OpenZFS 9330 - stack overflow when creating a deeply nested dataset

Datasets that are deeply nested (~100 levels) are impractical. We just
put a limit of 50 levels to newly created datasets. Existing datasets
should work without a problem.

The problem can be seen by attempting to create a dataset using the -p
option with many levels:

    panic[cpu0]/thread=ffffff01cd282c20: BAD TRAP: type=8 (#df Double fault) rp=ffffffff

    fffffffffbc3aa60 unix:die+100 ()
    fffffffffbc3ab70 unix:trap+157d ()
    ffffff00083d7020 unix:_patch_xrstorq_rbx+196 ()
    ffffff00083d7050 zfs:dbuf_rele+2e ()
    ...
    ffffff00083d7080 zfs:dsl_dir_close+32 ()
    ffffff00083d70b0 zfs:dsl_dir_evict+30 ()
    ffffff00083d70d0 zfs:dbuf_evict_user+4a ()
    ffffff00083d7100 zfs:dbuf_rele_and_unlock+87 ()
    ffffff00083d7130 zfs:dbuf_rele+2e ()
    ... The block above repeats once per directory in the ...
    ... create -p command, working towards the root ...
    ffffff00083db9f0 zfs:dsl_dataset_drop_ref+19 ()
    ffffff00083dba20 zfs:dsl_dataset_rele+42 ()
    ffffff00083dba70 zfs:dmu_objset_prefetch+e4 ()
    ffffff00083dbaa0 zfs:findfunc+23 ()
    ffffff00083dbb80 zfs:dmu_objset_find_spa+38c ()
    ffffff00083dbbc0 zfs:dmu_objset_find+40 ()
    ffffff00083dbc20 zfs:zfs_ioc_snapshot_list_next+4b ()
    ffffff00083dbcc0 zfs:zfsdev_ioctl+347 ()
    ffffff00083dbd00 genunix:cdev_ioctl+45 ()
    ffffff00083dbd40 specfs:spec_ioctl+5a ()
    ffffff00083dbdc0 genunix:fop_ioctl+7b ()
    ffffff00083dbec0 genunix:ioctl+18e ()
    ffffff00083dbf10 unix:brand_sys_sysenter+1c9 ()

Porting notes:
* Added zfs_max_dataset_nesting module option with documentation.
* Updated zfs_rename_014_neg.ksh for Linux.
* Increase the zfs.sh stack warning to 15K.  Enough time has passed
  that 16K can be reasonably assumed to be the default value.  It
  was increased in the 3.15 kernel released in June of 2014.

Authored by: Serapheim Dimitropoulos <[email protected]>
Reviewed by: John Kennedy <[email protected]>
Reviewed by: Matt Ahrens <[email protected]>
Ported-by: Brian Behlendorf <[email protected]>
Approved by: Garrett D'Amore <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/9330
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/757a75a
Closes #7681

1 files changed, 67 insertions, 12 deletions

diff --git a/module/zcommon/zfs_namecheck.c b/module/zcommon/zfs_namecheck.c
index aefde9087..58b23b0e0 100644
--- a/module/zcommon/zfs_namecheck.c
+++ b/module/zcommon/zfs_namecheck.c
@@ -23,7 +23,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
  */
 
 /*
@@ -34,8 +34,6 @@
  * name is invalid.  In the kernel, we only care whether it's valid or not.
  * Each routine therefore takes a 'namecheck_err_t' which describes exactly why
  * the name failed to validate.
- *
- * Each function returns 0 on success, -1 on error.
  */
 
 #if !defined(_KERNEL)
@@ -48,6 +46,14 @@
 #include "zfs_namecheck.h"
 #include "zfs_deleg.h"
 
+/*
+ * Deeply nested datasets can overflow the stack, so we put a limit
+ * in the amount of nesting a path can have. zfs_max_dataset_nesting
+ * can be tuned temporarily to fix existing datasets that exceed our
+ * predefined limit.
+ */
+int zfs_max_dataset_nesting = 50;
+
 static int
 valid_char(char c)
 {
@@ -58,10 +64,35 @@ valid_char(char c)
 }
 
 /*
+ * Looks at a path and returns its level of nesting (depth).
+ */
+int
+get_dataset_depth(const char *path)
+{
+	const char *loc = path;
+	int nesting = 0;
+
+	/*
+	 * Keep track of nesting until you hit the end of the
+	 * path or found the snapshot/bookmark seperator.
+	 */
+	for (int i = 0; loc[i] != '\0' &&
+	    loc[i] != '@' &&
+	    loc[i] != '#'; i++) {
+		if (loc[i] == '/')
+			nesting++;
+	}
+
+	return (nesting);
+}
+
+/*
  * Snapshot names must be made up of alphanumeric characters plus the following
  * characters:
  *
- * 	[-_.: ]
+ *	[-_.: ]
+ *
+ * Returns 0 on success, -1 on error.
  */
 int
 zfs_component_namecheck(const char *path, namecheck_err_t *why, char *what)
@@ -97,6 +128,8 @@ zfs_component_namecheck(const char *path, namecheck_err_t *why, char *what)
  * Permissions set name must start with the letter '@' followed by the
  * same character restrictions as snapshot names, except that the name
  * cannot exceed 64 characters.
+ *
+ * Returns 0 on success, -1 on error.
  */
 int
 permset_namecheck(const char *path, namecheck_err_t *why, char *what)
@@ -119,28 +152,40 @@ permset_namecheck(const char *path, namecheck_err_t *why, char *what)
 }
 
 /*
+ * Dataset paths should not be deeper than zfs_max_dataset_nesting
+ * in terms of nesting.
+ *
+ * Returns 0 on success, -1 on error.
+ */
+int
+dataset_nestcheck(const char *path)
+{
+	return ((get_dataset_depth(path) < zfs_max_dataset_nesting) ? 0 : -1);
+}
+
+/*
  * Entity names must be of the following form:
  *
- * 	[component/]*[component][(@|#)component]?
+ *	[component/]*[component][(@|#)component]?
  *
  * Where each component is made up of alphanumeric characters plus the following
  * characters:
  *
- * 	[-_.:%]
+ *	[-_.:%]
  *
  * We allow '%' here as we use that character internally to create unique
  * names for temporary clones (for online recv).
+ *
+ * Returns 0 on success, -1 on error.
  */
 int
 entity_namecheck(const char *path, namecheck_err_t *why, char *what)
 {
-	const char *start, *end;
-	int found_delim;
+	const char *end;
 
 	/*
 	 * Make sure the name is not too long.
 	 */
-
 	if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN) {
 		if (why)
 			*why = NAME_ERR_TOOLONG;
@@ -160,8 +205,8 @@ entity_namecheck(const char *path, namecheck_err_t *why, char *what)
 		return (-1);
 	}
 
-	start = path;
-	found_delim = 0;
+	const char *start = path;
+	boolean_t found_delim = B_FALSE;
 	for (;;) {
 		/* Find the end of this component */
 		end = start;
@@ -196,7 +241,7 @@ entity_namecheck(const char *path, namecheck_err_t *why, char *what)
 				return (-1);
 			}
 
-			found_delim = 1;
+			found_delim = B_TRUE;
 		}
 
 		/* Zero-length components are not allowed */
@@ -248,6 +293,8 @@ dataset_namecheck(const char *path, namecheck_err_t *why, char *what)
  * mountpoint names must be of the following form:
  *
  *	/[component][/]*[component][/]
+ *
+ * Returns 0 on success, -1 on error.
  */
 int
 mountpoint_namecheck(const char *path, namecheck_err_t *why)
@@ -292,6 +339,8 @@ mountpoint_namecheck(const char *path, namecheck_err_t *why)
  * dataset names, with the additional restriction that the pool name must begin
  * with a letter.  The pool names 'raidz' and 'mirror' are also reserved names
  * that cannot be used.
+ *
+ * Returns 0 on success, -1 on error.
  */
 int
 pool_namecheck(const char *pool, namecheck_err_t *why, char *what)
@@ -351,4 +400,10 @@ pool_namecheck(const char *pool, namecheck_err_t *why, char *what)
 EXPORT_SYMBOL(pool_namecheck);
 EXPORT_SYMBOL(dataset_namecheck);
 EXPORT_SYMBOL(zfs_component_namecheck);
+EXPORT_SYMBOL(dataset_nestcheck);
+EXPORT_SYMBOL(get_dataset_depth);
+EXPORT_SYMBOL(zfs_max_dataset_nesting);
+
+module_param(zfs_max_dataset_nesting, int, 0644);
+MODULE_PARM_DESC(zfs_max_dataset_nesting, "Maximum depth of nested datasets");
 #endif