its value is the number of segments passed in */
- seg_array = (unsigned int *) &nr_segs;
- /* We dont have to free up anything */
- to_free = 0;
- }
- ptr = iovecptr;
- pvfs2_print("pvfs2_file_writev writing [EMAIL PROTECTED]", (int) count,
*offset);
- pvfs2_print("pvfs2_file_writev: new_nr_segs: %lu, seg_count: %u\n",
- new_nr_segs, seg_count);
- for (seg = 0; seg < new_nr_segs; seg++)
- {
- pvfs2_print("pvfs2_file_writev: %d) %p to %p [%d bytes]\n",
- seg + 1, iovecptr[seg].iov_base,
- iovecptr[seg].iov_base + iovecptr[seg].iov_len,
- (int) iovecptr[seg].iov_len);
- }
- for (seg = 0; seg < seg_count; seg++)
- {
- pvfs2_print("pvfs2_file_writev: %d) %u\n", seg + 1, seg_array[seg]);
- }
- seg = 0;
- while (total_count < count)
- {
- new_op = op_alloc();
- if (!new_op)
- {
- *offset = original_offset;
- if (to_free) {
- kfree(iovecptr);
- kfree(seg_array);
- }
- return -ENOMEM;
- }
-
- new_op->upcall.type = PVFS2_VFS_OP_FILE_IO;
- new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO; /* synchronous
I/O */
- new_op->upcall.req.io.io_type = PVFS_IO_WRITE;
- new_op->upcall.req.io.refn = pvfs2_inode->refn;
-
- ret = pvfs_bufmap_get(&buffer_index);
- if (ret < 0)
- {
- pvfs2_error("pvfs2_file_writev: pvfs_bufmap_get() "
- "failure (%d)\n", ret);
- op_release(new_op);
- *offset = original_offset;
- if (to_free) {
- kfree(iovecptr);
- kfree(seg_array);
- }
- return ret;
- }
-
- /* how much to transfer in this loop iteration */
- each_count = (((count - total_count) > pvfs_bufmap_size_query()) ?
- pvfs_bufmap_size_query() : (count - total_count));
-
- new_op->upcall.req.io.buf_index = buffer_index;
- new_op->upcall.req.io.count = each_count;
- new_op->upcall.req.io.offset = *offset;
- pvfs2_print("pvfs2_file_writev nr_segs %u, offset: %llu each_count:
%d\n",
- seg_array[seg], *offset, (int) each_count);
-
- /*
- * copy data from application by pulling it out of the iovec.
- * Number of segments to copy so that we don't overflow the block-size
- * is set in seg_array[], and ptr points to the appropriate
- * beginning of the iovec from where data needs to be copied out,
- * and each_count indicates the size in bytes that needs to be pulled
- * out. */
- ret = pvfs_bufmap_copy_iovec_from_user(
- buffer_index, ptr, seg_array[seg], each_count);
- if (ret < 0)
- {
- pvfs2_error("Failed to copy user buffer. Please make sure "
- "that the pvfs2-client is running. %d\n", ret);
- op_release(new_op);
- pvfs_bufmap_put(buffer_index);
- *offset = original_offset;
- if (to_free) {
- kfree(seg_array);
- kfree(iovecptr);
- }
- return ret;
- }
-
- ret = service_operation(
- new_op, "pvfs2_file_writev", PVFS2_OP_RETRY_COUNT,
- get_interruptible_flag(inode));
-
- if (ret < 0)
- {
- /* this macro is defined in pvfs2-kernel.h */
- handle_io_error();
-
- /*
- don't write an error to syslog on signaled operation
- termination unless we've got debugging turned on, as
- this can happen regularly (i.e. ctrl-c)
- */
- if (ret == -EINTR)
- {
- pvfs2_print("pvfs2_file_writev: returning error %d\n", ret);
- }
- else
- {
- pvfs2_error(
- "pvfs2_file_writev: error writing to handle %llu, "
- "FILE: %s\n -- returning %d\n",
- llu(pvfs2_ino_to_handle(inode->i_ino)),
- (file && file->f_dentry && file->f_dentry->d_name.name
?
- (char *)file->f_dentry->d_name.name : "UNKNOWN"),
- ret);
- }
- *offset = original_offset;
- if (to_free) {
- kfree(seg_array);
- kfree(iovecptr);
- }
- return ret;
- }
- /* advance the iovec pointer */
- ptr += seg_array[seg];
- seg++;
- *offset += new_op->downcall.resp.io.amt_complete;
- total_count += new_op->downcall.resp.io.amt_complete;
- amt_complete = new_op->downcall.resp.io.amt_complete;
-
- /*
- tell the device file owner waiting on I/O that this read has
- completed and it can return now. in this exact case, on
- wakeup the device will free the op, so we *cannot* touch it
- after this.
- */
- wake_up_device_for_return(new_op);
- pvfs_bufmap_put(buffer_index);
-
- /* if we got a short write, fall out and return what we got so
- * far TODO: define semantics here- kind of depends on pvfs2
- * semantics that don't really exist yet
- */
- if (amt_complete < each_count)
- {
- break;
- }
- }
-
- if (to_free) {
- kfree(iovecptr);
- kfree(seg_array);
- }
- if (total_count)
- {
-#ifdef HAVE_TOUCH_ATIME
- touch_atime(file->f_vfsmnt, file->f_dentry);
-#else
- update_atime(inode);
-#endif
- }
- return total_count;
+ return do_readv_writev(IO_WRITEV, file, iov, nr_segs, offset);
}
#ifdef HAVE_AIO_VFS_SUPPORT
Index: src/kernel/linux-2.6/pvfs2-kernel.h
===================================================================
RCS file: /anoncvs/pvfs2/src/kernel/linux-2.6/pvfs2-kernel.h,v
retrieving revision 1.116
diff -u -r1.116 pvfs2-kernel.h
--- src/kernel/linux-2.6/pvfs2-kernel.h 19 May 2006 21:37:11 -0000 1.116
+++ src/kernel/linux-2.6/pvfs2-kernel.h 26 May 2006 02:59:21 -0000
@@ -752,8 +752,9 @@
{ \
wake_up_device_for_return(new_op); \
} \
+ new_op = NULL; \
pvfs_bufmap_put(buffer_index); \
- *offset = original_offset; \
+ buffer_index = -1; \
} while(0)
#ifdef HAVE_AIO_VFS_SUPPORT
_______________________________________________
Pvfs2-developers mailing list
Pvfs2-developers@beowulf-underground.org
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
