Linux kernel -stable discussions
 help / color / mirror / Atom feed
From: Ben Hutchings <ben@decadent.org.uk>
To: linux-kernel@vger.kernel.org, stable@vger.kernel.org
Cc: akpm@linux-foundation.org, "Liu Song" <liu.song11@zte.com.cn>,
	"David Woodhouse" <David.Woodhouse@intel.com>
Subject: [PATCH 3.2 39/62] Fix directory hardlinks from deleted directories
Date: Tue, 29 Mar 2016 20:18:22 +0100	[thread overview]
Message-ID: <lsq.1459279102.705866553@decadent.org.uk> (raw)
In-Reply-To: <lsq.1459279101.951687763@decadent.org.uk>

3.2.79-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: David Woodhouse <David.Woodhouse@intel.com>

commit be629c62a603e5935f8177fd8a19e014100a259e upstream.

When a directory is deleted, we don't take too much care about killing off
all the dirents that belong to it — on the basis that on remount, the scan
will conclude that the directory is dead anyway.

This doesn't work though, when the deleted directory contained a child
directory which was moved *out*. In the early stages of the fs build
we can then end up with an apparent hard link, with the child directory
appearing both in its true location, and as a child of the original
directory which are this stage of the mount process we don't *yet* know
is defunct.

To resolve this, take out the early special-casing of the "directories
shall not have hard links" rule in jffs2_build_inode_pass1(), and let the
normal nlink processing happen for directories as well as other inodes.

Then later in the build process we can set ic->pino_nlink to the parent
inode#, as is required for directories during normal operaton, instead
of the nlink. And complain only *then* about hard links which are still
in evidence even after killing off all the unreachable paths.

Reported-by: Liu Song <liu.song11@zte.com.cn>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
 fs/jffs2/build.c    | 75 ++++++++++++++++++++++++++++++++++++++++-------------
 fs/jffs2/nodelist.h |  6 ++++-
 2 files changed, 62 insertions(+), 19 deletions(-)

--- a/fs/jffs2/build.c
+++ b/fs/jffs2/build.c
@@ -47,7 +47,8 @@ next_inode(int *i, struct jffs2_inode_ca
 
 
 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
-				    struct jffs2_inode_cache *ic)
+				    struct jffs2_inode_cache *ic,
+				    int *dir_hardlinks)
 {
 	struct jffs2_full_dirent *fd;
 
@@ -66,19 +67,21 @@ static void jffs2_build_inode_pass1(stru
 			dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
 				  fd->name, fd->ino, ic->ino);
 			jffs2_mark_node_obsolete(c, fd->raw);
+			/* Clear the ic/raw union so it doesn't cause problems later. */
+			fd->ic = NULL;
 			continue;
 		}
 
+		/* From this point, fd->raw is no longer used so we can set fd->ic */
+		fd->ic = child_ic;
+		child_ic->pino_nlink++;
+		/* If we appear (at this stage) to have hard-linked directories,
+		 * set a flag to trigger a scan later */
 		if (fd->type == DT_DIR) {
-			if (child_ic->pino_nlink) {
-				JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u appears to be a hard link\n",
-					    fd->name, fd->ino, ic->ino);
-				/* TODO: What do we do about it? */
-			} else {
-				child_ic->pino_nlink = ic->ino;
-			}
-		} else
-			child_ic->pino_nlink++;
+			child_ic->flags |= INO_FLAGS_IS_DIR;
+			if (child_ic->pino_nlink > 1)
+				*dir_hardlinks = 1;
+		}
 
 		dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
 		/* Can't free scan_dents so far. We might need them in pass 2 */
@@ -92,8 +95,7 @@ static void jffs2_build_inode_pass1(stru
 */
 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
 {
-	int ret;
-	int i;
+	int ret, i, dir_hardlinks = 0;
 	struct jffs2_inode_cache *ic;
 	struct jffs2_full_dirent *fd;
 	struct jffs2_full_dirent *dead_fds = NULL;
@@ -117,7 +119,7 @@ static int jffs2_build_filesystem(struct
 	/* Now scan the directory tree, increasing nlink according to every dirent found. */
 	for_each_inode(i, c, ic) {
 		if (ic->scan_dents) {
-			jffs2_build_inode_pass1(c, ic);
+			jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
 			cond_resched();
 		}
 	}
@@ -153,6 +155,20 @@ static int jffs2_build_filesystem(struct
 	}
 
 	dbg_fsbuild("pass 2a complete\n");
+
+	if (dir_hardlinks) {
+		/* If we detected directory hardlinks earlier, *hopefully*
+		 * they are gone now because some of the links were from
+		 * dead directories which still had some old dirents lying
+		 * around and not yet garbage-collected, but which have
+		 * been discarded above. So clear the pino_nlink field
+		 * in each directory, so that the final scan below can
+		 * print appropriate warnings. */
+		for_each_inode(i, c, ic) {
+			if (ic->flags & INO_FLAGS_IS_DIR)
+				ic->pino_nlink = 0;
+		}
+	}
 	dbg_fsbuild("freeing temporary data structures\n");
 
 	/* Finally, we can scan again and free the dirent structs */
@@ -160,6 +176,33 @@ static int jffs2_build_filesystem(struct
 		while(ic->scan_dents) {
 			fd = ic->scan_dents;
 			ic->scan_dents = fd->next;
+			/* We do use the pino_nlink field to count nlink of
+			 * directories during fs build, so set it to the
+			 * parent ino# now. Now that there's hopefully only
+			 * one. */
+			if (fd->type == DT_DIR) {
+				if (!fd->ic) {
+					/* We'll have complained about it and marked the coresponding
+					   raw node obsolete already. Just skip it. */
+					continue;
+				}
+
+				/* We *have* to have set this in jffs2_build_inode_pass1() */
+				BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
+
+				/* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
+				 * is set. Otherwise, we know this should never trigger anyway, so
+				 * we don't do the check. And ic->pino_nlink still contains the nlink
+				 * value (which is 1). */
+				if (dir_hardlinks && fd->ic->pino_nlink) {
+					JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
+						    fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
+					/* Should we unlink it from its previous parent? */
+				}
+
+				/* For directories, ic->pino_nlink holds that parent inode # */
+				fd->ic->pino_nlink = ic->ino;
+			}
 			jffs2_free_full_dirent(fd);
 		}
 		ic->scan_dents = NULL;
@@ -238,11 +281,7 @@ static void jffs2_build_remove_unlinked_
 
 			/* Reduce nlink of the child. If it's now zero, stick it on the
 			   dead_fds list to be cleaned up later. Else just free the fd */
-
-			if (fd->type == DT_DIR)
-				child_ic->pino_nlink = 0;
-			else
-				child_ic->pino_nlink--;
+			child_ic->pino_nlink--;
 
 			if (!child_ic->pino_nlink) {
 				dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
--- a/fs/jffs2/nodelist.h
+++ b/fs/jffs2/nodelist.h
@@ -194,6 +194,7 @@ struct jffs2_inode_cache {
 #define INO_STATE_CLEARING	6	/* In clear_inode() */
 
 #define INO_FLAGS_XATTR_CHECKED	0x01	/* has no duplicate xattr_ref */
+#define INO_FLAGS_IS_DIR	0x02	/* is a directory */
 
 #define RAWNODE_CLASS_INODE_CACHE	0
 #define RAWNODE_CLASS_XATTR_DATUM	1
@@ -249,7 +250,10 @@ struct jffs2_readinode_info
 
 struct jffs2_full_dirent
 {
-	struct jffs2_raw_node_ref *raw;
+	union {
+		struct jffs2_raw_node_ref *raw;
+		struct jffs2_inode_cache *ic; /* Just during part of build */
+	};
 	struct jffs2_full_dirent *next;
 	uint32_t version;
 	uint32_t ino; /* == zero for unlink */

  parent reply	other threads:[~2016-03-29 19:18 UTC|newest]

Thread overview: 66+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-03-29 19:18 [PATCH 3.2 00/62] 3.2.79-rc1 review Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 19/62] ALSA: seq: Fix double port list deletion Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 45/62] ALSA: timer: Fix broken compat timer user status ioctl Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 36/62] KVM: async_pf: do not warn on page allocation failures Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 57/62] ubi: Fix out of bounds write in volume update code Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 16/62] ALSA: seq: Fix leak of pool buffer at concurrent writes Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 04/62] wext: fix message delay/ordering Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 61/62] include/linux/poison.h: fix LIST_POISON{1,2} offset Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 25/62] usb: dwc3: Fix assignment of EP transfer resources Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 34/62] mac80211: minstrel_ht: set default tx aggregation timeout to 0 Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 23/62] x86/uaccess/64: Make the __copy_user_nocache() assembly code more readable Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 09/62] cifs: fix erroneous return value Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 22/62] x86, extable: Remove open-coded exception table entries in arch/x86/lib/copy_user_nocache_64.S Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 26/62] net/mlx4_en: Count HW buffer overrun only once Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 31/62] ext4: fix bh->b_state corruption Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 08/62] drm/i915: fix error path in intel_setup_gmbus() Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 41/62] ipr: Fix out-of-bounds null overwrite Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 29/62] USB: option: add "4G LTE usb-modem U901" Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 06/62] mac80211: fix use of uninitialised values in RX aggregation Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 47/62] ALSA: hdsp: Fix wrong boolean ctl value accesses Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 20/62] af_unix: Don't set err in unix_stream_read_generic unless there was an error Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 38/62] jffs2: Fix page lock / f->sem deadlock Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 27/62] USB: option: add support for SIM7100E Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 05/62] cfg80211/wext: fix message ordering Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 03/62] iommu/vt-d: Fix 64-bit accesses to 32-bit DMAR_GSTS_REG Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 11/62] s390/dasd: fix refcount for PAV reassignment Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 30/62] Adding Intel Lewisburg device IDs for SATA Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 35/62] sunrpc/cache: fix off-by-one in qword_get() Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 32/62] kernel/resource.c: fix muxed resource handling in __request_region() Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 10/62] s390/dasd: prevent incorrect length error under z/VM after PAV changes Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 02/62] crypto: {blk,giv}cipher: Set has_setkey Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 40/62] libata: Align ata_device's id on a cacheline Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 55/62] mld, igmp: Fix reserved tailroom calculation Ben Hutchings
2016-03-29 19:18 ` Ben Hutchings [this message]
2016-03-29 19:18 ` [PATCH 3.2 62/62] HID: usbhid: fix recursive deadlock Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 24/62] x86/uaccess/64: Handle the caching of 4-byte nocache copies properly in __copy_user_nocache() Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 42/62] ipr: Fix regression when loading firmware Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 28/62] USB: cp210x: add IDs for GE B650V3 and B850V3 boards Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 15/62] xen/pcifront: Fix mysterious crashes when NUMA locality information was extracted Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 13/62] xen/pciback: Check PF instead of VF for PCI_COMMAND_MEMORY Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 01/62] Revert "crypto: algif_skcipher - Do not dereference ctx without socket lock" Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 52/62] ALSA: seq: oss: Don't drain at closing a client Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 07/62] nfs: fix nfs_size_to_loff_t Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 49/62] ASoC: wm8958: Fix enum ctl accesses in a wrong type Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 12/62] libata: fix HDIO_GET_32BIT ioctl Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 53/62] PM / sleep / x86: Fix crash on graph trace through x86 suspend Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 46/62] ALSA: hdspm: Fix wrong boolean ctl value accesses Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 56/62] MIPS: traps: Fix SIGFPE information leak from `do_ov' and `do_trap_or_bp' Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 37/62] Revert "jffs2: Fix lock acquisition order bug in jffs2_write_begin" Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 58/62] Revert "drm/radeon: call hpd_irq_event on resume" Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 59/62] s390/mm: four page table levels vs. fork Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 54/62] IB/core: Use GRH when the path hop-limit > 0 Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 18/62] tracing: Fix freak link error caused by branch tracer Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 51/62] USB: serial: option: add support for Quectel UC20 Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 50/62] ASoC: wm8994: Fix enum ctl accesses in a wrong type Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 21/62] af_unix: Guard against other == sk in unix_dgram_sendmsg Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 33/62] can: ems_usb: Fix possible tx overflow Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 17/62] tracepoints: Do not trace when cpu is offline Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 48/62] USB: cp210x: Add ID for Parrot NMEA GPS Flight Recorder Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 43/62] mm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 44/62] hpfs: don't truncate the file when delete fails Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 14/62] xen/pciback: Save the number of MSI-X entries to be copied later Ben Hutchings
2016-03-29 19:18 ` [PATCH 3.2 60/62] Input: aiptek - fix crash on detecting device without endpoints Ben Hutchings
2016-03-29 20:19 ` [PATCH 3.2 00/62] 3.2.79-rc1 review Ben Hutchings
2016-03-29 20:26 ` Guenter Roeck
2016-03-29 21:03   ` Ben Hutchings

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=lsq.1459279102.705866553@decadent.org.uk \
    --to=ben@decadent.org.uk \
    --cc=David.Woodhouse@intel.com \
    --cc=akpm@linux-foundation.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=liu.song11@zte.com.cn \
    --cc=stable@vger.kernel.org \
    /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