* [PATCH] tcm/fileio: Convert to iov allocations for fd_do_readv() and fd_do_writev()
@ 2010-09-22 8:02 Nicholas A. Bellinger
2010-09-22 11:26 ` Boaz Harrosh
0 siblings, 1 reply; 3+ messages in thread
From: Nicholas A. Bellinger @ 2010-09-22 8:02 UTC (permalink / raw)
To: linux-kernel, linux-scsi, Christoph Hellwig, FUJITA Tomonori
Cc: Fubo Chen, Nicholas Bellinger
From: Nicholas Bellinger <nab@linux-iscsi.org>
Greetings all,
This patch converts target_core_file.c fd_do_readv() and fd_do_writev() code to
use dynamic allocation for *iov[] instead of local scope static memory to
properly handle cases with TCM_Loop that can potentially exceed the 8K stack
on 64-bit kernels with per 1:1 struct scatterlist <-> 512-byte block_size SGL length
mapping up to the current FD_MAX_SECTORS=1024. For more information please see:
http://lkml.org/lkml/2010/9/22/42
Many thanks to Fubo Chen for spotting this one!
Signed-off-by: Nicholas A. Bellinger <nab@linux-iscsi.org>
---
drivers/target/target_core_file.c | 20 ++++++++++++++++----
1 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/drivers/target/target_core_file.c b/drivers/target/target_core_file.c
index 221b41e..93d3f63 100644
--- a/drivers/target/target_core_file.c
+++ b/drivers/target/target_core_file.c
@@ -476,12 +476,16 @@ static int fd_do_readv(struct fd_request *req, struct se_task *task)
{
struct file *fd = req->fd_dev->fd_file;
struct scatterlist *sg = task->task_sg;
- struct iovec iov[req->fd_sg_count];
+ struct iovec *iov;
mm_segment_t old_fs;
loff_t pos = (req->fd_lba * DEV_ATTRIB(task->se_dev)->block_size);
int ret = 0, i;
- memset(iov, 0, sizeof(struct iovec) * req->fd_sg_count);
+ iov = kzalloc(sizeof(struct iovec) * req->fd_sg_count, GFP_KERNEL);
+ if (!(iov)) {
+ printk(KERN_ERR "Unable to allocate fd_do_readv iov[]\n");
+ return -1;
+ }
for (i = 0; i < req->fd_sg_count; i++) {
iov[i].iov_len = sg[i].length;
@@ -492,6 +496,8 @@ static int fd_do_readv(struct fd_request *req, struct se_task *task)
set_fs(get_ds());
ret = vfs_readv(fd, &iov[0], req->fd_sg_count, &pos);
set_fs(old_fs);
+
+ kfree(iov);
/*
* Return zeros and GOOD status even if the READ did not return
* the expected virt_size for struct file w/o a backing struct
@@ -583,12 +589,16 @@ static int fd_do_writev(struct fd_request *req, struct se_task *task)
{
struct file *fd = req->fd_dev->fd_file;
struct scatterlist *sg = task->task_sg;
- struct iovec iov[req->fd_sg_count];
+ struct iovec *iov;
mm_segment_t old_fs;
loff_t pos = (req->fd_lba * DEV_ATTRIB(task->se_dev)->block_size);
int ret, i = 0;
- memset(iov, 0, sizeof(struct iovec) * req->fd_sg_count);
+ iov = kzalloc(sizeof(struct iovec) * req->fd_sg_count, GFP_KERNEL);
+ if (!(iov)) {
+ printk(KERN_ERR "Unable to allocate fd_do_writev iov[]\n");
+ return -1;
+ }
for (i = 0; i < req->fd_sg_count; i++) {
iov[i].iov_len = sg[i].length;
@@ -600,6 +610,8 @@ static int fd_do_writev(struct fd_request *req, struct se_task *task)
ret = vfs_writev(fd, &iov[0], req->fd_sg_count, &pos);
set_fs(old_fs);
+ kfree(iov);
+
if (ret < 0 || ret != req->fd_size) {
printk(KERN_ERR "vfs_writev() returned %d\n", ret);
return -1;
--
1.5.6.5
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH] tcm/fileio: Convert to iov allocations for fd_do_readv() and fd_do_writev()
2010-09-22 8:02 [PATCH] tcm/fileio: Convert to iov allocations for fd_do_readv() and fd_do_writev() Nicholas A. Bellinger
@ 2010-09-22 11:26 ` Boaz Harrosh
2010-09-22 11:58 ` Nicholas A. Bellinger
0 siblings, 1 reply; 3+ messages in thread
From: Boaz Harrosh @ 2010-09-22 11:26 UTC (permalink / raw)
To: Nicholas A. Bellinger
Cc: linux-kernel, linux-scsi, Christoph Hellwig, FUJITA Tomonori,
Mike Christie, Fubo Chen
On 09/22/2010 10:02 AM, Nicholas A. Bellinger wrote:
> From: Nicholas Bellinger <nab@linux-iscsi.org>
>
> Greetings all,
>
> This patch converts target_core_file.c fd_do_readv() and fd_do_writev() code to
> use dynamic allocation for *iov[] instead of local scope static memory to
> properly handle cases with TCM_Loop that can potentially exceed the 8K stack
> on 64-bit kernels with per 1:1 struct scatterlist <-> 512-byte block_size SGL length
> mapping up to the current FD_MAX_SECTORS=1024. For more information please see:
>
> http://lkml.org/lkml/2010/9/22/42
>
> Many thanks to Fubo Chen for spotting this one!
>
> Signed-off-by: Nicholas A. Bellinger <nab@linux-iscsi.org>
> ---
> drivers/target/target_core_file.c | 20 ++++++++++++++++----
> 1 files changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/target/target_core_file.c b/drivers/target/target_core_file.c
> index 221b41e..93d3f63 100644
> --- a/drivers/target/target_core_file.c
> +++ b/drivers/target/target_core_file.c
> @@ -476,12 +476,16 @@ static int fd_do_readv(struct fd_request *req, struct se_task *task)
> {
> struct file *fd = req->fd_dev->fd_file;
> struct scatterlist *sg = task->task_sg;
> - struct iovec iov[req->fd_sg_count];
> + struct iovec *iov;
> mm_segment_t old_fs;
> loff_t pos = (req->fd_lba * DEV_ATTRIB(task->se_dev)->block_size);
> int ret = 0, i;
>
> - memset(iov, 0, sizeof(struct iovec) * req->fd_sg_count);
> + iov = kzalloc(sizeof(struct iovec) * req->fd_sg_count, GFP_KERNEL);
A kzalloc allocation bigger then PAG_SIZE is very unstable, more so when
the system is up for a long time.
Better prepare to loop here, up to PAG_SIZE iov at a time.
Boaz
> + if (!(iov)) {
> + printk(KERN_ERR "Unable to allocate fd_do_readv iov[]\n");
> + return -1;
> + }
>
> for (i = 0; i < req->fd_sg_count; i++) {
> iov[i].iov_len = sg[i].length;
> @@ -492,6 +496,8 @@ static int fd_do_readv(struct fd_request *req, struct se_task *task)
> set_fs(get_ds());
> ret = vfs_readv(fd, &iov[0], req->fd_sg_count, &pos);
> set_fs(old_fs);
> +
> + kfree(iov);
> /*
> * Return zeros and GOOD status even if the READ did not return
> * the expected virt_size for struct file w/o a backing struct
> @@ -583,12 +589,16 @@ static int fd_do_writev(struct fd_request *req, struct se_task *task)
> {
> struct file *fd = req->fd_dev->fd_file;
> struct scatterlist *sg = task->task_sg;
> - struct iovec iov[req->fd_sg_count];
> + struct iovec *iov;
> mm_segment_t old_fs;
> loff_t pos = (req->fd_lba * DEV_ATTRIB(task->se_dev)->block_size);
> int ret, i = 0;
>
> - memset(iov, 0, sizeof(struct iovec) * req->fd_sg_count);
> + iov = kzalloc(sizeof(struct iovec) * req->fd_sg_count, GFP_KERNEL);
> + if (!(iov)) {
> + printk(KERN_ERR "Unable to allocate fd_do_writev iov[]\n");
> + return -1;
> + }
>
> for (i = 0; i < req->fd_sg_count; i++) {
> iov[i].iov_len = sg[i].length;
> @@ -600,6 +610,8 @@ static int fd_do_writev(struct fd_request *req, struct se_task *task)
> ret = vfs_writev(fd, &iov[0], req->fd_sg_count, &pos);
> set_fs(old_fs);
>
> + kfree(iov);
> +
> if (ret < 0 || ret != req->fd_size) {
> printk(KERN_ERR "vfs_writev() returned %d\n", ret);
> return -1;
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] tcm/fileio: Convert to iov allocations for fd_do_readv() and fd_do_writev()
2010-09-22 11:26 ` Boaz Harrosh
@ 2010-09-22 11:58 ` Nicholas A. Bellinger
0 siblings, 0 replies; 3+ messages in thread
From: Nicholas A. Bellinger @ 2010-09-22 11:58 UTC (permalink / raw)
To: Boaz Harrosh
Cc: linux-kernel, linux-scsi, Christoph Hellwig, FUJITA Tomonori,
Mike Christie, Fubo Chen
On Wed, 2010-09-22 at 13:26 +0200, Boaz Harrosh wrote:
> On 09/22/2010 10:02 AM, Nicholas A. Bellinger wrote:
> > From: Nicholas Bellinger <nab@linux-iscsi.org>
> >
> > Greetings all,
> >
> > This patch converts target_core_file.c fd_do_readv() and fd_do_writev() code to
> > use dynamic allocation for *iov[] instead of local scope static memory to
> > properly handle cases with TCM_Loop that can potentially exceed the 8K stack
> > on 64-bit kernels with per 1:1 struct scatterlist <-> 512-byte block_size SGL length
> > mapping up to the current FD_MAX_SECTORS=1024. For more information please see:
> >
> > http://lkml.org/lkml/2010/9/22/42
> >
> > Many thanks to Fubo Chen for spotting this one!
> >
> > Signed-off-by: Nicholas A. Bellinger <nab@linux-iscsi.org>
> > ---
> > drivers/target/target_core_file.c | 20 ++++++++++++++++----
> > 1 files changed, 16 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/target/target_core_file.c b/drivers/target/target_core_file.c
> > index 221b41e..93d3f63 100644
> > --- a/drivers/target/target_core_file.c
> > +++ b/drivers/target/target_core_file.c
> > @@ -476,12 +476,16 @@ static int fd_do_readv(struct fd_request *req, struct se_task *task)
> > {
> > struct file *fd = req->fd_dev->fd_file;
> > struct scatterlist *sg = task->task_sg;
> > - struct iovec iov[req->fd_sg_count];
> > + struct iovec *iov;
> > mm_segment_t old_fs;
> > loff_t pos = (req->fd_lba * DEV_ATTRIB(task->se_dev)->block_size);
> > int ret = 0, i;
> >
> > - memset(iov, 0, sizeof(struct iovec) * req->fd_sg_count);
> > + iov = kzalloc(sizeof(struct iovec) * req->fd_sg_count, GFP_KERNEL);
>
> A kzalloc allocation bigger then PAG_SIZE is very unstable, more so when
> the system is up for a long time.
>
> Better prepare to loop here, up to PAG_SIZE iov at a time.
>
Hmmm, good point for the extreme TCM_Loop single 512-byte block per SGL
case. Looks like we need to split up the struct iovec mappings for
FILEIO into multiple allocations in fd_do_readv() and fd_do_writev()..?
Best,
--nab
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2010-09-22 11:58 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2010-09-22 8:02 [PATCH] tcm/fileio: Convert to iov allocations for fd_do_readv() and fd_do_writev() Nicholas A. Bellinger
2010-09-22 11:26 ` Boaz Harrosh
2010-09-22 11:58 ` Nicholas A. Bellinger
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).