<div dir="ltr"><br><div class="gmail_extra"><div class="gmail_quote">On Mon, Nov 7, 2016 at 6:19 AM, James Simmons <span dir="ltr"><<a href="mailto:jsimmons@infradead.org" target="_blank">jsimmons@infradead.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5"><br>
> On Thu, Nov 3, 2016 at 1:05 AM, Dilger, Andreas <<a href="mailto:andreas.dilger@intel.com">andreas.dilger@intel.com</a>> wrote:<br>
> On Oct 25, 2016, at 10:47, Aya Mahfouz <<a href="mailto:mahfouz.saif.elyazal@gmail.com">mahfouz.saif.elyazal@gmail.<wbr>com</a>> wrote:<br>
> ><br>
> > On Mon, Oct 17, 2016 at 10:38:31PM +0000, Dilger, Andreas wrote:<br>
> >> On Oct 17, 2016, at 15:46, Aya Mahfouz <<a href="mailto:mahfouz.saif.elyazal@gmail.com">mahfouz.saif.elyazal@gmail.<wbr>com</a>> wrote:<br>
> >>><br>
> >>> class_devno_max is an inline function that returns<br>
> >>> MAX_OBD_DEVICES. Replace all calls to the function<br>
> >>> by MAX_OBD_DEVICES.<br>
> >><br>
> >> Thanks for your patch, but unfortunately it can't be accepted.<br>
> >><br>
> >> This function was added in preparation of being able to tune the maximum<br>
> >> number of storage devices dynamically, rather than having to hard code it<br>
> >> to the maximum possible number of servers that a client can possibly<br>
> >> connect to.<br>
> >><br>
> >> While the current maximum of 8192 servers has been enough for current<br>
> >> filesystems, I'd rather move in the direction of dynamically handling this<br>
> >> limit rather than re-introducing a hard-coded constant throughout the code.<br>
> >><br>
> > Hello,<br>
> ><br>
> > I would like to proceed with implementing the function if possible.<br>
> > Kindly direct me to some starting pointers.<br>
><br>
> Hi Aya,<br>
> thanks for offering to look into this.<br>
><br>
> There are several ways to approach this problem to make the allocation<br>
> of the obd_devs[] array dynamic. In most cases, there isn't any value<br>
> to dynamically shrink this array, since the filesystem(s) will typically<br>
> be mounted until the node is rebooted, and it is only in the tens of KB<br>
> size range, so this will not affect ongoing operations, and that simplifies<br>
> the implementation.<br>
><br>
> The easiest way would be to have a dynamically-sized obd_devs[] array that<br>
> is reallocated in class_newdev() in PAGE_SIZE chunks whenever the current<br>
> array has no more free slots and copied to the new array, using obd_dev_lock<br>
> to protect the array while it is being reallocated and copied. In most<br>
> cases, this would save memory over the static array (not many filesystems<br>
> have so many servers), but for the few sites that have 10000+ servers they<br>
> don't need to change the source to handle this. Using libcfs_kvzalloc()<br>
> would avoid issues with allocating large chunks of memory.<br>
><br>
> There are a few places where obd_devs[] is accessed outside obd_dev_lock<br>
> that would need to be fixed now that this array may be changed at runtime.<br>
><br>
> A second approach that may scale better is to change obd_devs from an array<br>
> to a doubly linked list (using standard list_head helpers). In many cases<br>
> the whole list is seached linearly, and most of the uses of class_num2obd()<br>
> are just used to walk that list in order, which could be replaced with<br>
> list_for_each_entry() list traversal. The class_name2dev() function should<br>
> be changed to return the pointer to the obd_device structure, and a new<br>
> helper class_dev2num() would just return the obd_minor number from the<br>
> obd_device struct for the one use in class_resolve_dev_name(). Using a<br>
> linked list has the advantage that there is no need to search for free slots<br>
> in the array, since devices would be removed from the list when it is freed.<br>
><br>
> Cheers, Andreas<br>
><br>
> Thanks Andreas! Will start looking into it.<br>
<br>
</div></div>Just to let you know I opened a ticket for you.<br>
<br>
<a href="https://jira.hpdd.intel.com/browse/LU-8802" rel="noreferrer" target="_blank">https://jira.hpdd.intel.com/<wbr>browse/LU-8802</a><br>
<br>
This way wwe can track the progress and have Lustre developers assigened<br>
to look at your work. Thanks for stepping forward.<br>
<div class="HOEnZb"><div class="h5"><br></div></div></blockquote><div>Thanks, I've created an account: asaifm <br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">
<br>
> --<br>
> Kind Regards,<br>
> Aya Saif El-yazal Mahfouz<br>
> <br>
> >> One comment inline below, if you still want to submit a patch.<br>
> >><br>
> >>> Signed-off-by: Aya Mahfouz <<a href="mailto:mahfouz.saif.elyazal@gmail.com">mahfouz.saif.elyazal@gmail.<wbr>com</a>><br>
> >>> ---<br>
> >>> drivers/staging/lustre/lustre/<wbr>obdclass/class_obd.c | 6 +++---<br>
> >>> drivers/staging/lustre/lustre/<wbr>obdclass/genops.c | 22 +++++++++++-----------<br>
> >>> .../lustre/lustre/obdclass/<wbr>linux/linux-module.c | 6 +++---<br>
> >>> 3 files changed, 17 insertions(+), 17 deletions(-)<br>
> >>><br>
> >>> diff --git a/drivers/staging/lustre/<wbr>lustre/obdclass/class_obd.c<br>
> b/drivers/staging/lustre/<wbr>lustre/obdclass/class_obd.c<br>
> >>> index 2b21675..b775c74 100644<br>
> >>> --- a/drivers/staging/lustre/<wbr>lustre/obdclass/class_obd.c<br>
> >>> +++ b/drivers/staging/lustre/<wbr>lustre/obdclass/class_obd.c<br>
> >>> @@ -345,7 +345,7 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg)<br>
> >>> goto out;<br>
> >>> }<br>
> >>> obd = class_name2obd(data->ioc_<wbr>inlbuf4);<br>
> >>> - } else if (data->ioc_dev < class_devno_max()) {<br>
> >>> + } else if (data->ioc_dev < MAX_OBD_DEVICES) {<br>
> >>> obd = class_num2obd(data->ioc_dev);<br>
> >>> } else {<br>
> >>> CERROR("OBD ioctl: No device\n");<br>
> >>> @@ -498,7 +498,7 @@ static int __init obdclass_init(void)<br>
> >>> }<br>
> >>><br>
> >>> /* This struct is already zeroed for us (static global) */<br>
> >>> - for (i = 0; i < class_devno_max(); i++)<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++)<br>
> >>> obd_devs[i] = NULL;<br>
> >><br>
> >> This block can just be removed entirely. It used to do something useful,<br>
> >> but through a series of changes it has become useless.<br>
> >><br>
> >> Cheers, Andreas<br>
> >><br>
> >>> /* Default the dirty page cache cap to 1/2 of system memory.<br>
> >>> @@ -548,7 +548,7 @@ static void obdclass_exit(void)<br>
> >>> lustre_unregister_fs();<br>
> >>><br>
> >>> misc_deregister(&obd_psdev);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (obd && obd->obd_set_up &&<br>
> >>> diff --git a/drivers/staging/lustre/<wbr>lustre/obdclass/genops.c<br>
> b/drivers/staging/lustre/<wbr>lustre/obdclass/genops.c<br>
> >>> index 99c2da6..af4fc58 100644<br>
> >>> --- a/drivers/staging/lustre/<wbr>lustre/obdclass/genops.c<br>
> >>> +++ b/drivers/staging/lustre/<wbr>lustre/obdclass/genops.c<br>
> >>> @@ -290,7 +290,7 @@ struct obd_device *class_newdev(const char *type_name, const char *name)<br>
> >>> LASSERT(newdev->obd_magic == OBD_DEVICE_MAGIC);<br>
> >>><br>
> >>> write_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (obd && (strcmp(name, obd->obd_name) == 0)) {<br>
> >>> @@ -322,9 +322,9 @@ struct obd_device *class_newdev(const char *type_name, const char *name)<br>
> >>> }<br>
> >>> write_unlock(&obd_dev_lock);<br>
> >>><br>
> >>> - if (!result && i >= class_devno_max()) {<br>
> >>> + if (!result && i >= MAX_OBD_DEVICES) {<br>
> >>> CERROR("all %u OBD devices used, increase MAX_OBD_DEVICES\n",<br>
> >>> - class_devno_max());<br>
> >>> + MAX_OBD_DEVICES);<br>
> >>> result = ERR_PTR(-EOVERFLOW);<br>
> >>> goto out;<br>
> >>> }<br>
> >>> @@ -372,7 +372,7 @@ int class_name2dev(const char *name)<br>
> >>> return -1;<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (obd && strcmp(name, obd->obd_name) == 0) {<br>
> >>> @@ -397,7 +397,7 @@ struct obd_device *class_name2obd(const char *name)<br>
> >>> {<br>
> >>> int dev = class_name2dev(name);<br>
> >>><br>
> >>> - if (dev < 0 || dev > class_devno_max())<br>
> >>> + if (dev < 0 || dev > MAX_OBD_DEVICES)<br>
> >>> return NULL;<br>
> >>> return class_num2obd(dev);<br>
> >>> }<br>
> >>> @@ -408,7 +408,7 @@ int class_uuid2dev(struct obd_uuid *uuid)<br>
> >>> int i;<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (obd && obd_uuid_equals(uuid, &obd->obd_uuid)) {<br>
> >>> @@ -435,7 +435,7 @@ struct obd_device *class_num2obd(int num)<br>
> >>> {<br>
> >>> struct obd_device *obd = NULL;<br>
> >>><br>
> >>> - if (num < class_devno_max()) {<br>
> >>> + if (num < MAX_OBD_DEVICES) {<br>
> >>> obd = obd_devs[num];<br>
> >>> if (!obd)<br>
> >>> return NULL;<br>
> >>> @@ -463,7 +463,7 @@ struct obd_device *class_find_client_obd(struct obd_uuid *tgt_uuid,<br>
> >>> int i;<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (!obd)<br>
> >>> @@ -496,13 +496,13 @@ struct obd_device *class_devices_in_group(struct obd_uuid *grp_uuid, int<br>
> *next)<br>
> >>><br>
> >>> if (!next)<br>
> >>> i = 0;<br>
> >>> - else if (*next >= 0 && *next < class_devno_max())<br>
> >>> + else if (*next >= 0 && *next < MAX_OBD_DEVICES)<br>
> >>> i = *next;<br>
> >>> else<br>
> >>> return NULL;<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (; i < class_devno_max(); i++) {<br>
> >>> + for (; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd = class_num2obd(i);<br>
> >>><br>
> >>> if (!obd)<br>
> >>> @@ -533,7 +533,7 @@ int class_notify_sptlrpc_conf(<wbr>const char *fsname, int namelen)<br>
> >>> LASSERT(namelen > 0);<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> obd = class_num2obd(i);<br>
> >>><br>
> >>> if (!obd || obd->obd_set_up == 0 || obd->obd_stopping)<br>
> >>> diff --git a/drivers/staging/lustre/<wbr>lustre/obdclass/linux/linux-<wbr>module.c<br>
> b/drivers/staging/lustre/<wbr>lustre/obdclass/linux/linux-<wbr>module.c<br>
> >>> index 33342bf..ca5b466 100644<br>
> >>> --- a/drivers/staging/lustre/<wbr>lustre/obdclass/linux/linux-<wbr>module.c<br>
> >>> +++ b/drivers/staging/lustre/<wbr>lustre/obdclass/linux/linux-<wbr>module.c<br>
> >>> @@ -228,7 +228,7 @@ static ssize_t health_show(struct kobject *kobj, struct attribute *attr,<br>
> >>> return sprintf(buf, "LBUG\n");<br>
> >>><br>
> >>> read_lock(&obd_dev_lock);<br>
> >>> - for (i = 0; i < class_devno_max(); i++) {<br>
> >>> + for (i = 0; i < MAX_OBD_DEVICES; i++) {<br>
> >>> struct obd_device *obd;<br>
> >>><br>
> >>> obd = class_num2obd(i);<br>
> >>> @@ -326,7 +326,7 @@ static struct attribute *lustre_attrs[] = {<br>
> >>><br>
> >>> static void *obd_device_list_seq_start(<wbr>struct seq_file *p, loff_t *pos)<br>
> >>> {<br>
> >>> - if (*pos >= class_devno_max())<br>
> >>> + if (*pos >= MAX_OBD_DEVICES)<br>
> >>> return NULL;<br>
> >>><br>
> >>> return pos;<br>
> >>> @@ -339,7 +339,7 @@ static void obd_device_list_seq_stop(<wbr>struct seq_file *p, void *v)<br>
> >>> static void *obd_device_list_seq_next(<wbr>struct seq_file *p, void *v, loff_t *pos)<br>
> >>> {<br>
> >>> ++*pos;<br>
> >>> - if (*pos >= class_devno_max())<br>
> >>> + if (*pos >= MAX_OBD_DEVICES)<br>
> >>> return NULL;<br>
> >>><br>
> >>> return pos;<br>
> >>> --<br>
> >>> 2.5.0<br>
> >>><br>
> >>><br>
> >>> --<br>
> >>> Kind Regards,<br>
> >>> Aya Saif El-yazal Mahfouz<br>
> >>> ______________________________<wbr>_________________<br>
> >>> lustre-devel mailing list<br>
> >>> <a href="mailto:lustre-devel@lists.lustre.org">lustre-devel@lists.lustre.org</a><br>
> >>> <a href="http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org" rel="noreferrer" target="_blank">http://lists.lustre.org/<wbr>listinfo.cgi/lustre-devel-<wbr>lustre.org</a><br>
> >><br>
> ><br>
> > --<br>
> > Kind Regards,<br>
> > Aya Saif El-yazal Mahfouz<br>
> > ______________________________<wbr>_________________<br>
> > lustre-devel mailing list<br>
> > <a href="mailto:lustre-devel@lists.lustre.org">lustre-devel@lists.lustre.org</a><br>
> > <a href="http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org" rel="noreferrer" target="_blank">http://lists.lustre.org/<wbr>listinfo.cgi/lustre-devel-<wbr>lustre.org</a><br>
><br>
><br>
><br>
> </div></div></blockquote></div><br></div></div>