2015-04-10 talloc 3 Samba System Administration tools 4.0 talloc hierarchical reference counted memory pool system with destructors #include <talloc.h> If you are used to talloc from Samba3 then please read this carefully, as talloc has changed a lot. The new talloc is a hierarchical, reference counted memory pool system with destructors. Quite a mouthful really, but not too bad once you get used to it. Perhaps the biggest change from Samba3 is that there is no distinction between a "talloc context" and a "talloc pointer". Any pointer returned from talloc() is itself a valid talloc context. This means you can do this: struct foo *X = talloc(mem_ctx, struct foo); X->name = talloc_strdup(X, "foo"); and the pointer X->name would be a "child" of the talloc context X which is itself a child of mem_ctx. So if you do talloc_free(mem_ctx) then it is all destroyed, whereas if you do talloc_free(X) then just X and X->name are destroyed, and if you do talloc_free(X->name) then just the name element of X is destroyed. If you think about this, then what this effectively gives you is an n-ary tree, where you can free any part of the tree with talloc_free(). If you find this confusing, then I suggest you run the testsuite program to watch talloc in action. You may also like to add your own tests to testsuite.c to clarify how some particular situation is handled. The following is a complete guide to the talloc API. Read it all at least twice. The talloc() macro is the core of the talloc library. It takes a memory ctx and a type, and returns a pointer to a new area of memory of the given type. The returned pointer is itself a talloc context, so you can use it as the ctx argument to more calls to talloc() if you wish. The returned pointer is a "child" of the supplied context. This means that if you talloc_free() the ctx then the new child disappears as well. Alternatively you can free just the child. The ctx argument to talloc() can be NULL, in which case a new top level context is created. The function talloc_size() should be used when you don't have a convenient type to pass to talloc(). Unlike talloc(), it is not type safe (as it returns a void *), so you are on your own for type checking. The talloc_ptrtype() macro should be used when you have a pointer and want to allocate memory to point at with this pointer. When compiling with gcc >= 3 it is typesafe. Note this is a wrapper of talloc_size() and talloc_get_name() will return the current location in the source file. and not the type. The talloc_free() function frees a piece of talloc memory, and all its children. You can call talloc_free() on any pointer returned by talloc(). The return value of talloc_free() indicates success or failure, with 0 returned for success and -1 for failure. The only possible failure condition is if ptr had a destructor attached to it and the destructor returned -1. See talloc_set_destructor() for details on destructors. If this pointer has an additional parent when talloc_free() is called then the memory is not actually released, but instead the most recently established parent is destroyed. See talloc_reference() for details on establishing additional parents. For more control on which parent is removed, see talloc_unlink(). talloc_free() operates recursively on its children. From the 2.0 version of talloc, as a special case, talloc_free() is refused on pointers that have more than one parent, as talloc would have no way of knowing which parent should be removed. To free a pointer that has more than one parent please use talloc_unlink(). To help you find problems in your code caused by this behaviour, if you do try and free a pointer with more than one parent then the talloc logging function will be called to give output like this: ERROR: talloc_free with references at some_dir/source/foo.c:123 reference at some_dir/source/other.c:325 reference at some_dir/source/third.c:121 Please see the documentation for talloc_set_log_fn() and talloc_set_log_stderr() for more information on talloc logging functions. The talloc_reference() function makes ctx an additional parent of ptr. The return value of talloc_reference() is always the original pointer ptr, unless talloc ran out of memory in creating the reference in which case it will return NULL (each additional reference consumes around 48 bytes of memory on intel x86 platforms). If ptr is NULL, then the function is a no-op, and simply returns NULL. After creating a reference you can free it in one of the following ways: you can talloc_free() any parent of the original pointer. That will reduce the number of parents of this pointer by 1, and will cause this pointer to be freed if it runs out of parents. you can talloc_free() the pointer itself if it has at maximum one parent. This behaviour has been changed since the release of version 2.0. Further information in the description of "talloc_free". For more control on which parent to remove, see talloc_unlink(). The talloc_unlink() function removes a specific parent from ptr. The ctx passed must either be a context used in talloc_reference() with this pointer, or must be a direct parent of ptr. Note that if the parent has already been removed using talloc_free() then this function will fail and will return -1. Likewise, if ptr is NULL, then the function will make no modifications and return -1. Usually you can just use talloc_free() instead of talloc_unlink(), but sometimes it is useful to have the additional control on which parent is removed. The function talloc_set_destructor() sets the destructor for the pointer ptr. A destructor is a function that is called when the memory used by a pointer is about to be released. The destructor receives ptr as an argument, and should return 0 for success and -1 for failure. The destructor can do anything it wants to, including freeing other pieces of memory. A common use for destructors is to clean up operating system resources (such as open file descriptors) contained in the structure the destructor is placed on. You can only place one destructor on a pointer. If you need more than one destructor then you can create a zero-length child of the pointer and place an additional destructor on that. To remove a destructor call talloc_set_destructor() with NULL for the destructor. If your destructor attempts to talloc_free() the pointer that it is the destructor for then talloc_free() will return -1 and the free will be ignored. This would be a pointless operation anyway, as the destructor is only called when the memory is just about to go away. The talloc_increase_ref_count(ptr) function is exactly equivalent to: talloc_reference(NULL, ptr); You can use either syntax, depending on which you think is clearer in your code. It returns 0 on success and -1 on failure. Return the number of references to the pointer. Each talloc pointer has a "name". The name is used principally for debugging purposes, although it is also possible to set and get the name on a pointer in as a way of "marking" pointers in your code. The main use for names on pointer is for "talloc reports". See talloc_report_depth_cb(), talloc_report_depth_file(), talloc_report() talloc_report() and talloc_report_full() for details. Also see talloc_enable_leak_report() and talloc_enable_leak_report_full(). The talloc_set_name() function allocates memory as a child of the pointer. It is logically equivalent to: talloc_set_name_const(ptr, talloc_asprintf(ptr, fmt, ...)); Note that multiple calls to talloc_set_name() will allocate more memory without releasing the name. All of the memory is released when the ptr is freed using talloc_free(). The function talloc_set_name_const() is just like talloc_set_name(), but it takes a string constant, and is much faster. It is extensively used by the "auto naming" macros, such as talloc_p(). This function does not allocate any memory. It just copies the supplied pointer into the internal representation of the talloc ptr. This means you must not pass a name pointer to memory that will disappear before ptr is freed with talloc_free(). The talloc_named() function creates a named talloc pointer. It is equivalent to: ptr = talloc_size(ctx, size); talloc_set_name(ptr, fmt, ....); This is equivalent to: ptr = talloc_size(ctx, size); talloc_set_name_const(ptr, name); This returns the current name for the given talloc pointer, ptr. See talloc_set_name() for details. This function creates a zero length named talloc context as a top level context. It is equivalent to: talloc_named(NULL, 0, fmt, ...); This is a utility macro that creates a new memory context hanging off an existing context, automatically naming it "talloc_new: __location__" where __location__ is the source line it is called from. It is particularly useful for creating a new temporary working context. The talloc_realloc() macro changes the size of a talloc pointer. It has the following equivalences: talloc_realloc(ctx, NULL, type, 1) ==> talloc(ctx, type); talloc_realloc(ctx, ptr, type, 0) ==> talloc_free(ptr); The ctx argument is only used if ptr is not NULL, otherwise it is ignored. talloc_realloc() returns the new pointer, or NULL on failure. The call will fail either due to a lack of memory, or because the pointer has more than one parent (see talloc_reference()). the talloc_realloc_size() function is useful when the type is not known so the type-safe talloc_realloc() cannot be used. The talloc_steal() function changes the parent context of a talloc pointer. It is typically used when the context that the pointer is currently a child of is going to be freed and you wish to keep the memory for a longer time. The talloc_steal() function returns the pointer that you pass it. It does not have any failure modes. It is possible to produce loops in the parent/child relationship if you are not careful with talloc_steal(). No guarantees are provided as to your sanity or the safety of your data if you do this. Note that if you try and call talloc_steal() on a pointer that has more than one parent then the result is ambiguous. Talloc will choose to remove the parent that is currently indicated by talloc_parent() and replace it with the chosen parent. You will also get a message like this via the talloc logging functions: WARNING: talloc_steal with references at some_dir/source/foo.c:123 reference at some_dir/source/other.c:325 reference at some_dir/source/third.c:121 To unambiguously change the parent of a pointer please see the function talloc_reparent(). See the talloc_set_log_fn() documentation for more information on talloc logging. The talloc_reparent() function changes the parent context of a talloc pointer. It is typically used when the context that the pointer is currently a child of is going to be freed and you wish to keep the memory for a longer time. The talloc_reparent() function returns the pointer that you pass it. It does not have any failure modes. The difference between talloc_reparent() and talloc_steal() is that talloc_reparent() can specify which parent you wish to change. This is useful when a pointer has multiple parents via references. The talloc_move() function is a wrapper around talloc_steal() which zeros the source pointer after the move. This avoids a potential source of bugs where a programmer leaves a pointer in two structures, and uses the pointer from the old structure after it has been moved to a new one. The talloc_total_size() function returns the total size in bytes used by this pointer and all child pointers. Mostly useful for debugging. Passing NULL is allowed, but it will only give a meaningful result if talloc_enable_leak_report() or talloc_enable_leak_report_full() has been called. The talloc_total_blocks() function returns the total memory block count used by this pointer and all child pointers. Mostly useful for debugging. Passing NULL is allowed, but it will only give a meaningful result if talloc_enable_leak_report() or talloc_enable_leak_report_full() has been called. The talloc_report() function prints a summary report of all memory used by ptr. One line of report is printed for each immediate child of ptr, showing the total memory and number of blocks used by that child. You can pass NULL for the pointer, in which case a report is printed for the top level memory context, but only if talloc_enable_leak_report() or talloc_enable_leak_report_full() has been called. This provides a more detailed report than talloc_report(). It will recursively print the entire tree of memory referenced by the pointer. References in the tree are shown by giving the name of the pointer that is referenced. You can pass NULL for the pointer, in which case a report is printed for the top level memory context, but only if talloc_enable_leak_report() or talloc_enable_leak_report_full() has been called. void talloc_report_depth_cb const void *ptr int depth int max_depth void (*callback)(const void *ptr, int depth, int max_depth, int is_ref, void *priv) void *priv This provides a more flexible reports than talloc_report(). It will recursively call the callback for the entire tree of memory referenced by the pointer. References in the tree are passed with is_ref = 1 and the pointer that is referenced. You can pass NULL for the pointer, in which case a report is printed for the top level memory context, but only if talloc_enable_leak_report() or talloc_enable_leak_report_full() has been called. The recursion is stopped when depth >= max_depth. max_depth = -1 means only stop at leaf nodes. void talloc_report_depth_file const void *ptr int depth int max_depth FILE *f This provides a more flexible reports than talloc_report(). It will let you specify the depth and max_depth. This enables calling of talloc_report(NULL, stderr) when the program exits. In Samba4 this is enabled by using the --leak-report command line option. For it to be useful, this function must be called before any other talloc function as it establishes a "null context" that acts as the top of the tree. If you don't call this function first then passing NULL to talloc_report() or talloc_report_full() won't give you the full tree printout. Here is a typical talloc report: talloc report on 'null_context' (total 267 bytes in 15 blocks) libcli/auth/spnego_parse.c:55 contains 31 bytes in 2 blocks libcli/auth/spnego_parse.c:55 contains 31 bytes in 2 blocks iconv(UTF8,CP850) contains 42 bytes in 2 blocks libcli/auth/spnego_parse.c:55 contains 31 bytes in 2 blocks iconv(CP850,UTF8) contains 42 bytes in 2 blocks iconv(UTF8,UTF-16LE) contains 45 bytes in 2 blocks iconv(UTF-16LE,UTF8) contains 45 bytes in 2 blocks This enables calling of talloc_report_full(NULL, stderr) when the program exits. In Samba4 this is enabled by using the --leak-report-full command line option. For it to be useful, this function must be called before any other talloc function as it establishes a "null context" that acts as the top of the tree. If you don't call this function first then passing NULL to talloc_report() or talloc_report_full() won't give you the full tree printout. Here is a typical full report: full talloc report on 'root' (total 18 bytes in 8 blocks) p1 contains 18 bytes in 7 blocks (ref 0) r1 contains 13 bytes in 2 blocks (ref 0) reference to: p2 p2 contains 1 bytes in 1 blocks (ref 1) x3 contains 1 bytes in 1 blocks (ref 0) x2 contains 1 bytes in 1 blocks (ref 0) x1 contains 1 bytes in 1 blocks (ref 0) The talloc_zero() macro is equivalent to: ptr = talloc(ctx, type); if (ptr) memset(ptr, 0, sizeof(type)); The talloc_zero_size() function is useful when you don't have a known type. The talloc_memdup() function is equivalent to: ptr = talloc_size(ctx, size); if (ptr) memcpy(ptr, p, size); The talloc_strdup() function is equivalent to: ptr = talloc_size(ctx, strlen(p)+1); if (ptr) memcpy(ptr, p, strlen(p)+1); This function sets the name of the new pointer to the passed string. This is equivalent to: talloc_set_name_const(ptr, ptr) The talloc_strndup() function is the talloc equivalent of the C library function strndup(3). This function sets the name of the new pointer to the passed string. This is equivalent to: talloc_set_name_const(ptr, ptr) The talloc_vasprintf() function is the talloc equivalent of the C library function vasprintf(3). This function sets the name of the new pointer to the new string. This is equivalent to: talloc_set_name_const(ptr, ptr) The talloc_asprintf() function is the talloc equivalent of the C library function asprintf(3). This function sets the name of the new pointer to the passed string. This is equivalent to: talloc_set_name_const(ptr, ptr) The talloc_asprintf_append() function appends the given formatted string to the given string. This function sets the name of the new pointer to the new string. This is equivalent to: talloc_set_name_const(ptr, ptr) The talloc_array() macro is equivalent to: (type *)talloc_size(ctx, sizeof(type) * count); except that it provides integer overflow protection for the multiply, returning NULL if the multiply overflows. The talloc_array_size() function is useful when the type is not known. It operates in the same way as talloc_array(), but takes a size instead of a type. The talloc_ptrtype() macro should be used when you have a pointer to an array and want to allocate memory of an array to point at with this pointer. When compiling with gcc >= 3 it is typesafe. Note this is a wrapper of talloc_array_size() and talloc_get_name() will return the current location in the source file. and not the type. This is a non-macro version of talloc_realloc(), which is useful as libraries sometimes want a realloc function pointer. A realloc(3) implementation encapsulates the functionality of malloc(3), free(3) and realloc(3) in one call, which is why it is useful to be able to pass around a single function pointer. This is a handy utility function that returns a talloc context which will be automatically freed on program exit. This can be used to reduce the noise in memory leak reports. This function checks if a pointer has the specified name. If it does then the pointer is returned. It it doesn't then NULL is returned. This macro allows you to do type checking on talloc pointers. It is particularly useful for void* private pointers. It is equivalent to this: (type *)talloc_check_name(ptr, #type) This macro allows you to force the name of a pointer to be a particular type. This can be used in conjunction with talloc_get_type() to do type checking on void* pointers. It is equivalent to this: talloc_set_name_const(ptr, #type) This function sets a logging function that talloc will use for warnings and errors. By default talloc will not print any warnings or errors. This sets the talloc log function to write log messages to stderr All the additional features of talloc(3) over malloc(3) do come at a price. We have a simple performance test in Samba4 that measures talloc() versus malloc() performance, and it seems that talloc() is about 10% slower than malloc() on my x86 Debian Linux box. For Samba, the great reduction in code complexity that we get by using talloc makes this worthwhile, especially as the total overhead of talloc/malloc in Samba is already quite small. malloc(3), strndup(3), vasprintf(3), asprintf(3), The original Samba software and related utilities were created by Andrew Tridgell. Samba is now developed by the Samba Team as an Open Source project similar to the way the Linux kernel is developed. Copyright (C) Andrew Tridgell 2004 This program is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, see http://www.gnu.org/licenses/.