From: Oleg Drokin <green@namesys.com>
To: "David S. Miller" <davem@redhat.com>
Cc: szepe@pinerecords.com, mason@suse.com, reiser@namesys.com,
shaggy@austin.ibm.com, marcelo@conectiva.com.br,
linux-kernel@vger.kernel.org, reiserfs-dev@namesys.com,
linuxjfs@us.ibm.com
Subject: Re: [reiserfs-dev] Re: [PATCH] sparc32: wrong type of nlink_t
Date: Thu, 5 Sep 2002 13:54:42 +0400 [thread overview]
Message-ID: <20020905135442.A19682@namesys.com> (raw)
In-Reply-To: <20020904.223651.79770866.davem@redhat.com>
Hello!
On Wed, Sep 04, 2002 at 10:36:51PM -0700, David S. Miller wrote:
> > Our disk format has link counts > 32k
> Does the internal reiserfs nlink value translate directly
> to what stat() puts in st_nlink?
> It really doesn't matter. Even if you have some huge value that can't
> be represented in st_nlink, you can report to the user that st_nlink
> is NLINK_MAX.
Ok, since I really like this approach, below is the patch (for 2.4) that
demonstrates my solution.
Also it correctly calculates maximal number given type may hold ( does not work
with unsigned long long, though) with my own way ;)
FS compatibility is still intact and older kernel will work.
If nobody will object against this code, then I'll implement something
similar for 2.5 too and we will fix reiserfsck to understand this correct
nlink accounting.
The only downside of this patch is that v3.5 filesystems won't allow to
create more than 64k-2 subdirs in a dir, like old code did, but this is
not an issue I think. (already existent dirs should still work)
Bye,
Oleg
===== fs/reiserfs/inode.c 1.35 vs edited =====
--- 1.35/fs/reiserfs/inode.c Fri Aug 9 19:22:33 2002
+++ edited/fs/reiserfs/inode.c Thu Sep 5 13:43:26 2002
@@ -891,7 +891,8 @@
set_inode_item_key_version (inode, KEY_FORMAT_3_5);
set_inode_sd_version (inode, STAT_DATA_V1);
inode->i_mode = sd_v1_mode(sd);
- inode->i_nlink = sd_v1_nlink(sd);
+ inode->i_nlink = ( sd_v1_nlink(sd) > REISERFS_LINK_MAX ) ? REISERFS_LINK_MAX:sd_v1_nlink(sd);
+ inode->u.reiserfs_i.i_nlink_real = sd_v1_nlink(sd);
inode->i_uid = sd_v1_uid(sd);
inode->i_gid = sd_v1_gid(sd);
inode->i_size = sd_v1_size(sd);
@@ -923,7 +924,8 @@
struct stat_data * sd = (struct stat_data *)B_I_PITEM (bh, ih);
inode->i_mode = sd_v2_mode(sd);
- inode->i_nlink = sd_v2_nlink(sd);
+ inode->i_nlink = (sd_v2_nlink(sd)>REISERFS_LINK_MAX)?REISERFS_LINK_MAX:sd_v2_nlink(sd);
+ inode->u.reiserfs_i.i_nlink_real = sd_v2_nlink(sd);
inode->i_uid = sd_v2_uid(sd);
inode->i_size = sd_v2_size(sd);
inode->i_gid = sd_v2_gid(sd);
@@ -975,7 +977,7 @@
__u16 flags;
set_sd_v2_mode(sd_v2, inode->i_mode );
- set_sd_v2_nlink(sd_v2, inode->i_nlink );
+ set_sd_v2_nlink(sd_v2, inode->u.reiserfs_i.i_nlink_real );
set_sd_v2_uid(sd_v2, inode->i_uid );
set_sd_v2_size(sd_v2, inode->i_size );
set_sd_v2_gid(sd_v2, inode->i_gid );
@@ -1001,7 +1003,7 @@
set_sd_v1_mode(sd_v1, inode->i_mode );
set_sd_v1_uid(sd_v1, inode->i_uid );
set_sd_v1_gid(sd_v1, inode->i_gid );
- set_sd_v1_nlink(sd_v1, inode->i_nlink );
+ set_sd_v1_nlink(sd_v1, inode->u.reiserfs_i.i_nlink_real );
set_sd_v1_size(sd_v1, inode->i_size );
set_sd_v1_atime(sd_v1, inode->i_atime );
set_sd_v1_ctime(sd_v1, inode->i_ctime );
@@ -1537,7 +1539,7 @@
/* fill stat data */
inode->i_mode = mode;
- inode->i_nlink = (S_ISDIR (mode) ? 2 : 1);
+ inode->u.reiserfs_i.i_nlink_real = inode->i_nlink = (S_ISDIR (mode) ? 2 : 1);
inode->i_uid = current->fsuid;
if (dir->i_mode & S_ISGID) {
inode->i_gid = dir->i_gid;
===== fs/reiserfs/namei.c 1.24 vs edited =====
--- 1.24/fs/reiserfs/namei.c Fri Aug 9 19:22:33 2002
+++ edited/fs/reiserfs/namei.c Thu Sep 5 13:34:43 2002
@@ -8,8 +8,16 @@
#include <linux/reiserfs_fs.h>
#include <linux/smp_lock.h>
-#define INC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) { i->i_nlink++; if (i->i_nlink >= REISERFS_LINK_MAX) i->i_nlink=1; }
-#define DEC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) i->i_nlink--;
+#define INC_INODE_NLINK(i) { if (i->i_nlink < REISERFS_LINK_MAX) i->i_nlink++; i->u.reiserfs_i.i_nlink_real++; }
+#define DEC_INODE_NLINK(i) {if ( --i->u.reiserfs_i.i_nlink_real < REISERFS_LINK_MAX) i->i_nlink--;}
+
+// v3.5 files have 16bit nlink_t, v3.6 files have 32bit nlink_t.
+#define CAN_INCREASE_NLINK(i) ( i->u.reiserfs_i.i_nlink_real < ((get_inode_item_key_version (inode) != KEY_FORMAT_3_5)?MAX_UL_INT:MAX_US_INT))
+
+// Compatibility stuff with old trick that allowed to have lots of subdirs in
+// one dir. Such dirs had 1 as their nlink count.
+#define INC_DIR_INODE_NLINK(i) { if (i->i_nlink != 1) INC_INODE_NLINK(i);}
+#define DEC_DIR_INODE_NLINK(i) { if (i->i_nlink != 1) DEC_INODE_NLINK(i);}
// directory item contains array of entry headers. This performs
// binary search through that array
@@ -518,7 +526,7 @@
retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
inode, 1/*visible*/);
if (retval) {
- inode->i_nlink--;
+ DEC_INODE_NLINK(inode);
reiserfs_update_sd (&th, inode);
pop_journal_writer(windex) ;
// FIXME: should we put iput here and have stat data deleted
@@ -569,7 +577,7 @@
retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
inode, 1/*visible*/);
if (retval) {
- inode->i_nlink--;
+ DEC_INODE_NLINK(inode);
reiserfs_update_sd (&th, inode);
pop_journal_writer(windex) ;
journal_end(&th, dir->i_sb, jbegin_count) ;
@@ -613,7 +621,7 @@
dentry, inode, &retval);
if (!inode) {
pop_journal_writer(windex) ;
- dir->i_nlink-- ;
+ DEC_DIR_INODE_NLINK(dir);
journal_end(&th, dir->i_sb, jbegin_count) ;
return retval;
}
@@ -627,7 +635,7 @@
retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
inode, 1/*visible*/);
if (retval) {
- inode->i_nlink = 0;
+ inode->u.reiserfs_i.i_nlink_real = inode->i_nlink = 0;
DEC_DIR_INODE_NLINK(dir);
reiserfs_update_sd (&th, inode);
pop_journal_writer(windex) ;
@@ -711,7 +719,7 @@
if ( inode->i_nlink != 2 && inode->i_nlink != 1 )
printk ("reiserfs_rmdir: empty directory has nlink != 2 (%d)\n", inode->i_nlink);
- inode->i_nlink = 0;
+ inode->u.reiserfs_i.i_nlink_real = inode->i_nlink = 0;
inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
reiserfs_update_sd (&th, inode);
@@ -780,14 +788,14 @@
if (!inode->i_nlink) {
printk("reiserfs_unlink: deleting nonexistent file (%s:%lu), %d\n",
kdevname(inode->i_dev), inode->i_ino, inode->i_nlink);
- inode->i_nlink = 1;
+ inode->u.reiserfs_i.i_nlink_real = inode->i_nlink = 1;
}
retval = reiserfs_cut_from_item (&th, &path, &(de.de_entry_key), dir, NULL, 0);
if (retval < 0)
goto end_unlink;
- inode->i_nlink--;
+ DEC_INODE_NLINK(inode);
inode->i_ctime = CURRENT_TIME;
reiserfs_update_sd (&th, inode);
@@ -868,7 +876,7 @@
retval = reiserfs_add_entry (&th, parent_dir, dentry->d_name.name, dentry->d_name.len,
inode, 1/*visible*/);
if (retval) {
- inode->i_nlink--;
+ DEC_INODE_NLINK(inode);
reiserfs_update_sd (&th, inode);
pop_journal_writer(windex) ;
journal_end(&th, parent_dir->i_sb, jbegin_count) ;
@@ -896,7 +904,7 @@
if (S_ISDIR(inode->i_mode))
return -EPERM;
- if (inode->i_nlink >= REISERFS_LINK_MAX) {
+ if (CAN_INCREASE_NLINK(inode)) {
//FIXME: sd_nlink is 32 bit for new files
return -EMLINK;
}
@@ -917,7 +925,7 @@
return retval;
}
- inode->i_nlink++;
+ INC_INODE_NLINK(inode);
ctime = CURRENT_TIME;
inode->i_ctime = ctime;
reiserfs_update_sd (&th, inode);
@@ -1157,9 +1165,9 @@
if (new_dentry_inode) {
// adjust link number of the victim
if (S_ISDIR(new_dentry_inode->i_mode)) {
- new_dentry_inode->i_nlink = 0;
+ new_dentry_inode->u.reiserfs_i.i_nlink_real = new_dentry_inode->i_nlink = 0;
} else {
- new_dentry_inode->i_nlink--;
+ DEC_INODE_NLINK(new_dentry_inode);
}
ctime = CURRENT_TIME;
new_dentry_inode->i_ctime = ctime;
===== include/linux/reiserfs_fs.h 1.22 vs edited =====
--- 1.22/include/linux/reiserfs_fs.h Tue Aug 20 13:40:53 2002
+++ edited/include/linux/reiserfs_fs.h Thu Sep 5 13:42:08 2002
@@ -1163,7 +1163,6 @@
#define INODE_PKEY(inode) ((struct key *)((inode)->u.reiserfs_i.i_key))
#define MAX_UL_INT 0xffffffff
-#define MAX_INT 0x7ffffff
#define MAX_US_INT 0xffff
// reiserfs version 2 has max offset 60 bits. Version 1 - 32 bit offset
@@ -1186,8 +1185,9 @@
#define MAX_FC_NUM MAX_US_INT
-/* the purpose is to detect overflow of an unsigned short */
-#define REISERFS_LINK_MAX (MAX_US_INT - 1000)
+/* Find maximal number, that nlink_t can hold. GCC is able to calculate this
+ value at compile time, so do not worry about extra CPU overhead. */
+#define REISERFS_LINK_MAX ((((nlink_t) -1) > 0)?~0:((1u<<(sizeof(nlink_t)*8-1))-1))
/* The following defines are used in reiserfs_insert_item and reiserfs_append_item */
===== include/linux/reiserfs_fs_i.h 1.8 vs edited =====
--- 1.8/include/linux/reiserfs_fs_i.h Fri Aug 9 19:22:34 2002
+++ edited/include/linux/reiserfs_fs_i.h Thu Sep 5 13:41:01 2002
@@ -53,6 +53,8 @@
** flushed */
unsigned long i_trans_id ;
unsigned long i_trans_index ;
+ unsigned int i_nlink_real; /* We store real nlink number since field in
+ struct inode is too short for us */
};
#endif
next prev parent reply other threads:[~2002-09-05 9:50 UTC|newest]
Thread overview: 44+ messages / expand[flat|nested] mbox.gz Atom feed top
2002-09-01 8:55 [PATCH] sparc32: wrong type of nlink_t Tomas Szepe
2002-09-01 8:52 ` David S. Miller
2002-09-01 9:44 ` Tomas Szepe
2002-09-04 20:18 ` Dave Kleikamp
2002-09-04 20:29 ` [reiserfs-dev] " Chris Mason
2002-09-04 23:34 ` David S. Miller
2002-09-06 8:52 ` David Woodhouse
2002-09-04 20:31 ` Hans Reiser
2002-09-04 21:18 ` Tomas Szepe
2002-09-04 21:44 ` Thunder from the hill
2002-09-04 21:57 ` Thunder from the hill
2002-09-04 23:35 ` David S. Miller
2002-09-05 0:36 ` Hans Reiser
2002-09-05 0:32 ` David S. Miller
2002-09-05 0:49 ` Chris Mason
2002-09-05 5:40 ` Tomas Szepe
2002-09-05 5:36 ` David S. Miller
2002-09-05 5:48 ` Tomas Szepe
2002-09-05 5:45 ` David S. Miller
2002-09-05 9:46 ` Nikita Danilov
2002-09-05 5:56 ` Oleg Drokin
2002-09-05 5:52 ` David S. Miller
2002-09-05 6:07 ` Oleg Drokin
2002-09-05 5:59 ` Tomas Szepe
2002-09-05 9:54 ` Oleg Drokin [this message]
2002-09-05 10:50 ` David S. Miller
2002-09-05 13:49 ` Oleg Drokin
2002-09-05 13:57 ` Oleg Drokin
2002-09-05 14:03 ` Chris Mason
2002-09-05 14:17 ` Oleg Drokin
2002-09-05 16:45 ` Chris Mason
2002-09-05 17:25 ` Oleg Drokin
2002-09-05 21:18 ` jw schultz
2002-09-05 22:02 ` Ragnar Kjørstad
2002-09-05 22:57 ` jw schultz
2002-09-06 0:01 ` Ragnar Kjørstad
2002-09-06 1:41 ` jw schultz
2002-09-06 2:29 ` Ragnar Kjørstad
2002-09-05 16:09 ` Dave Kleikamp
2002-09-05 16:13 ` Oleg Drokin
2002-09-05 17:58 ` Dave Kleikamp
2002-09-06 13:54 ` Oleg Drokin
2002-09-05 17:24 ` Linus Torvalds
2002-09-04 23:33 ` David S. Miller
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20020905135442.A19682@namesys.com \
--to=green@namesys.com \
--cc=davem@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linuxjfs@us.ibm.com \
--cc=marcelo@conectiva.com.br \
--cc=mason@suse.com \
--cc=reiser@namesys.com \
--cc=reiserfs-dev@namesys.com \
--cc=shaggy@austin.ibm.com \
--cc=szepe@pinerecords.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox