xref: /titanic_44/usr/src/man/man3c/dldump.3c (revision f38cb554a534c6df738be3f4d23327e69888e634)
te
Copyright (c) 2004 by Sun Microsystems, Inc. All rights reserved.
The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License.
You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License.
When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner]
DLDUMP 3C "Mar 1, 2004"
NAME
dldump - create a new file from a dynamic object component of the calling process
SYNOPSIS

#include <dlfcn.h>

int dldump(const char * ipath, const char * opath, int flags);
DESCRIPTION

The dldump() function creates a new dynamic object opath from an existing dynamic object ipath that is bound to the current process. An ipath value of 0 is interpreted as the dynamic object that started the process. The new object is constructed from the existing objects' disc file. Relocations can be applied to the new object to pre-bind it to other dynamic objects, or fix the object to a specific memory location. In addition, data elements within the new object can be obtained from the objects' memory image as this data exists in the calling process.

These techniques allow the new object to be executed with a lower startup cost. This reduction can be because of less relocations being required to load the object, or because of a reduction in the data processing requirements of the object. However, limitations can exist in using these techniques. The application of relocations to the new dynamic object opath can restrict its flexibility within a dynamically changing environment. In addition, limitations in regards to data usage can make dumping a memory image impractical. See EXAMPLES.

The runtime linker verifies that the dynamic object ipath is mapped as part of the current process. Thus, the object must either be the dynamic object that started the process, one of the process's dependencies, or an object that has been preloaded. See exec(2), and ld.so.1(1).

As part of the runtime processing of a dynamic object, relocation records within the object are interpreted and applied to offsets within the object. These offsets are said to be relocated. Relocations can be categorized into two basic types: non-symbolic and symbolic.

The non-symbolic relocation is a simple relative relocation that requires the base address at which the object is mapped to perform the relocation. The symbolic relocation requires the address of an associated symbol, and results in a binding to the dynamic object that defines this symbol. The symbol definition can originate from any of the dynamic objects that make up the process, that is, the object that started the process, one of the process's dependencies, an object that has been preloaded, or the dynamic object being relocated.

The flags parameter controls the relocation processing and other attributes of producing the new dynamic object opath. Without any flags, the new object is constructed solely from the contents of the ipath disc file without any relocations applied.

Various relocation flags can be or'ed into the flags parameter to affect the relocations that are applied to the new object. Non-symbolic relocations can be applied using the following: RTLD_REL_RELATIVE

Relocation records from the object ipath, that define relative relocations, are applied to the object opath.

A variety of symbolic relocations can be applied using the following flags (each of these flags also implies RTLD_REL_RELATIVE is in effect): RTLD_REL_EXEC

Symbolic relocations that result in binding ipath to the dynamic object that started the process, commonly a dynamic executable, are applied to the object opath.

RTLD_REL_DEPENDS

Symbolic relocations that result in binding ipath to any of the dynamic dependencies of the process are applied to the object opath.

RTLD_REL_PRELOAD

Symbolic relocations that result in binding ipath to any objects preloaded with the process are applied to the object opath. See LD_PRELOAD in ld.so.1(1).

RTLD_REL_SELF

Symbolic relocations that result in binding ipath to itself, are applied to the object opath.

RTLD_REL_WEAK

Weak relocations that remain unresolved are applied to the object opath as 0.

RTLD_REL_ALL

All relocation records defined in the object ipath are applied to the new object opath. This is basically a concatenation of all the above relocation flags.

Note that for dynamic executables, RTLD_REL_RELATIVE, RTLD_REL_EXEC, and RTLD_REL_SELF have no effect. See EXAMPLES.

If relocations, knowledgeable of the base address of the mapped object, are applied to the new object opath, then the new object becomes fixed to the location that the ipath image is mapped within the current process.

Any relocations applied to the new object opath will have the original relocation record removed so that the relocation will not be applied more than once. Otherwise, the new object opath will retain the relocation records as they exist in the ipath disc file.

The following additional attributes for creating the new dynamic object opath can be specified using the flags parameter: RTLD_MEMORY

The new object opath is constructed from the current memory contents of the ipath image as it exists in the calling process. This option allows data modified by the calling process to be captured in the new object. Note that not all data modifications may be applicable for capture; significant restrictions exist in using this technique. See EXAMPLES. By default, when processing a dynamic executable, any allocated memory that follows the end of the data segment is captured in the new object (see malloc(3C) and brk(2)). This data, which represents the process heap, is saved as a new .SUNW_heap section in the object opath. The objects' program headers and symbol entries, such as _end, are adjusted accordingly. See also RTLD_NOHEAP. When using this attribute, any relocations that have been applied to the ipath memory image that do not fall into one of the requested relocation categories are undone, that is, the relocated element is returned to the value as it existed in the ipath disc file.

RTLD_STRIP

Only collect allocatable sections within the object opath. Sections that are not part of the dynamic objects' memory image are removed. RTLD_STRIP reduces the size of the opath disc file and is comparable to having run the new object through strip(1).

RTLD_NOHEAP

Do not save any heap to the new object. This option is only meaningful when processing a dynamic executable with the RTLD_MEMORY attribute and allows for reducing the size of the opath disc file. The executable must confine its data initialization to data elements within its data segment, and must not use any allocated data elements that comprise the heap.

It should be emphasized, that an object created by dldump() is simply an updated ELF object file. No additional state regarding the process at the time dldump() is called is maintained in the new object. dldump() does not provide a panacea for checkpoint and resume. A new dynamic executable, for example, will not start where the original executable called dldump(). It will gain control at the executable's normal entry point. See EXAMPLES.

RETURN VALUES

On successful creation of the new object, dldump() returns 0. Otherwise, a non-zero value is returned and more detailed diagnostic information is available through dlerror().

EXAMPLES

Example 1 Sample code using dldump().

The following code fragment, which can be part of a dynamic executable a.out, can be used to create a new shared object from one of the dynamic executables' dependencies libfoo.so.1:

const char * ipath = "libfoo.so.1";
const char * opath = "./tmp/libfoo.so.1";
...
if (dldump(ipath, opath, RTLD_REL_RELATIVE) != 0)
 (void) printf("dldump failed: %s\en", dlerror(\|));

The new shared object opath is fixed to the address of the mapped ipath bound to the dynamic executable a.out. All relative relocations are applied to this new shared object, which will reduce its relocation overhead when it is used as part of another process.

By performing only relative relocations, any symbolic relocation records remain defined within the new object, and thus the dynamic binding to external symbols will be preserved when the new object is used.

Use of the other relocation flags can fix specific relocations in the new object and thus can reduce even more the runtime relocation startup cost of the new object. However, this will also restrict the flexibility of using the new object within a dynamically changing environment, as it will bind the new object to some or all of the dynamic objects presently mapped as part of the process.

For example, the use of RTLD_REL_SELF will cause any references to symbols from ipath to be bound to definitions within itself if no other preceding object defined the same symbol. In other words, a call to foo(\|) within ipath will bind to the definition foo within the same object. Therefore, opath will have one less binding that must be computed at runtime. This reduces the startup cost of using opath by other applications; however, interposition of the symbol foo will no longer be possible.

Using a dumped shared object with applied relocations as an applications dependency normally requires that the application have the same dependencies as the application that produced the dumped image. Dumping shared objects, and the various flags associated with relocation processing, have some specialized uses. However, the technique is intended as a building block for future technology.

The following code fragment, which is part of the dynamic executable a.out, can be used to create a new version of the dynamic executable:

static char * dumped = 0;
const char * opath = "./a.out.new";
...
if (dumped == 0) {
 char buffer[100];
 int size;
 time_t seconds;
 ...
 /* Perform data initialization */
 seconds = time((time_t *)0);
 size = cftime(buffer, (char *)0, &seconds);
 if ((dumped = (char *)malloc(size + 1)) == 0) {
 (void) printf("malloc failed: %s\en", strerror(errno));
 return (1);
 }
 (void) strcpy(dumped, buffer);
 ...
 /*
 * Tear down any undesirable data initializations and
 * dump the dynamic executables memory image.
 */
 _exithandle(\|);
 _exit(dldump(0, opath, RTLD_MEMORY));
}
(void) printf("Dumped: %s\en", dumped);

Any modifications made to the dynamic executable, up to the point the dldump() call is made, are saved in the new object a.out.new. This mechanism allows the executable to update parts of its data segment and heap prior to creating the new object. In this case, the date the executable is dumped is saved in the new object. The new object can then be executed without having to carry out the same (presumably expensive) initialization.

For greatest flexibility, this example does not save any relocated information. The elements of the dynamic executable ipath that have been modified by relocations at process startup, that is, references to external functions, are returned to the values of these elements as they existed in the ipath disc file. This preservation of relocation records allows the new dynamic executable to be flexible, and correctly bind and initialize to its dependencies when executed on the same or newer upgrades of the OS.

Fixing relocations by applying some of the relocation flags would bind the new object to the dependencies presently mapped as part of the process calling dldump(). It may also remove necessary copy relocation processing required for the correct initialization of its shared object dependencies. Therefore, if the new dynamic executables' dependencies have no specialized initialization requirements, the executable may still only interact correctly with the dependencies to which it binds if they were mapped to the same locations as they were when dldump() was called.

Note that for dynamic executables, RTLD_REL_RELATIVE, RTLD_REL_EXEC, and RTLD_REL_SELF have no effect, as relocations within the dynamic executable will have been fixed when it was created by ld(1).

When RTLD_MEMORY is used, care should be taken to insure that dumped data sections that reference external objects are not reused without appropriate re-initialization. For example, if a data item contains a file descriptor, a variable returned from a shared object, or some other external data, and this data item has been initialized prior to the dldump() call, its value will have no meaning in the new dumped image.

When RTLD_MEMORY is used, any modification to a data item that is initialized via a relocation whose relocation record will be retained in the new image will effectively be lost or invalidated within the new image. For example, if a pointer to an external object is incremented prior to the dldump() call, this data item will be reset to its disc file contents so that it can be relocated when the new image is used; hence, the previous increment is lost.

Non-idempotent data initializations may prevent the use of RTLD_MEMORY. For example, the addition of elements to a linked-list via init sections can result in the linked-list data being captured in the new image. Running this new image may result in init sections continuing to add new elements to the list without the prerequisite initialization of the list head. It is recommended that _exithandle(3C) be called before dldump() to tear down any data initializations established via initialization code. Note that this may invalidate the calling image; thus, following the call to dldump(), only a call to _Exit(2) should be made.

USAGE

The dldump() function is one of a family of functions that give the user direct access to the dynamic linking facilities. These facilities are available to dynamically-linked processes only. See Linker and Libraries Guide).

ATTRIBUTES

See attributes(5) for descriptions of the following attributes:

ATTRIBUTE TYPE ATTRIBUTE VALUE
MT-Level MT-Safe
SEE ALSO

ld(1), ld.so.1(1), strip(1), _Exit(2), brk(2), exec(2), _exithandle(3C), dladdr(3C), dlclose(3C), dlerror(3C), dlopen(3C), dlsym(3C), end(3C), malloc(3C), attributes(5)

Linker and Libraries Guide

NOTES

These functions are available to dynamically-linked processes only.

Any NOBITS sections within the ipath are expanded to PROGBITS sections within the opath. NOBITS sections occupy no space within an ELF file image. NOBITS sections declare memory that must be created and zero-filled when the object is mapped into the runtime environment. .bss is a typical example of this section type. PROGBITS sections, on the other hand, hold information defined by the object within the ELF file image. This section conversion reduces the runtime initialization cost of the new dumped object but increases the objects' disc space requirement.

When a shared object is dumped, and relocations are applied which are knowledgeable of the base address of the mapped object, the new object is fixed to this new base address. The dumped object has its ELF type reclassified to be a dynamic executable. The dumped object can be processed by the runtime linker, but is not valid as input to the link-editor.

If relocations are applied to the new object, any remaining relocation records are reorganized for better locality of reference. The relocation sections are renamed to .SUNW_reloc and the association with the section to relocate, is lost. Only the offset of the relocation record is meaningful. .SUNW_reloc relocations do not make the new object invalid to either the runtime linker or link-editor, but can reduce the objects analysis with some ELF readers.