commit d4625a19fe64f664119a541b317fb83de01bb273
Author: Florian Weimer <fweimer@redhat.com>
Date:   Tue Nov 12 12:25:49 2019 +0100
 
    login: Use pread64 in utmp implementation
    
    This reduces the possible error scenarios considerably because
    no longer can file seek fail, leaving the file descriptor in an
    inconsistent state and out of sync with the cache.
    
    As a result, it is possible to avoid setting file_offset to -1
    to make an error persistent.  Instead, subsequent calls will retry
    the operation and report any errors returned by the kernel.
    
    This change also avoids reading the file from the start if pututline
    is called multiple times, to work around lock acquisition failures
    due to timeouts.
    
    Change-Id: If21ea0c162c38830a89331ea93cddec14c0974de
 
diff --git a/login/utmp_file.c b/login/utmp_file.c
index e653d14967c4fb7a..c828a28ac54c150e 100644
--- a/login/utmp_file.c
+++ b/login/utmp_file.c
@@ -162,12 +162,35 @@ maybe_setutent (void)
   return file_fd >= 0 || __libc_setutent ();
 }
 
+/* Reads the entry at file_offset, storing it in last_entry and
+   updating file_offset on success.  Returns -1 for a read error, 0
+   for EOF, and 1 for a successful read.  last_entry and file_offset
+   are only updated on a successful and complete read.  */
+static ssize_t
+read_last_entry (void)
+{
+  struct utmp buffer;
+  ssize_t nbytes = __pread64_nocancel (file_fd, &buffer, sizeof (buffer),
+				       file_offset);
+  if (nbytes < 0)
+    return -1;
+  else if (nbytes != sizeof (buffer))
+    /* Assume EOF.  */
+    return 0;
+  else
+    {
+      last_entry = buffer;
+      file_offset += sizeof (buffer);
+      return 1;
+    }
+}
+
 int
 __libc_getutent_r (struct utmp *buffer, struct utmp **result)
 {
-  ssize_t nbytes;
+  int saved_errno = errno;
 
-  if (!maybe_setutent () || file_offset == -1l)
+  if (!maybe_setutent ())
     {
       /* Not available.  */
       *result = NULL;
@@ -175,25 +198,22 @@ __libc_getutent_r (struct utmp *buffer, struct utmp **result)
     }
 
   if (try_file_lock (file_fd, F_RDLCK))
-    nbytes = 0;
-  else
-    {
-      /* Read the next entry.  */
-      nbytes = __read_nocancel (file_fd, &last_entry, sizeof (struct utmp));
-      file_unlock (file_fd);
-    }
+    return -1;
 
-  if (nbytes != sizeof (struct utmp))
+  ssize_t nbytes = read_last_entry ();
+  file_unlock (file_fd);
+
+  if (nbytes <= 0)		/* Read error or EOF.  */
     {
-      if (nbytes != 0)
-	file_offset = -1l;
+      if (nbytes == 0)
+	/* errno should be unchanged to indicate success.  A premature
+	   EOF is treated like an EOF (missing complete record at the
+	   end).  */
+	__set_errno (saved_errno);
       *result = NULL;
       return -1;
     }
 
-  /* Update position pointer.  */
-  file_offset += sizeof (struct utmp);
-
   memcpy (buffer, &last_entry, sizeof (struct utmp));
   *result = buffer;
 
@@ -209,15 +229,15 @@ internal_getut_nolock (const struct utmp *id)
 {
   while (1)
     {
-      /* Read the next entry.  */
-      if (__read_nocancel (file_fd, &last_entry, sizeof (struct utmp))
-	  != sizeof (struct utmp))
+      ssize_t nbytes = read_last_entry ();
+      if (nbytes < 0)
+	return -1;
+      if (nbytes == 0)
 	{
+	  /* End of file reached.  */
 	  __set_errno (ESRCH);
-	  file_offset = -1l;
 	  return -1;
 	}
-      file_offset += sizeof (struct utmp);
 
       if (matches_last_entry (id))
 	break;
@@ -249,7 +269,7 @@ int
 __libc_getutid_r (const struct utmp *id, struct utmp *buffer,
 		  struct utmp **result)
 {
-  if (!maybe_setutent () || file_offset == -1l)
+  if (!maybe_setutent ())
     {
       *result = NULL;
       return -1;
@@ -276,7 +296,7 @@ int
 __libc_getutline_r (const struct utmp *line, struct utmp *buffer,
 		    struct utmp **result)
 {
-  if (!maybe_setutent () || file_offset == -1l)
+  if (!maybe_setutent ())
     {
       *result = NULL;
       return -1;
@@ -290,16 +310,21 @@ __libc_getutline_r (const struct utmp *line, struct utmp *buffer,
 
   while (1)
     {
-      /* Read the next entry.  */
-      if (__read_nocancel (file_fd, &last_entry, sizeof (struct utmp))
-	  != sizeof (struct utmp))
+      ssize_t nbytes = read_last_entry ();
+      if (nbytes < 0)
 	{
+	  file_unlock (file_fd);
+	  *result = NULL;
+	  return -1;
+	}
+      if (nbytes == 0)
+	{
+	  /* End of file reached.  */
+	  file_unlock (file_fd);
 	  __set_errno (ESRCH);
-	  file_offset = -1l;
 	  *result = NULL;
-	  goto unlock_return;
+	  return -1;
 	}
-      file_offset += sizeof (struct utmp);
 
       /* Stop if we found a user or login entry.  */
       if ((last_entry.ut_type == USER_PROCESS
@@ -309,20 +334,18 @@ __libc_getutline_r (const struct utmp *line, struct utmp *buffer,
 	break;
     }
 
+  file_unlock (file_fd);
   memcpy (buffer, &last_entry, sizeof (struct utmp));
   *result = buffer;
 
-unlock_return:
-  file_unlock (file_fd);
-
-  return ((*result == NULL) ? -1 : 0);
+  return 0;
 }
 
 
 struct utmp *
 __libc_pututline (const struct utmp *data)
 {
-  if (!maybe_setutent () || file_offset == -1l)
+  if (!maybe_setutent ())
     return NULL;
 
   struct utmp *pbuf;
@@ -337,8 +360,7 @@ __libc_pututline (const struct utmp *data)
       if (new_fd == -1)
 	return NULL;
 
-      if (__lseek64 (new_fd, __lseek64 (file_fd, 0, SEEK_CUR), SEEK_SET) == -1
-	  || __dup2 (new_fd, file_fd) < 0)
+      if (__dup2 (new_fd, file_fd) < 0)
 	{
 	  __close_nocancel_nostatus (new_fd);
 	  return NULL;
@@ -355,69 +377,70 @@ __libc_pututline (const struct utmp *data)
   bool found = false;
   if (matches_last_entry (data))
     {
-      if (__lseek64 (file_fd, file_offset, SEEK_SET) < 0)
+      /* Read back the entry under the write lock.  */
+      file_offset -= sizeof (last_entry);
+      ssize_t nbytes = read_last_entry ();
+      if (nbytes < 0)
 	{
 	  file_unlock (file_fd);
 	  return NULL;
 	}
-      if (__read_nocancel (file_fd, &last_entry, sizeof (last_entry))
-	  != sizeof (last_entry))
-	{
-	  if (__lseek64 (file_fd, file_offset, SEEK_SET) < 0)
-	    {
-	      file_unlock (file_fd);
-	      return NULL;
-	    }
-	  found = false;
-	}
+
+      if (nbytes == 0)
+	/* End of file reached.  */
+	found = false;
       else
 	found = matches_last_entry (data);
     }
 
   if (!found)
+    /* Search forward for the entry.  */
     found = internal_getut_nolock (data) >= 0;
 
+  off64_t write_offset;
   if (!found)
     {
       /* We append the next entry.  */
-      file_offset = __lseek64 (file_fd, 0, SEEK_END);
-      if (file_offset % sizeof (struct utmp) != 0)
-	{
-	  file_offset -= file_offset % sizeof (struct utmp);
-	  __ftruncate64 (file_fd, file_offset);
-
-	  if (__lseek64 (file_fd, 0, SEEK_END) < 0)
-	    {
-	      pbuf = NULL;
-	      goto unlock_return;
-	    }
-	}
+      write_offset = __lseek64 (file_fd, 0, SEEK_END);
+
+      /* Round down to the next multiple of the entry size.  This
+	 ensures any partially-written record is overwritten by the
+	 new record.  */
+      write_offset = (write_offset / sizeof (struct utmp)
+		      * sizeof (struct utmp));
     }
   else
+    /* Overwrite last_entry.  */
+    write_offset = file_offset - sizeof (struct utmp);
+
+  /* Write the new data.  */
+  ssize_t nbytes;
+  if (__lseek64 (file_fd, write_offset, SEEK_SET) < 0
+      || (nbytes = __write_nocancel (file_fd, data, sizeof (struct utmp))) < 0)
     {
-      /* We replace the just read entry.  */
-      file_offset -= sizeof (struct utmp);
-      __lseek64 (file_fd, file_offset, SEEK_SET);
+      /* There is no need to recover the file position because all
+	 reads use pread64, and any future write is preceded by
+	 another seek.  */
+      file_unlock (file_fd);
+      return NULL;
     }
 
-  /* Write the new data.  */
-  if (__write_nocancel (file_fd, data, sizeof (struct utmp))
-      != sizeof (struct utmp))
+  if (nbytes != sizeof (struct utmp))
     {
       /* If we appended a new record this is only partially written.
 	 Remove it.  */
       if (!found)
-	(void) __ftruncate64 (file_fd, file_offset);
-      pbuf = NULL;
-    }
-  else
-    {
-      file_offset += sizeof (struct utmp);
-      pbuf = (struct utmp *) data;
+	(void) __ftruncate64 (file_fd, write_offset);
+      file_unlock (file_fd);
+      /* Assume that the write failure was due to missing disk
+	 space.  */
+      __set_errno (ENOSPC);
+      return NULL;
     }
 
- unlock_return:
   file_unlock (file_fd);
+  file_offset = write_offset + sizeof (struct utmp);
+  pbuf = (struct utmp *) data;
 
   return pbuf;
 }