FAILED: patch "[PATCH] erofs: fix unexpected EIO under memory pressure" failed to apply to 6.18-stable tree

PATCH New [email protected] 3 participants 5 messages Dec 29, 2025 – Jan 5

Subsystems (2)

Stable branch EROFS file system

lore

Share & tools

lore

[email protected]

author maintainer

2025-12-29 11:35 UTC #0 lore raw

The patch below does not apply to the 6.18-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.18.y
git checkout FETCH_HEAD
git cherry-pick -x 4012d78562193ef5eb613bad4b0c0fa187637cfe
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2025122915-kitchen-june-49ec@gregkh' --subject-prefix 'PATCH 6.18.y' HEAD^..

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 4012d78562193ef5eb613bad4b0c0fa187637cfe Mon Sep 17 00:00:00 2001
From: Junbeom Yeom <[email protected]>
Date: Fri, 19 Dec 2025 21:40:31 +0900
Subject: [PATCH] erofs: fix unexpected EIO under memory pressure

erofs readahead could fail with ENOMEM under the memory pressure because it tries to alloc_page with GFP_NOWAIT | GFP_NORETRY, while GFP_KERNEL for a regular read. And if readahead fails (with non-uptodate folios), the original request will then fall back to synchronous read, and .read_folio() should return appropriate errnos.

However, in scenarios where readahead and read operations compete, read operation could return an unintended EIO because of an incorrect error propagation.

To resolve this, this patch modifies the behavior so that, when the PCL is for read(which means pcl.besteffort is true), it attempts actual decompression instead of propagating the privios error except initial EIO.

- Page size: 4K
- The original size of FileA: 16K
- Compress-ratio per PCL: 50% (Uncompressed 8K -> Compressed 4K) [page0, page1] [page2, page3]
[PCL0]---------[PCL1]

- functions declaration:
. pread(fd, buf, count, offset)
. readahead(fd, offset, count)
- Thread A tries to read the last 4K
- Thread B tries to do readahead 8K from 4K
- RA, besteffort == false
- R, besteffort == true

pread(FileA, buf, 4K, 12K)
do readahead(page3) // failed with ENOMEM
wait_lock(page3)

    if (!uptodate(page3))
      goto do_read
                               readahead(FileA, 4K, 8K)
                               // Here create PCL-chain like below:
                               // [null, page1] [page2, null]
                               //   [PCL0:RA]-----[PCL1:RA]
...
  do read(page3)        // found [PCL1:RA] and add page3 into it,
                        // and then, change PCL1 from RA to R
...
                               // Now, PCL-chain is as below:
                               // [null, page1] [page2, page3]
                               //   [PCL0:RA]-----[PCL1:R]

                                 // try to decompress PCL-chain...
                                 z_erofs_decompress_queue
                                   err = 0;

                                   // failed with ENOMEM, so page 1
                                   // only for RA will not be uptodated.
                                   // it's okay.
                                   err = decompress([PCL0:RA], err)

                                   // However, ENOMEM propagated to next
                                   // PCL, even though PCL is not only
                                   // for RA but also for R. As a result,
                                   // it just failed with ENOMEM without
                                   // trying any decompression, so page2
                                   // and page3 will not be uptodated.
                ** BUG HERE ** --> err = decompress([PCL1:R], err)

                                   return err as ENOMEM

...
wait_lock(page3)
if (!uptodate(page3))
return EIO <-- Return an unexpected EIO!
...

Fixes 2349d2fa02db ("erofs: sunset unneeded NOFAILs")

Cc [email protected]

Reviewed-by Jaewook Kim <[email protected]>

Reviewed-by Sungjong Seo <[email protected]>

Signed-off-by Junbeom Yeom <[email protected]>

Reviewed-by Gao Xiang <[email protected]>

Signed-off-by Gao Xiang <[email protected]>

fs/erofs/zdata.c +4 -4

index 65da21504632..3d31f7840ca0 100644
@@ -1262,7 +1262,7 @@ static int z_erofs_parse_in_bvecs(struct z_erofs_backend *be, bool *overlapped)
 	return err;
 }
 
-static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
+static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, bool eio)
 {
 	struct erofs_sb_info *const sbi = EROFS_SB(be->sb);
 	struct z_erofs_pcluster *pcl = be->pcl;

@@ -1270,7 +1270,7 @@ static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
 	const struct z_erofs_decompressor *alg =
 				z_erofs_decomp[pcl->algorithmformat];
 	bool try_free = true;
-	int i, j, jtop, err2;
+	int i, j, jtop, err2, err = eio ? -EIO : 0;
 	struct page *page;
 	bool overlapped;
 	const char *reason;

@@ -1413,12 +1413,12 @@ static int z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io,
 		.pcl = io->head,
 	};
 	struct z_erofs_pcluster *next;
-	int err = io->eio ? -EIO : 0;
+	int err = 0;
 
 	for (; be.pcl != Z_EROFS_PCLUSTER_TAIL; be.pcl = next) {
 		DBG_BUGON(!be.pcl);
 		next = READ_ONCE(be.pcl->next);
-		err = z_erofs_decompress_pcluster(&be, err) ?: err;
+		err = z_erofs_decompress_pcluster(&be, io->eio) ?: err;
 	}
 	return err;
 }

Sasha Levin

maintainer

2025-12-29 18:54 UTC #1 lore raw

PATCH 1/2 erofs: improve decompression error reporting

From: Gao Xiang <[email protected]>

[ Upstream commit 831faabed8129246c9802af9ad9581a2c1e9eeb9 ]

Change the return type of decompress() from int to const char * to provide more informative error diagnostics:

- A NULL return indicates successful decompression;

- If IS_ERR(ptr) is true, the return value encodes a standard negative errno (e.g., -ENOMEM, -EOPNOTSUPP) identifying the specific error;

- Otherwise, a non-NULL return points to a human-readable error string, and the corresponding error code should be treated as -EFSCORRUPTED.

Signed-off-by Gao Xiang <[email protected]>

Stable-dep-of: 4012d7856219 ("erofs: fix unexpected EIO under memory pressure")

Signed-off-by Sasha Levin <[email protected]>

---
 fs/erofs/compress.h             |  4 ++--
 fs/erofs/decompressor.c         | 20 +++++++++-----------
 fs/erofs/decompressor_deflate.c | 10 ++++------
 fs/erofs/decompressor_lzma.c    | 10 ++++------
 fs/erofs/decompressor_zstd.c    | 12 ++++--------
 fs/erofs/zdata.c                | 21 +++++++++++++++++----
 6 files changed, 40 insertions(+), 37 deletions(-)

fs/erofs/compress.h +2 -2

index 510e922c5193..1ee4ad934c1f 100644
@@ -23,8 +23,8 @@ struct z_erofs_decompress_req {
 struct z_erofs_decompressor {
 	int (*config)(struct super_block *sb, struct erofs_super_block *dsb,
 		      void *data, int size);
-	int (*decompress)(struct z_erofs_decompress_req *rq,
-			  struct page **pagepool);
+	const char *(*decompress)(struct z_erofs_decompress_req *rq,
+				  struct page **pagepool);
 	int (*init)(void);
 	void (*exit)(void);
 	char *name;

fs/erofs/decompressor.c +9 ~~-11~~

index 354762c9723f..2d6b765dae65 100644
@@ -228,8 +228,6 @@ static int z_erofs_lz4_decompress_mem(struct z_erofs_decompress_req *rq, u8 *dst
 					  rq->inputsize, rq->outputsize);
 
 	if (ret != rq->outputsize) {
-		erofs_err(rq->sb, "failed to decompress %d in[%u, %u] out[%u]",
-			  ret, rq->inputsize, inputmargin, rq->outputsize);
 		if (ret >= 0)
 			memset(out + ret, 0, rq->outputsize - ret);
 		ret = -EFSCORRUPTED;

@@ -250,8 +248,8 @@ static int z_erofs_lz4_decompress_mem(struct z_erofs_decompress_req *rq, u8 *dst
 	return ret;
 }
 
-static int z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
-				  struct page **pagepool)
+static const char *z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
+					  struct page **pagepool)
 {
 	unsigned int dst_maptype;
 	void *dst;

@@ -266,14 +264,14 @@ static int z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
 		/* general decoding path which can be used for all cases */
 		ret = z_erofs_lz4_prepare_dstpages(rq, pagepool);
 		if (ret < 0)
-			return ret;
+			return ERR_PTR(ret);
 		if (ret > 0) {
 			dst = page_address(*rq->out);
 			dst_maptype = 1;
 		} else {
 			dst = erofs_vm_map_ram(rq->out, rq->outpages);
 			if (!dst)
-				return -ENOMEM;
+				return ERR_PTR(-ENOMEM);
 			dst_maptype = 2;
 		}
 	}

@@ -282,11 +280,11 @@ static int z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
 		kunmap_local(dst);
 	else if (dst_maptype == 2)
 		vm_unmap_ram(dst, rq->outpages);
-	return ret;
+	return ERR_PTR(ret);
 }
 
-static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
-				   struct page **pagepool)
+static const char *z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
+					   struct page **pagepool)
 {
 	const unsigned int nrpages_in = rq->inpages, nrpages_out = rq->outpages;
 	const unsigned int bs = rq->sb->s_blocksize;

@@ -294,7 +292,7 @@ static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
 	u8 *kin;
 
 	if (rq->outputsize > rq->inputsize)
-		return -EOPNOTSUPP;
+		return ERR_PTR(-EOPNOTSUPP);
 	if (rq->alg == Z_EROFS_COMPRESSION_INTERLACED) {
 		cur = bs - (rq->pageofs_out & (bs - 1));
 		pi = (rq->pageofs_in + rq->inputsize - cur) & ~PAGE_MASK;

@@ -334,7 +332,7 @@ static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
 		kunmap_local(kin);
 	}
 	DBG_BUGON(ni > nrpages_in);
-	return 0;
+	return NULL;
 }
 
 int z_erofs_stream_switch_bufs(struct z_erofs_stream_dctx *dctx, void **dst,

fs/erofs/decompressor_deflate.c +4 -6

index 6909b2d529c7..e9c4b740ef89 100644
@@ -157,8 +157,6 @@ static int __z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
 				break;
 			if (zerr == Z_STREAM_END && !rq->outputsize)
 				break;
-			erofs_err(sb, "failed to decompress %d in[%u] out[%u]",
-				  zerr, rq->inputsize, rq->outputsize);
 			err = -EFSCORRUPTED;
 			break;
 		}

@@ -178,8 +176,8 @@ static int __z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
 	return err;
 }
 
-static int z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
-				      struct page **pgpl)
+static const char *z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
+					      struct page **pgpl)
 {
 #ifdef CONFIG_EROFS_FS_ZIP_ACCEL
 	int err;

@@ -187,11 +185,11 @@ static int z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
 	if (!rq->partial_decoding) {
 		err = z_erofs_crypto_decompress(rq, pgpl);
 		if (err != -EOPNOTSUPP)
-			return err;
+			return ERR_PTR(err);
 
 	}
 #endif
-	return __z_erofs_deflate_decompress(rq, pgpl);
+	return ERR_PTR(__z_erofs_deflate_decompress(rq, pgpl));
 }
 
 const struct z_erofs_decompressor z_erofs_deflate_decomp = {

fs/erofs/decompressor_lzma.c +4 -6

index 832cffb83a66..7784ced90145 100644
@@ -146,8 +146,8 @@ static int z_erofs_load_lzma_config(struct super_block *sb,
 	return err;
 }
 
-static int z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
-				   struct page **pgpl)
+static const char *z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
+					   struct page **pgpl)
 {
 	struct super_block *sb = rq->sb;
 	struct z_erofs_stream_dctx dctx = { .rq = rq, .no = -1, .ni = 0 };

@@ -162,7 +162,7 @@ static int z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
 			min(rq->inputsize, sb->s_blocksize - rq->pageofs_in));
 	if (err) {
 		kunmap_local(dctx.kin);
-		return err;
+		return ERR_PTR(err);
 	}
 
 	/* 2. get an available lzma context */

@@ -207,8 +207,6 @@ static int z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
 		if (xz_err != XZ_OK) {
 			if (xz_err == XZ_STREAM_END && !rq->outputsize)
 				break;
-			erofs_err(sb, "failed to decompress %d in[%u] out[%u]",
-				  xz_err, rq->inputsize, rq->outputsize);
 			err = -EFSCORRUPTED;
 			break;
 		}

@@ -223,7 +221,7 @@ static int z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
 	z_erofs_lzma_head = strm;
 	spin_unlock(&z_erofs_lzma_lock);
 	wake_up(&z_erofs_lzma_wq);
-	return err;
+	return ERR_PTR(err);
 }
 
 const struct z_erofs_decompressor z_erofs_lzma_decomp = {

fs/erofs/decompressor_zstd.c +4 -8

index e38d93bb2104..50fadff89cbc 100644
@@ -135,8 +135,8 @@ static int z_erofs_load_zstd_config(struct super_block *sb,
 	return strm ? -ENOMEM : 0;
 }
 
-static int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
-				   struct page **pgpl)
+static const char *z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
+					   struct page **pgpl)
 {
 	struct super_block *sb = rq->sb;
 	struct z_erofs_stream_dctx dctx = { .rq = rq, .no = -1, .ni = 0 };

@@ -152,7 +152,7 @@ static int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
 			min(rq->inputsize, sb->s_blocksize - rq->pageofs_in));
 	if (err) {
 		kunmap_local(dctx.kin);
-		return err;
+		return ERR_PTR(err);
 	}
 
 	/* 2. get an available ZSTD context */

@@ -191,10 +191,6 @@ static int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
 		if (zstd_is_error(zerr) ||
 		    ((rq->outputsize + dctx.avail_out) && (!zerr || (zerr > 0 &&
 				!(rq->inputsize + in_buf.size - in_buf.pos))))) {
-			erofs_err(sb, "failed to decompress in[%u] out[%u]: %s",
-				  rq->inputsize, rq->outputsize,
-				  zstd_is_error(zerr) ? zstd_get_error_name(zerr) :
-					"unexpected end of stream");
 			err = -EFSCORRUPTED;
 			break;
 		}

@@ -210,7 +206,7 @@ static int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
 	z_erofs_zstd_head = strm;
 	spin_unlock(&z_erofs_zstd_lock);
 	wake_up(&z_erofs_zstd_wq);
-	return err;
+	return ERR_PTR(err);
 }
 
 const struct z_erofs_decompressor z_erofs_zstd_decomp = {

fs/erofs/zdata.c +17 -4

index bc80cfe482f7..461a929e0825 100644
@@ -1267,12 +1267,13 @@ static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
 	struct erofs_sb_info *const sbi = EROFS_SB(be->sb);
 	struct z_erofs_pcluster *pcl = be->pcl;
 	unsigned int pclusterpages = z_erofs_pclusterpages(pcl);
-	const struct z_erofs_decompressor *decomp =
+	const struct z_erofs_decompressor *alg =
 				z_erofs_decomp[pcl->algorithmformat];
+	bool try_free = true;
 	int i, j, jtop, err2;
 	struct page *page;
 	bool overlapped;
-	bool try_free = true;
+	const char *reason;
 
 	mutex_lock(&pcl->lock);
 	be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT;

@@ -1304,8 +1305,8 @@ static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
 	err2 = z_erofs_parse_in_bvecs(be, &overlapped);
 	if (err2)
 		err = err2;
-	if (!err)
-		err = decomp->decompress(&(struct z_erofs_decompress_req) {
+	if (!err) {
+		reason = alg->decompress(&(struct z_erofs_decompress_req) {
 					.sb = be->sb,
 					.in = be->compressed_pages,
 					.out = be->decompressed_pages,

@@ -1322,6 +1323,18 @@ static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
 					.gfp = pcl->besteffort ? GFP_KERNEL :
 						GFP_NOWAIT | __GFP_NORETRY
 				 }, be->pagepool);
+		if (IS_ERR(reason)) {
+			erofs_err(be->sb, "failed to decompress (%s) %ld @ pa %llu size %u => %u",
+				  alg->name, PTR_ERR(reason), pcl->pos,
+				  pcl->pclustersize, pcl->length);
+			err = PTR_ERR(reason);
+		} else if (unlikely(reason)) {
+			erofs_err(be->sb, "failed to decompress (%s) %s @ pa %llu size %u => %u",
+				  alg->name, reason, pcl->pos,
+				  pcl->pclustersize, pcl->length);
+			err = -EFSCORRUPTED;
+		}
+	}
 
 	/* must handle all compressed pages before actual file pages */
 	if (pcl->from_meta) {

Signature · 1 line

2.51.0

Sasha Levin

maintainer

2025-12-29 18:54 UTC #2 lore raw

PATCH 2/2 erofs: fix unexpected EIO under memory pressure

From: Junbeom Yeom <[email protected]>

[ Upstream commit 4012d78562193ef5eb613bad4b0c0fa187637cfe ]

However, in scenarios where readahead and read operations compete, read operation could return an unintended EIO because of an incorrect error propagation.

- Page size: 4K
- The original size of FileA: 16K
- Compress-ratio per PCL: 50% (Uncompressed 8K -> Compressed 4K) [page0, page1] [page2, page3]
[PCL0]---------[PCL1]

pread(FileA, buf, 4K, 12K)
do readahead(page3) // failed with ENOMEM
wait_lock(page3)

    if (!uptodate(page3))
      goto do_read
                               readahead(FileA, 4K, 8K)
                               // Here create PCL-chain like below:
                               // [null, page1] [page2, null]
                               //   [PCL0:RA]-----[PCL1:RA]
...
  do read(page3)        // found [PCL1:RA] and add page3 into it,
                        // and then, change PCL1 from RA to R
...
                               // Now, PCL-chain is as below:
                               // [null, page1] [page2, page3]
                               //   [PCL0:RA]-----[PCL1:R]

                                 // try to decompress PCL-chain...
                                 z_erofs_decompress_queue
                                   err = 0;

                                   // failed with ENOMEM, so page 1
                                   // only for RA will not be uptodated.
                                   // it's okay.
                                   err = decompress([PCL0:RA], err)

                                   // However, ENOMEM propagated to next
                                   // PCL, even though PCL is not only
                                   // for RA but also for R. As a result,
                                   // it just failed with ENOMEM without
                                   // trying any decompression, so page2
                                   // and page3 will not be uptodated.
                ** BUG HERE ** --> err = decompress([PCL1:R], err)

                                   return err as ENOMEM

...
wait_lock(page3)
if (!uptodate(page3))
return EIO <-- Return an unexpected EIO!
...

Fixes 2349d2fa02db ("erofs: sunset unneeded NOFAILs")

Cc [email protected]

Reviewed-by Jaewook Kim <[email protected]>

Reviewed-by Sungjong Seo <[email protected]>

Signed-off-by Junbeom Yeom <[email protected]>

Reviewed-by Gao Xiang <[email protected]>

Signed-off-by Gao Xiang <[email protected]>

Signed-off-by Sasha Levin <[email protected]>

---
 fs/erofs/zdata.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

fs/erofs/zdata.c +4 -4

index 461a929e0825..c23e0278c373 100644
@@ -1262,7 +1262,7 @@ static int z_erofs_parse_in_bvecs(struct z_erofs_backend *be, bool *overlapped)
 	return err;
 }
 
-static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
+static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, bool eio)
 {
 	struct erofs_sb_info *const sbi = EROFS_SB(be->sb);
 	struct z_erofs_pcluster *pcl = be->pcl;

@@ -1270,7 +1270,7 @@ static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
 	const struct z_erofs_decompressor *alg =
 				z_erofs_decomp[pcl->algorithmformat];
 	bool try_free = true;
-	int i, j, jtop, err2;
+	int i, j, jtop, err2, err = eio ? -EIO : 0;
 	struct page *page;
 	bool overlapped;
 	const char *reason;

@@ -1413,12 +1413,12 @@ static int z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io,
 		.pcl = io->head,
 	};
 	struct z_erofs_pcluster *next;
-	int err = io->eio ? -EIO : 0;
+	int err = 0;
 
 	for (; be.pcl != Z_EROFS_PCLUSTER_TAIL; be.pcl = next) {
 		DBG_BUGON(!be.pcl);
 		next = READ_ONCE(be.pcl->next);
-		err = z_erofs_decompress_pcluster(&be, err) ?: err;
+		err = z_erofs_decompress_pcluster(&be, io->eio) ?: err;
 	}
 	return err;
 }

Signature · 1 line

2.51.0

Gao Xiang

2025-12-30 02:31 UTC #3 lore raw

PATCH v2 1/1 erofs: fix unexpected EIO under memory pressure

From: Junbeom Yeom <[email protected]>

However, in scenarios where readahead and read operations compete, read operation could return an unintended EIO because of an incorrect error propagation.

- Page size: 4K
- The original size of FileA: 16K
- Compress-ratio per PCL: 50% (Uncompressed 8K -> Compressed 4K) [page0, page1] [page2, page3]
[PCL0]---------[PCL1]

pread(FileA, buf, 4K, 12K)
do readahead(page3) // failed with ENOMEM
wait_lock(page3)

    if (!uptodate(page3))
      goto do_read
                               readahead(FileA, 4K, 8K)
                               // Here create PCL-chain like below:
                               // [null, page1] [page2, null]
                               //   [PCL0:RA]-----[PCL1:RA]
...
  do read(page3)        // found [PCL1:RA] and add page3 into it,
                        // and then, change PCL1 from RA to R
...
                               // Now, PCL-chain is as below:
                               // [null, page1] [page2, page3]
                               //   [PCL0:RA]-----[PCL1:R]

                                 // try to decompress PCL-chain...
                                 z_erofs_decompress_queue
                                   err = 0;

                                   // failed with ENOMEM, so page 1
                                   // only for RA will not be uptodated.
                                   // it's okay.
                                   err = decompress([PCL0:RA], err)

                                   // However, ENOMEM propagated to next
                                   // PCL, even though PCL is not only
                                   // for RA but also for R. As a result,
                                   // it just failed with ENOMEM without
                                   // trying any decompression, so page2
                                   // and page3 will not be uptodated.
                ** BUG HERE ** --> err = decompress([PCL1:R], err)

                                   return err as ENOMEM

...
wait_lock(page3)
if (!uptodate(page3))
return EIO <-- Return an unexpected EIO!
...

Fixes 2349d2fa02db ("erofs: sunset unneeded NOFAILs")

Cc [email protected]

Reviewed-by Jaewook Kim <[email protected]>

Reviewed-by Sungjong Seo <[email protected]>

Signed-off-by Junbeom Yeom <[email protected]>

Reviewed-by Gao Xiang <[email protected]>

Signed-off-by Gao Xiang <[email protected]>

---
Hi Greg and Sasha,

Let's just merge this directly.
No need to backport commit 831faabed812 ("erofs: improve decompression error reporting") for now.

Thanks,
Gao Xiang

 fs/erofs/zdata.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

fs/erofs/zdata.c +5 -5

index bc80cfe482f7..683703aee5ef 100644
@@ -1262,17 +1262,17 @@ static int z_erofs_parse_in_bvecs(struct z_erofs_backend *be, bool *overlapped)
 	return err;
 }
 
-static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, int err)
+static int z_erofs_decompress_pcluster(struct z_erofs_backend *be, bool eio)
 {
 	struct erofs_sb_info *const sbi = EROFS_SB(be->sb);
 	struct z_erofs_pcluster *pcl = be->pcl;
 	unsigned int pclusterpages = z_erofs_pclusterpages(pcl);
 	const struct z_erofs_decompressor *decomp =
 				z_erofs_decomp[pcl->algorithmformat];
-	int i, j, jtop, err2;
+	bool try_free = true;
+	int i, j, jtop, err2, err = eio ? -EIO : 0;
 	struct page *page;
 	bool overlapped;
-	bool try_free = true;
 
 	mutex_lock(&pcl->lock);
 	be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT;

@@ -1400,12 +1400,12 @@ static int z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io,
 		.pcl = io->head,
 	};
 	struct z_erofs_pcluster *next;
-	int err = io->eio ? -EIO : 0;
+	int err = 0;
 
 	for (; be.pcl != Z_EROFS_PCLUSTER_TAIL; be.pcl = next) {
 		DBG_BUGON(!be.pcl);
 		next = READ_ONCE(be.pcl->next);
-		err = z_erofs_decompress_pcluster(&be, err) ?: err;
+		err = z_erofs_decompress_pcluster(&be, io->eio) ?: err;
 	}
 	return err;
 }

Signature · 1 line

2.43.5

Greg KH

author maintainer

2026-01-05 13:56 UTC #4 lore raw

Re: [PATCH 6.18.y v2] erofs: fix unexpected EIO under memory pressure

Quoted 80 lines · Gao Xiang

From: Junbeom Yeom <[email protected]>
erofs readahead could fail with ENOMEM under the memory pressure because it tries to alloc_page with GFP_NOWAIT | GFP_NORETRY, while GFP_KERNEL for a regular read. And if readahead fails (with non-uptodate folios), the original request will then fall back to synchronous read, and .read_folio() should return appropriate errnos.
However, in scenarios where readahead and read operations compete, read operation could return an unintended EIO because of an incorrect error propagation.
To resolve this, this patch modifies the behavior so that, when the PCL is for read(which means pcl.besteffort is true), it attempts actual decompression instead of propagating the privios error except initial EIO.
- Page size: 4K
- The original size of FileA: 16K
- Compress-ratio per PCL: 50% (Uncompressed 8K -> Compressed 4K) [page0, page1] [page2, page3]
[PCL0]---------[PCL1]
- functions declaration:
. pread(fd, buf, count, offset)
. readahead(fd, offset, count)
- Thread A tries to read the last 4K
- Thread B tries to do readahead 8K from 4K
- RA, besteffort == false
- R, besteffort == true
<process A> <process B>
pread(FileA, buf, 4K, 12K)
do readahead(page3) // failed with ENOMEM
wait_lock(page3)
    if (!uptodate(page3))
      goto do_read
                               readahead(FileA, 4K, 8K)
                               // Here create PCL-chain like below:
                               // [null, page1] [page2, null]
                               //   [PCL0:RA]-----[PCL1:RA]
...
  do read(page3)        // found [PCL1:RA] and add page3 into it,
                        // and then, change PCL1 from RA to R
...
                               // Now, PCL-chain is as below:
                               // [null, page1] [page2, page3]
                               //   [PCL0:RA]-----[PCL1:R]

                                 // try to decompress PCL-chain...
                                 z_erofs_decompress_queue
                                   err = 0;

                                   // failed with ENOMEM, so page 1
                                   // only for RA will not be uptodated.
                                   // it's okay.
                                   err = decompress([PCL0:RA], err)

                                   // However, ENOMEM propagated to next
                                   // PCL, even though PCL is not only
                                   // for RA but also for R. As a result,
                                   // it just failed with ENOMEM without
                                   // trying any decompression, so page2
                                   // and page3 will not be uptodated.
                ** BUG HERE ** --> err = decompress([PCL1:R], err)

                                   return err as ENOMEM
...
wait_lock(page3)
if (!uptodate(page3))
return EIO <-- Return an unexpected EIO!
...
Fixes 2349d2fa02db ("erofs: sunset unneeded NOFAILs")
Cc [email protected]
Reviewed-by Jaewook Kim <[email protected]>
Reviewed-by Sungjong Seo <[email protected]>
Signed-off-by Junbeom Yeom <[email protected]>
Reviewed-by Gao Xiang <[email protected]>
Signed-off-by Gao Xiang <[email protected]>

Quoted 6 lines

---
Hi Greg and Sasha,
Let's just merge this directly.
No need to backport commit 831faabed812 ("erofs: improve decompression error reporting") for now.

Now taken, thanks!

greg k-h