Skip to contentSkip to conversation

staging: erofs: fix an error handling in erofs_readdir()

PATCH Superseded Gao Xiang 8 participants 27 messages Aug 18, 2019 – Aug 20, 2019

Subsystems (3)
lore
Share & tools
lore

Gao Xiang

author
#0 lore raw

PATCH 1/1 staging: erofs: fix an error handling in erofs_readdir()

From: Gao Xiang <[email protected]>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Let's fix it now.

[1] https://lore.kernel.org/r/1746679415.68815.1566076790942.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>

---

Which is based on the following patch as well
https://lore.kernel.org/r/20190816071142.8633-1-gaoxiang25 at huawei.com/

and
https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao at aol.com/ can still be properly applied after this patch.

 drivers/staging/erofs/dir.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/staging/erofs/dir.c +6 -2
index 5f38382637e6..f2d7539589e4 100644
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
 
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH v2 1/1 staging: erofs: fix an error handling in erofs_readdir()

From: Gao Xiang <[email protected]>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Let's fix it now.

[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>

---

changelog from v1:
- fix the incorrect external link in commit message.

This patch is based on the following patch as well
https://lore.kernel.org/r/20190816071142.8633-1-gaoxiang25 at huawei.com/

and
https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao at aol.com/ can still be properly applied after this patch.

 drivers/staging/erofs/dir.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/staging/erofs/dir.c +6 -2
index 5f38382637e6..f2d7539589e4 100644
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
 
Signature · 1 line
2.17.1
#2 lore raw

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 11 lines · Gao Xiang
drivers/staging/erofs/dir.c +5 -2
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

author
#3 lore raw

Hi Willy,

Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

Thanks,
Gao Xiang

#4 lore raw
Quoted 33 lines · Gao Xiang
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

Quoted 5 lines

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

I don't think it matters whether it's due to a disk error or a corrupted image. We can't read the directory entry, so we should probably return -EFSCORRUPTED. Thinking about it some more, read_mapping_page() can also return -ENOMEM, so it should probably look something like this:

		err = 0;
		if (dentry_page == ERR_PTR(-ENOMEM))
			err = -ENOMEM;
		else if (IS_ERR(dentry_page)) {
			errln("fail to readdir of logical block %u of nid %llu",
			      i, EROFS_V(dir)->nid);
			err = -EFSCORRUPTED;
		}

		if (err)
			break;
author
#5 lore raw
Quoted 37 lines · Matthew Wilcox
Quoted 36 lines · Gao Xiang
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

Quoted 7 lines
Quoted 2 lines

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

I don't think it matters whether it's due to a disk error or a corrupted image. We can't read the directory entry, so we should probably return -EFSCORRUPTED. Thinking about it some more, read_mapping_page() can also return -ENOMEM, so it should probably look something like this:

OK, I will send the next version like what you described below. I realized that at first but I have no tendency to return EFSCORRUPTED or EIO here.

Thanks,
Gao Xiang

Quoted 12 lines
		err = 0;
		if (dentry_page == ERR_PTR(-ENOMEM))
			err = -ENOMEM;
		else if (IS_ERR(dentry_page)) {
			errln("fail to readdir of logical block %u of nid %llu",
			      i, EROFS_V(dir)->nid);
			err = -EFSCORRUPTED;
		}

		if (err)
			break;
author
#6 lore raw

PATCH 1/1 staging: erofs: fix an error handling in erofs_readdir()

From: Gao Xiang <[email protected]>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Let's fix it now.

[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>

---
changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

changelog from v1:
- fix the incorrect external link in commit message.

This patch is based on the following patch as well
https://lore.kernel.org/r/20190816071142.8633-1-gaoxiang25 at huawei.com/

and
https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao at aol.com/ can still be properly applied after this patch.

 drivers/staging/erofs/dir.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/staging/erofs/dir.c +11 -2
index 5f38382637e6..eb430a031b20 100644
@@ -82,8 +82,17 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (dentry_page == ERR_PTR(-ENOMEM)) {
+			errln("no memory to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -ENOMEM;
+			break;
+		} else if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -EFSCORRUPTED;
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
 
Signature · 1 line
2.17.1
#7 lore raw
I love your patch! Yet something to improve:

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-an-error-handling-in-erofs_readdir/20190818-191344
config: sparc64-allmodconfig (attached as .config)
compiler: sparc64-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=sparc64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging//erofs/dir.c: In function 'erofs_readdir':

>> drivers/staging//erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared (first use in this function)
       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~

drivers/staging//erofs/dir.c:110:11: note: each undeclared identifier is reported only once for each function it appears in

vim +/EFSCORRUPTED +110 drivers/staging//erofs/dir.c

    85	
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
    94		bool initial = true;
    95	
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;
   100	
   101			dentry_page = read_mapping_page(mapping, i, NULL);
   102			if (dentry_page == ERR_PTR(-ENOMEM)) {
   103				errln("no memory to readdir of logical block %u of nid %llu",
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of nid %llu",
   109				      i, EROFS_V(dir)->nid);
 > 110				err = -EFSCORRUPTED;

111 break;
112 }
113
114 de = (struct erofs_dirent *)kmap(dentry_page);
115
116 nameoff = le16_to_cpu(de->nameoff);
117
118 if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
119 nameoff >= PAGE_SIZE)) {
120 errln("%s, invalid de[0].nameoff %u",
121 __func__, nameoff);
122
123 err = -EIO;
124 goto skip_this;
125 }
126
127 maxsize = min_t(unsigned int,
128 dirsize - ctx->pos + ofs, PAGE_SIZE);
129
130 /* search dirents at the arbitrary position */
131 if (unlikely(initial)) {
132 initial = false;
133
134 ofs = roundup(ofs, sizeof(struct erofs_dirent));
135 if (unlikely(ofs >= nameoff))
136 goto skip_this;
137 }
138
139 err = erofs_fill_dentries(ctx, de, &ofs, nameoff, maxsize);
140 skip_this:
141 kunmap(dentry_page);
142
143 put_page(dentry_page);
144
145 ctx->pos = blknr_to_addr(i) + ofs;
146
147 if (unlikely(err))
148 break;
149 ++i;
150 ofs = 0;
151 }
152 return err < 0 ? err : 0;
153 }
154

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#8 lore raw
In reply to Gao Xiang
I love your patch! Yet something to improve:

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-an-error-handling-in-erofs_readdir/20190818-191344
config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

>> drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared (first use in this function); did you mean 'FS_NRSUPER'?
       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is reported only once for each function it appears in

vim +110 drivers/staging/erofs/dir.c

    85	
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
    94		bool initial = true;
    95	
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;
   100	
   101			dentry_page = read_mapping_page(mapping, i, NULL);
   102			if (dentry_page == ERR_PTR(-ENOMEM)) {
   103				errln("no memory to readdir of logical block %u of nid %llu",
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of nid %llu",
   109				      i, EROFS_V(dir)->nid);
 > 110				err = -EFSCORRUPTED;

111 break;
112 }
113
114 de = (struct erofs_dirent *)kmap(dentry_page);
115
116 nameoff = le16_to_cpu(de->nameoff);
117
118 if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
119 nameoff >= PAGE_SIZE)) {
120 errln("%s, invalid de[0].nameoff %u",
121 __func__, nameoff);
122
123 err = -EIO;
124 goto skip_this;
125 }
126
127 maxsize = min_t(unsigned int,
128 dirsize - ctx->pos + ofs, PAGE_SIZE);
129
130 /* search dirents at the arbitrary position */
131 if (unlikely(initial)) {
132 initial = false;
133
134 ofs = roundup(ofs, sizeof(struct erofs_dirent));
135 if (unlikely(ofs >= nameoff))
136 goto skip_this;
137 }
138
139 err = erofs_fill_dentries(ctx, de, &ofs, nameoff, maxsize);
140 skip_this:
141 kunmap(dentry_page);
142
143 put_page(dentry_page);
144
145 ctx->pos = blknr_to_addr(i) + ofs;
146
147 if (unlikely(err))
148 break;
149 ++i;
150 ofs = 0;
151 }
152 return err < 0 ? err : 0;
153 }
154

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

author
#9 lore raw
Quoted 7 lines · kbuild test robot

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

... those patches should be applied to staging tree since linux-next has not been updated yet...

Thanks,
Gao Xiang

Quoted 98 lines

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-an-error-handling-in-erofs_readdir/20190818-191344
config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag
Reported-by: kbuild test robot <lkp at intel.com>

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

Quoted 1 line
Quoted 1 line

drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared (first use in this function); did you mean 'FS_NRSUPER'?

       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is reported only once for each function it appears in

vim +110 drivers/staging/erofs/dir.c

    85	
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
    94		bool initial = true;
    95	
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;
   100	
   101			dentry_page = read_mapping_page(mapping, i, NULL);
   102			if (dentry_page == ERR_PTR(-ENOMEM)) {
   103				errln("no memory to readdir of logical block %u of nid %llu",
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of nid %llu",
   109				      i, EROFS_V(dir)->nid);
Quoted 1 line

110 err = -EFSCORRUPTED;

   111				break;
   112			}
   113	
   114			de = (struct erofs_dirent *)kmap(dentry_page);
   115	
   116			nameoff = le16_to_cpu(de->nameoff);
   117	
   118			if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
   119				     nameoff >= PAGE_SIZE)) {
   120				errln("%s, invalid de[0].nameoff %u",
   121				      __func__, nameoff);
   122	
   123				err = -EIO;
   124				goto skip_this;
   125			}
   126	
   127			maxsize = min_t(unsigned int,
   128					dirsize - ctx->pos + ofs, PAGE_SIZE);
   129	
   130			/* search dirents at the arbitrary position */
   131			if (unlikely(initial)) {
   132				initial = false;
   133	
   134				ofs = roundup(ofs, sizeof(struct erofs_dirent));
   135				if (unlikely(ofs >= nameoff))
   136					goto skip_this;
   137			}
   138	
   139			err = erofs_fill_dentries(ctx, de, &ofs, nameoff, maxsize);
   140	skip_this:
   141			kunmap(dentry_page);
   142	
   143			put_page(dentry_page);
   144	
   145			ctx->pos = blknr_to_addr(i) + ofs;
   146	
   147			if (unlikely(err))
   148				break;
   149			++i;
   150			ofs = 0;
   151		}
   152		return err < 0 ? err : 0;
   153	}
   154	

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#10 lore raw
Quoted 11 lines · Gao Xiang
Quoted 7 lines · kbuild test robot

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

... those patches should be applied to staging tree since linux-next has not been updated yet...

thanks for the feedback, we will consider this to our todo list.

Quoted 104 lines

Thanks,
Gao Xiang

Quoted 98 lines

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-an-error-handling-in-erofs_readdir/20190818-191344
config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

Quoted 1 line
Quoted 1 line

drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared (first use in this function); did you mean 'FS_NRSUPER'?

       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is reported only once for each function it appears in

vim +110 drivers/staging/erofs/dir.c

    85	
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
    94		bool initial = true;
    95	
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;
   100	
   101			dentry_page = read_mapping_page(mapping, i, NULL);
   102			if (dentry_page == ERR_PTR(-ENOMEM)) {
   103				errln("no memory to readdir of logical block %u of nid %llu",
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of nid %llu",
   109				      i, EROFS_V(dir)->nid);
Quoted 1 line

110 err = -EFSCORRUPTED;

   111				break;
   112			}
   113	
   114			de = (struct erofs_dirent *)kmap(dentry_page);
   115	
   116			nameoff = le16_to_cpu(de->nameoff);
   117	
   118			if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
   119				     nameoff >= PAGE_SIZE)) {
   120				errln("%s, invalid de[0].nameoff %u",
   121				      __func__, nameoff);
   122	
   123				err = -EIO;
   124				goto skip_this;
   125			}
   126	
   127			maxsize = min_t(unsigned int,
   128					dirsize - ctx->pos + ofs, PAGE_SIZE);
   129	
   130			/* search dirents at the arbitrary position */
   131			if (unlikely(initial)) {
   132				initial = false;
   133	
   134				ofs = roundup(ofs, sizeof(struct erofs_dirent));
   135				if (unlikely(ofs >= nameoff))
   136					goto skip_this;
   137			}
   138	
   139			err = erofs_fill_dentries(ctx, de, &ofs, nameoff, maxsize);
   140	skip_this:
   141			kunmap(dentry_page);
   142	
   143			put_page(dentry_page);
   144	
   145			ctx->pos = blknr_to_addr(i) + ofs;
   146	
   147			if (unlikely(err))
   148				break;
   149			++i;
   150			ofs = 0;
   151		}
   152		return err < 0 ? err : 0;
   153	}
   154	

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#11 lore raw

Hi Philip,

Quoted 6 lines · Philip Li
Quoted 5 lines · Gao Xiang
Quoted 4 lines · kbuild test robot

Hi Gao,

I love your patch! Yet something to improve:

Quoted 7 lines
Quoted 6 lines
Quoted 3 lines

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

... those patches should be applied to staging tree since linux-next has not been updated yet...

thanks for the feedback, we will consider this to our todo list.

Yes, many confusing reports anyway...
(Just my personal suggestion, maybe we can add some hints on the patch email to indicate which tree can be applied successfully for ci in the future...)

Thanks,
Gao Xiang

Quoted 105 lines
Quoted 105 lines

Thanks,
Gao Xiang

Quoted 98 lines

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-an-error-handling-in-erofs_readdir/20190818-191344
config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

Quoted 1 line
Quoted 1 line

drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared (first use in this function); did you mean 'FS_NRSUPER'?

       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is reported only once for each function it appears in

vim +110 drivers/staging/erofs/dir.c

    85	
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
    94		bool initial = true;
    95	
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;
   100	
   101			dentry_page = read_mapping_page(mapping, i, NULL);
   102			if (dentry_page == ERR_PTR(-ENOMEM)) {
   103				errln("no memory to readdir of logical block %u of nid %llu",
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of nid %llu",
   109				      i, EROFS_V(dir)->nid);
Quoted 1 line

110 err = -EFSCORRUPTED;

   111				break;
   112			}
   113	
   114			de = (struct erofs_dirent *)kmap(dentry_page);
   115	
   116			nameoff = le16_to_cpu(de->nameoff);
   117	
   118			if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
   119				     nameoff >= PAGE_SIZE)) {
   120				errln("%s, invalid de[0].nameoff %u",
   121				      __func__, nameoff);
   122	
   123				err = -EIO;
   124				goto skip_this;
   125			}
   126	
   127			maxsize = min_t(unsigned int,
   128					dirsize - ctx->pos + ofs, PAGE_SIZE);
   129	
   130			/* search dirents at the arbitrary position */
   131			if (unlikely(initial)) {
   132				initial = false;
   133	
   134				ofs = roundup(ofs, sizeof(struct erofs_dirent));
   135				if (unlikely(ofs >= nameoff))
   136					goto skip_this;
   137			}
   138	
   139			err = erofs_fill_dentries(ctx, de, &ofs, nameoff, maxsize);
   140	skip_this:
   141			kunmap(dentry_page);
   142	
   143			put_page(dentry_page);
   144	
   145			ctx->pos = blknr_to_addr(i) + ofs;
   146	
   147			if (unlikely(err))
   148				break;
   149			++i;
   150			ofs = 0;
   151		}
   152		return err < 0 ? err : 0;
   153	}
   154	

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#12 lore raw
Quoted 16 lines

Subject: Re: [PATCH] staging: erofs: fix an error handling in erofs_readdir()

Hi Philip,

Quoted 9 lines · Philip Li
Quoted 8 lines · Gao Xiang
Quoted 7 lines · kbuild test robot

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help

improve the system]

Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 3 lines
Quoted 2 lines

... those patches should be applied to staging tree since linux-next has not been updated yet...

thanks for the feedback, we will consider this to our todo list.

Yes, many confusing reports anyway...
(Just my personal suggestion, maybe we can add some hints on the patch email to indicate which tree can be applied successfully for ci in the future...)

thanks, this is good idea. On the other side, we support to add --base option to git format-patch to automatically suggest the base, refer to https://stackoverflow.com/a/37406982 for detail. Meanwhile, we will enhance the internal logic to find suitable base if possible.

Quoted 119 lines

Thanks,
Gao Xiang

Quoted 7 lines
Quoted 7 lines

Thanks,
Gao Xiang

Quoted 2 lines

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-

an-error-handling-in-erofs_readdir/20190818-191344

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0 reproduce:
wget https://raw.githubusercontent.com/intel/lkp-

tests/master/sbin/make.cross -O ~/bin/make.cross

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

chmod +x ~/bin/make.cross
# save the attached .config to linux build tree GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

Quoted 1 line
Quoted 1 line

drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared

(first use in this function); did you mean 'FS_NRSUPER'?

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is

reported only once for each function it appears in

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

vim +110 drivers/staging/erofs/dir.c

    85
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
    94		bool initial = true;
    95
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;

100
101 dentry_page = read_mapping_page(mapping, i,

NULL);

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

102 if (dentry_page == ERR_PTR(-ENOMEM)) {
103 errln("no memory to readdir of logical

block %u of nid %llu",

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of

nid %llu",

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
   109				      i, EROFS_V(dir)->nid);
Quoted 1 line

110 err = -EFSCORRUPTED;

   111				break;
   112			}
   113
   114			de = (struct erofs_dirent *)kmap(dentry_page);
   115
   116			nameoff = le16_to_cpu(de->nameoff);
   117
   118			if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
   119				     nameoff >= PAGE_SIZE)) {
   120				errln("%s, invalid de[0].nameoff %u",
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
   121				      __func__, nameoff);
   122
   123				err = -EIO;
   124				goto skip_this;
   125			}
   126
   127			maxsize = min_t(unsigned int,
   128					dirsize - ctx->pos + ofs,

PAGE_SIZE);

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
   129
   130			/* search dirents at the arbitrary position */
   131			if (unlikely(initial)) {
   132				initial = false;
   133
   134				ofs = roundup(ofs, sizeof(struct

erofs_dirent));

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
   135				if (unlikely(ofs >= nameoff))
   136					goto skip_this;
   137			}
   138
   139			err = erofs_fill_dentries(ctx, de, &ofs, nameoff,

maxsize);

Quoted 21 lines
Quoted 21 lines
Quoted 19 lines
   140	skip_this:
   141			kunmap(dentry_page);
   142
   143			put_page(dentry_page);
   144
   145			ctx->pos = blknr_to_addr(i) + ofs;
   146
   147			if (unlikely(err))
   148				break;
   149			++i;
   150			ofs = 0;
   151		}
   152		return err < 0 ? err : 0;
   153	}
   154

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#13 lore raw
Quoted 18 lines · Philip
Quoted 18 lines

Subject: Re: [PATCH] staging: erofs: fix an error handling in erofs_readdir()

Hi Philip,

Quoted 9 lines · Philip Li
Quoted 8 lines · Gao Xiang
Quoted 7 lines · kbuild test robot

Hi Gao,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc4 next-20190816]
[if your patch is applied to the wrong git tree, please drop us a note to help

improve the system]

Quoted 3 lines
Quoted 3 lines

... those patches should be applied to staging tree since linux-next has not been updated yet...

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

thanks for the feedback, we will consider this to our todo list.

Yes, many confusing reports anyway...
(Just my personal suggestion, maybe we can add some hints on the patch email to indicate which tree can be applied successfully for ci in the future...)

thanks, this is good idea. On the other side, we support to add --base option to git format-patch to automatically suggest the base, refer to https://stackoverflow.com/a/37406982 for detail.

Thanks for your information :) It seems a new patch format, I will take a try later.

Thanks,
Gao Xiang

Quoted 121 lines

Meanwhile, we will enhance the internal logic to find suitable base if possible.

Quoted 119 lines

Thanks,
Gao Xiang

Quoted 7 lines
Quoted 7 lines

Thanks,
Gao Xiang

Quoted 2 lines

url: https://github.com/0day-ci/linux/commits/Gao-Xiang/staging-erofs-fix-

an-error-handling-in-erofs_readdir/20190818-191344

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

config: arm64-allyesconfig (attached as .config)
compiler: aarch64-linux-gcc (GCC) 7.4.0 reproduce:
wget https://raw.githubusercontent.com/intel/lkp-

tests/master/sbin/make.cross -O ~/bin/make.cross

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

chmod +x ~/bin/make.cross
# save the attached .config to linux build tree GCC_VERSION=7.4.0 make.cross ARCH=arm64

If you fix the issue, kindly add following tag

All errors (new ones prefixed by >>):

drivers/staging/erofs/dir.c: In function 'erofs_readdir':

Quoted 1 line
Quoted 1 line

drivers/staging/erofs/dir.c:110:11: error: 'EFSCORRUPTED' undeclared

(first use in this function); did you mean 'FS_NRSUPER'?

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
       err = -EFSCORRUPTED;
              ^~~~~~~~~~~~
              FS_NRSUPER

drivers/staging/erofs/dir.c:110:11: note: each undeclared identifier is

reported only once for each function it appears in

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

vim +110 drivers/staging/erofs/dir.c

    85
    86	static int erofs_readdir(struct file *f, struct dir_context *ctx)
    87	{
    88		struct inode *dir = file_inode(f);
    89		struct address_space *mapping = dir->i_mapping;
    90		const size_t dirsize = i_size_read(dir);
    91		unsigned int i = ctx->pos / EROFS_BLKSIZ;
    92		unsigned int ofs = ctx->pos % EROFS_BLKSIZ;
    93		int err = 0;
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
    94		bool initial = true;
    95
    96		while (ctx->pos < dirsize) {
    97			struct page *dentry_page;
    98			struct erofs_dirent *de;
    99			unsigned int nameoff, maxsize;

100
101 dentry_page = read_mapping_page(mapping, i,

NULL);

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

102 if (dentry_page == ERR_PTR(-ENOMEM)) {
103 errln("no memory to readdir of logical

block %u of nid %llu",

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
   104				      i, EROFS_V(dir)->nid);
   105				err = -ENOMEM;
   106				break;
   107			} else if (IS_ERR(dentry_page)) {
   108				errln("fail to readdir of logical block %u of

nid %llu",

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
   109				      i, EROFS_V(dir)->nid);
Quoted 1 line

110 err = -EFSCORRUPTED;

   111				break;
   112			}
   113
   114			de = (struct erofs_dirent *)kmap(dentry_page);
   115
   116			nameoff = le16_to_cpu(de->nameoff);
   117
   118			if (unlikely(nameoff < sizeof(struct erofs_dirent) ||
   119				     nameoff >= PAGE_SIZE)) {
   120				errln("%s, invalid de[0].nameoff %u",
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
   121				      __func__, nameoff);
   122
   123				err = -EIO;
   124				goto skip_this;
   125			}
   126
   127			maxsize = min_t(unsigned int,
   128					dirsize - ctx->pos + ofs,

PAGE_SIZE);

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
   129
   130			/* search dirents at the arbitrary position */
   131			if (unlikely(initial)) {
   132				initial = false;
   133
   134				ofs = roundup(ofs, sizeof(struct

erofs_dirent));

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
   135				if (unlikely(ofs >= nameoff))
   136					goto skip_this;
   137			}
   138
   139			err = erofs_fill_dentries(ctx, de, &ofs, nameoff,

maxsize);

Quoted 21 lines
Quoted 21 lines
Quoted 19 lines
   140	skip_this:
   141			kunmap(dentry_page);
   142
   143			put_page(dentry_page);
   144
   145			ctx->pos = blknr_to_addr(i) + ofs;
   146
   147			if (unlikely(err))
   148				break;
   149			++i;
   150			ofs = 0;
   151		}
   152		return err < 0 ? err : 0;
   153	}
   154

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

author
#14 lore raw

PATCH RESEND v3 1/1 staging: erofs: fix an error handling in erofs_readdir()

In reply to Gao Xiang

From: Gao Xiang <[email protected]>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Let's fix it now.

[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>

---
[RESEND] --> add the missing v3 version in subject, no logic change.

changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

changelog from v1:
- fix the incorrect external link in commit message.

This patch is based on the following patch as well
https://lore.kernel.org/r/20190816071142.8633-1-gaoxiang25 at huawei.com/

and
https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao at aol.com/ can still be properly applied after this patch.

 drivers/staging/erofs/dir.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/staging/erofs/dir.c +11 -2
index 5f38382637e6..eb430a031b20 100644
@@ -82,8 +82,17 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (dentry_page == ERR_PTR(-ENOMEM)) {
+			errln("no memory to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -ENOMEM;
+			break;
+		} else if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -EFSCORRUPTED;
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
 
Signature · 1 line
2.17.1
#15 lore raw

----- Urspr?ngliche Mail -----

Quoted 2 lines

changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

erofs does not define EFSCORRUPTED, so the build fails. At least on Linus' tree as of today.

Thanks,
//richard

author
#16 lore raw

Hi Richard,

Quoted 6 lines · Richard Weinberger

----- Urspr?ngliche Mail -----

Quoted 2 lines

changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

erofs does not define EFSCORRUPTED, so the build fails. At least on Linus' tree as of today.

Thanks for your reply :)

I write all patches based on staging tree and do more cleanups further than Linus' tree, EFSCORRUPTED was already introduced by Pavel days before...
https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git/commit/?h=staging-testing&id=a6b9b1d5eae61a68085030e50d56265dec5baa94

which can be fetched from
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git -b staging-next

Thanks,
Gao Xiang

Quoted 3 lines

Thanks,
//richard

author
#17 lore raw

Re: [PATCH v3 RESEND] staging: erofs: fix an error handling in erofs_readdir()y

Quoted 7 lines · Gao Xiang via Linux-erofs

Hi Richard,

Quoted 4 lines · Richard Weinberger

----- Urspr?ngliche Mail -----

Quoted 2 lines

changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

Quoted 7 lines
Quoted 2 lines

erofs does not define EFSCORRUPTED, so the build fails. At least on Linus' tree as of today.

Thanks for your reply :)

I write all patches based on staging tree and do more cleanups further than Linus' tree, EFSCORRUPTED was already introduced by Pavel days before...

Sorry, I mean "introduced which was suggested by Paval"...

Quoted 11 lines

https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git/commit/?h=staging-testing&id=a6b9b1d5eae61a68085030e50d56265dec5baa94

which can be fetched from
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git -b staging-next

Thanks,
Gao Xiang

Quoted 3 lines

Thanks,
//richard

maintainer
#18 lore raw

PATCH RESEND v3 1/1 staging: erofs: fix an error handling in erofs_readdir()

In reply to Gao Xiang

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag, fixing commit: 3aa8ec716e52 staging: erofs: add directory operations.

The bot has tested the following trees: v5.2.9, v4.19.67.

v5.2.9: Build failed! Errors:
drivers/staging/erofs/dir.c:111:11: error: ???EFSCORRUPTED??? undeclared (first use in this function); did you mean ???FS_NRSUPER????

v4.19.67: Build failed! Errors:
drivers/staging/erofs/dir.c:111:11: error: ???EFSCORRUPTED??? undeclared (first use in this function); did you mean ???FS_NRSUPER????

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

--
Thanks,
Sasha

Reviewed-by on v3 1/1
#19 lore raw
In reply to Gao Xiang
Quoted 11 lines · Gao Xiang

From: Gao Xiang <gaoxiang25 at huawei.com>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Quoted 8 lines

Let's fix it now.

[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>
Reviewed-by Chao Yu <yuchao0 at huawei.com>

Thanks,

#20 lore raw
In reply to Gao Xiang

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 87

Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +3 -0
+		if (dentry_page == ERR_PTR(-ENOMEM)) {
+			errln("no memory to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);

I don't think you need the error message. If we get a memory allocation failure, there's already going to be a lot of spew in the logs from the mm system. And if we do fail to allocate memory, we don't need to know the logical block number or the nid -- it has nothiing to do with those; the system simply ran out of memory.

Quoted 14 lines
drivers/staging/erofs/dir.c +8 -0
+			err = -ENOMEM;
+			break;
+		} else if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -EFSCORRUPTED;
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
Signature · 2 lines
-- 
2.17.1
author
#21 lore raw
Quoted 10 lines · Matthew Wilcox

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 87

Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +3 -0
+		if (dentry_page == ERR_PTR(-ENOMEM)) {
+			errln("no memory to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);

I don't think you need the error message. If we get a memory allocation failure, there's already going to be a lot of spew in the logs from the mm system. And if we do fail to allocate memory, we don't need to know the logical block number or the nid -- it has nothiing to do with those; the system simply ran out of memory.

OK, I agree with you. There is a messy of messages when memory allocation fail.

Since I don't really care apart from crashing or hanging the kernel, I will resend the patch to make you and Chao happy... :)

Thanks,
Gao Xiang

Quoted 15 lines
Quoted 15 lines
drivers/staging/erofs/dir.c +8 -0
+			err = -ENOMEM;
+			break;
+		} else if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -EFSCORRUPTED;
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
Signature · 2 lines
-- 
2.17.1
author
#22 lore raw

PATCH v4 1/1 staging: erofs: fix an error handling in erofs_readdir()

From: Gao Xiang <[email protected]>

Richard observed a forever loop of erofs_read_raw_page() [1] which can be generated by forcely setting ->u.i_blkaddr to 0xdeadbeef (as my understanding block layer can handle access beyond end of device correctly).

After digging into that, it seems the problem is highly related with directories and then I found the root cause is an improper error handling in erofs_readdir().

Let's fix it now.

[1] https://lore.kernel.org/r/1163995781.68824.1566084358245.JavaMail.zimbra at nod.at/

Reported-by Richard Weinberger <richard at nod.at>
Fixes 3aa8ec716e52 ("staging: erofs: add directory operations")
Cc <stable at vger.kernel.org> # 4.19+
Reviewed-by Chao Yu <yuchao0 at huawei.com>
Signed-off-by Gao Xiang <gaoxiang25 at huawei.com>

---
changelog from v3:
- kill message when memory allocation fails as suggested by Matthew;

[RESEND] --> add the missing v3 version in subject, no logic change.

changelog from v2:
- transform EIO to EFSCORRUPTED as suggested by Matthew;

changelog from v1:
- fix the incorrect external link in commit message.

This patch is based on staging-testing tree and
https://lore.kernel.org/r/20190817082313.21040-1-hsiangkao at aol.com/ can still be properly applied after this patch.

 drivers/staging/erofs/dir.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
drivers/staging/erofs/dir.c +9 -2
index 5f38382637e6..77ef856df9f3 100644
@@ -82,8 +82,15 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (dentry_page == ERR_PTR(-ENOMEM)) {
+			err = -ENOMEM;
+			break;
+		} else if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = -EFSCORRUPTED;
+			break;
+		}
 
 		de = (struct erofs_dirent *)kmap(dentry_page);
 
Signature · 1 line
2.17.1
maintainer
#23 lore raw

PATCH v4 1/1 staging: erofs: fix an error handling in erofs_readdir()

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag, fixing commit: 3aa8ec716e52 staging: erofs: add directory operations.

The bot has tested the following trees: v5.2.9, v4.19.67.

v5.2.9: Build failed! Errors:
drivers/staging/erofs/dir.c:109:11: error: ???EFSCORRUPTED??? undeclared (first use in this function); did you mean ???FS_NRSUPER????

v4.19.67: Build failed! Errors:
drivers/staging/erofs/dir.c:109:11: error: ???EFSCORRUPTED??? undeclared (first use in this function); did you mean ???FS_NRSUPER????

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

--
Thanks,
Sasha

#24 lore raw
In reply to Matthew Wilcox
Quoted 45 lines · Matthew Wilcox
Quoted 33 lines · Gao Xiang
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

Quoted 5 lines

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

I don't think it matters whether it's due to a disk error or a corrupted image. We can't read the directory entry, so we should probably return -EFSCORRUPTED. Thinking about it some more, read_mapping_page() can also return -ENOMEM, so it should probably look something like this:

Quoted 7 lines
		err = 0;
		if (dentry_page == ERR_PTR(-ENOMEM))
			err = -ENOMEM;
		else if (IS_ERR(dentry_page)) {
			errln("fail to readdir of logical block %u of nid %llu",
			      i, EROFS_V(dir)->nid);
			err = -EFSCORRUPTED;

Well, if there is real IO error happen under filesystem, we should return -EIO instead of EFSCORRUPTED?

The right fix may be that doing sanity check on on-disk blkaddr, and return -EFSCORRUPTED if the blkaddr is invalid and propagate the error to its caller erofs_readdir(), IIUC below error info.

Quoted 6 lines
[36297.354090] attempt to access beyond end of device
[36297.354098] loop17: rw=0, want=29887428984, limit=1953128
[36297.354107] attempt to access beyond end of device
[36297.354109] loop17: rw=0, want=29887428480, limit=1953128
[36301.827234] attempt to access beyond end of device
[36301.827243] loop17: rw=0, want=29887428480, limit=1953128

Thanks,

Quoted 5 lines
		}

		if (err)
			break;
author
#25 lore raw

Hi Chao,

Quoted 54 lines · Chao Yu
Quoted 45 lines · Matthew Wilcox
Quoted 33 lines · Gao Xiang
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

Quoted 5 lines

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

I don't think it matters whether it's due to a disk error or a corrupted image. We can't read the directory entry, so we should probably return -EFSCORRUPTED. Thinking about it some more, read_mapping_page() can also return -ENOMEM, so it should probably look something like this:

Quoted 7 lines
		err = 0;
		if (dentry_page == ERR_PTR(-ENOMEM))
			err = -ENOMEM;
		else if (IS_ERR(dentry_page)) {
			errln("fail to readdir of logical block %u of nid %llu",
			      i, EROFS_V(dir)->nid);
			err = -EFSCORRUPTED;
Quoted 6 lines

Well, if there is real IO error happen under filesystem, we should return -EIO instead of EFSCORRUPTED?

The right fix may be that doing sanity check on on-disk blkaddr, and return -EFSCORRUPTED if the blkaddr is invalid and propagate the error to its caller erofs_readdir(), IIUC below error info.

In my thought, I actually don't care what is actually returned (In other words, I have no tendency about EFSCORRUPTED / EIO) as long as it behaves normal for corrupted image.

A little concern is that I have no idea whether all user problems can handle EUCLEAN properly.

I don't want to limit blkaddr as what ->blocks recorded in erofs_super_block since it's already used for our hotpatching approach and that field is only used for statfs() for users to know its visible size, and block layer will block such invalid block access.

All in all, that is minor. I think we can fix as what Matthew said.

Thanks,
Gao Xiang

Quoted 15 lines
Quoted 7 lines
[36297.354090] attempt to access beyond end of device
[36297.354098] loop17: rw=0, want=29887428984, limit=1953128
[36297.354107] attempt to access beyond end of device
[36297.354109] loop17: rw=0, want=29887428480, limit=1953128
[36301.827234] attempt to access beyond end of device
[36301.827243] loop17: rw=0, want=29887428480, limit=1953128

Thanks,

Quoted 5 lines
		}

		if (err)
			break;
#26 lore raw

Hi Xiang,

Quoted 60 lines · Gao Xiang

Hi Chao,

Quoted 57 lines · Chao Yu
Quoted 45 lines · Matthew Wilcox
Quoted 33 lines · Gao Xiang
Quoted 4 lines · Matthew Wilcox
Quoted 3 lines · Gao Xiang
drivers/staging/erofs/dir.c +0 -0
@@ -82,8 +82,12 @@ static int erofs_readdir(struct file *f, struct dir_context *ctx)
 		unsigned int nameoff, maxsize;
 
Quoted 13 lines

drivers/staging/erofs/dir.c static int erofs_readdir(struct file *f, struct dir_context *ctx) · line 89

Quoted 8 lines
drivers/staging/erofs/dir.c +5 -2
 		dentry_page = read_mapping_page(mapping, i, NULL);
-		if (IS_ERR(dentry_page))
-			continue;
+		if (IS_ERR(dentry_page)) {
+			errln("fail to readdir of logical block %u of nid %llu",
+			      i, EROFS_V(dir)->nid);
+			err = PTR_ERR(dentry_page);
+			break;

I don't think you want to use the errno that came back from read_mapping_page() (which is, I think, always going to be -EIO). Rather you want -EFSCORRUPTED, at least if I understand the recent patches to ext2/ext4/f2fs/xfs/...

Thanks for your reply and noticing this. :)

Yes, as I talked with you about read_mapping_page() in a xfs related topic earlier, I think I fully understand what returns here.

I actually had some concern about that before sending out this patch. You know the status is
PG_uptodate is not set and PG_error is set here.

But we cannot know it is actually a disk read error or due to corrupted images (due to lack of page flags or some status, and I think it could be a waste of page structure space for such corrupted image or disk error)...

Quoted 5 lines

And some people also like propagate errors from insiders... (and they could argue about err = -EFSCORRUPTED as well..)

I'd like hear your suggestion about this after my words above? still return -EFSCORRUPTED?

I don't think it matters whether it's due to a disk error or a corrupted image. We can't read the directory entry, so we should probably return -EFSCORRUPTED. Thinking about it some more, read_mapping_page() can also return -ENOMEM, so it should probably look something like this:

Quoted 7 lines
		err = 0;
		if (dentry_page == ERR_PTR(-ENOMEM))
			err = -ENOMEM;
		else if (IS_ERR(dentry_page)) {
			errln("fail to readdir of logical block %u of nid %llu",
			      i, EROFS_V(dir)->nid);
			err = -EFSCORRUPTED;

Well, if there is real IO error happen under filesystem, we should return -EIO instead of EFSCORRUPTED?

Quoted 7 lines
Quoted 3 lines

The right fix may be that doing sanity check on on-disk blkaddr, and return -EFSCORRUPTED if the blkaddr is invalid and propagate the error to its caller erofs_readdir(), IIUC below error info.

In my thought, I actually don't care what is actually returned (In other words, I have no tendency about EFSCORRUPTED / EIO) as long as it behaves normal for corrupted image.

I doubt that user can and be willing to handle different error code in there error path.

Quoted 3 lines

A little concern is that I have no idea whether all user problems can handle EUCLEAN properly.

Yes, I can see it's widely used in fileystem, I guess it would be better to update manual of common fs interfaces to let user be aware of the meaning of it.

Quoted 8 lines

I don't want to limit blkaddr as what ->blocks recorded in erofs_super_block since it's already used for our hotpatching approach and that field is only used for statfs() for users to know its visible size, and block layer will block such invalid block access.

All in all, that is minor. I think we can fix as what Matthew said.

Yeah, as we discuss offline, we need keep flexibility on current version for android, and maybe we can add a feature to check blkaddr validation later.

Thanks,

Quoted 19 lines

Thanks,
Gao Xiang

Quoted 15 lines
Quoted 7 lines
[36297.354090] attempt to access beyond end of device
[36297.354098] loop17: rw=0, want=29887428984, limit=1953128
[36297.354107] attempt to access beyond end of device
[36297.354109] loop17: rw=0, want=29887428480, limit=1953128
[36301.827234] attempt to access beyond end of device
[36301.827243] loop17: rw=0, want=29887428480, limit=1953128

Thanks,

Quoted 5 lines
		}

		if (err)
			break;