Before now, we've been able to write CTF files to gzFile descriptors or
fds, and CTF archives to named files only.
Make this a bit less irregular by allowing CTF archives to be written
to fds with the new function ctf_arc_write_fd: also allow CTF
files to be written to a new memory buffer via ctf_write_mem.
(It would be nice to complete things by adding a new function to write
CTF archives to memory, but this is too difficult to do given the short
time the linker is expected to be writing them out: we will transition
to a better format in format v4, though we will always support reading
CTF archives that are stored in .ctf sections.)
include/
* ctf-api.h (ctf_arc_write_fd): New.
(ctf_write_mem): Likewise.
(ctf_gzwrite): Spacing fix.
libctf/
* ctf-archive.c (ctf_arc_write): Split off, and reimplement in terms
of...
(ctf_arc_write_fd): ... this new function.
* ctf-create.c (ctf_write_mem): New.
The CTF file format has always supported "external strtabs", which
internally are strtab offsets with their MSB on: such refs
get their strings from the strtab passed in at CTF file open time:
this is usually intended to be the ELF strtab, and that's what this
implementation is meant to support, though in theory the external
strtab could come from anywhere.
This commit adds support for these external strings in the ctf-string.c
strtab tracking layer. It's quite easy: we just add a field csa_offset
to the atoms table that tracks all strings: this field tracks the offset
of the string in the ELF strtab (with its MSB already on, courtesy of a
new macro CTF_SET_STID), and adds a new function that sets the
csa_offset to the specified offset (plus MSB). Then we just need to
avoid writing out strings to the internal strtab if they have csa_offset
set, and note that the internal strtab is shorter than it might
otherwise be.
(We could in theory save a little more time here by eschewing sorting
such strings, since we never actually write the strings out anywhere,
but that would mean storing them separately and it's just not worth the
complexity cost until profiling shows it's worth doing.)
We also have to go through a bit of extra effort at variable-sorting
time. This was previously using direct references to the internal
strtab: it couldn't use ctf_strptr or ctf_strraw because the new strtab
is not yet ready to put in its usual field (in a ctf_file_t that hasn't
even been allocated yet at this stage): but now we're using the external
strtab, this will no longer do because it'll be looking things up in the
wrong strtab, with disastrous results. Instead, pass the new internal
strtab in to a new ctf_strraw_explicit function which is just like
ctf_strraw except you can specify a ne winternal strtab to use.
But even now that it is using a new internal strtab, this is not quite
enough: it can't look up strings in the external strtab because ld
hasn't written it out yet, and when it does will write it straight to
disk. Instead, when we write the internal strtab, note all the offset
-> string mappings that we have noted belong in the *external* strtab to
a new "synthetic external strtab" dynhash, ctf_syn_ext_strtab, and look
in there at ctf_strraw time if it is set. This uses minimal extra
memory (because only strings in the external strtab that we actually use
are stored, and even those come straight out of the atoms table), but
let both variable sorting and name interning when ctf_bufopen is next
called work fine. (This also means that we don't need to filter out
spurious ECTF_STRTAB warnings from ctf_bufopen but can pass them back to
the caller, once we wrap ctf_bufopen so that we have a new internal
variant of ctf_bufopen etc that we can pass the synthetic external
strtab to. That error has been filtered out since the days of Solaris
libctf, which didn't try to handle the problem of getting external
strtabs right at construction time at all.)
v3: add the synthetic strtab and all associated machinery.
v5: fix tabdamage.
include/
* ctf.h (CTF_SET_STID): New.
libctf/
* ctf-impl.h (ctf_str_atom_t) <csa_offset>: New field.
(ctf_file_t) <ctf_syn_ext_strtab>: Likewise.
(ctf_str_add_ref): Name the last arg.
(ctf_str_add_external) New.
(ctf_str_add_strraw_explicit): Likewise.
(ctf_simple_open_internal): Likewise.
(ctf_bufopen_internal): Likewise.
* ctf-string.c (ctf_strraw_explicit): Split from...
(ctf_strraw): ... here, with new support for ctf_syn_ext_strtab.
(ctf_str_add_ref_internal): Return the atom, not the
string.
(ctf_str_add): Adjust accordingly.
(ctf_str_add_ref): Likewise. Move up in the file.
(ctf_str_add_external): New: update the csa_offset.
(ctf_str_count_strtab): Only account for strings with no csa_offset
in the internal strtab length.
(ctf_str_write_strtab): If the csa_offset is set, update the
string's refs without writing the string out, and update the
ctf_syn_ext_strtab. Make OOM handling less ugly.
* ctf-create.c (struct ctf_sort_var_arg_cb): New.
(ctf_update): Handle failure to populate the strtab. Pass in the
new ctf_sort_var arg. Adjust for ctf_syn_ext_strtab addition.
Call ctf_simple_open_internal, not ctf_simple_open.
(ctf_sort_var): Call ctf_strraw_explicit rather than looking up
strings by hand.
* ctf-hash.c (ctf_hash_insert_type): Likewise (but using
ctf_strraw). Adjust to diagnose ECTF_STRTAB nonetheless.
* ctf-open.c (init_types): No longer filter out ECTF_STRTAB.
(ctf_file_close): Destroy the ctf_syn_ext_strtab.
(ctf_simple_open): Rename to, and reimplement as a wrapper around...
(ctf_simple_open_internal): ... this new function, which calls
ctf_bufopen_internal.
(ctf_bufopen): Rename to, and reimplement as a wrapper around...
(ctf_bufopen_internal): ... this new function, which sets
ctf_syn_ext_strtab.
The existing function ctf_type_iter lets you iterate over root-visible
types (types you can look up by name). There is no way to iterate over
non-root-visible types, which is troublesome because both the linker
and dumper want to do that.
So add a new function that can do it: the callback it takes accepts
an extra parameter which indicates whether the type is root-visible
or not.
include/
* ctf-api.h (ctf_type_all_f): New.
(ctf_type_iter_all): New.
libctf/
* ctf_types.c (ctf_type_iter_all): New.
No code handles these yet, but our latest GCC patches are generating
them, so we have to be ready for them or erroneously conclude that we
have file corruption.
(This simultaneously fixes a longstanding bug, concealed because nothing
was generating anything in the object or function info sections, where
the end of the section was being tested against the wrong thing: it
would have walked over the entire contents of the variable section and
treated them as part of the function info section. This had to change
now anyway because the new sections have landed in between.)
include/
* ctf.h: Add object index and function index sections. Describe
them. Improve the description of the variable section and clarify
the constraints on backward-pointing type nodes.
(ctf_header): Add cth_objtidxoff, cth_funcidxoff.
libctf/
* ctf-open.c (init_symtab): Check for overflow against the right
section.
(upgrade_header): Set cth_objtidxoff, cth_funcidxoff to zero-length.
(upgrade_types_v1): Note that these sections are not checked.
(flip_header): Endian-swap the header fields.
(flip_ctf): Endian-swap the sections.
(flip_objts): Update comment.
(ctf_bufopen): Check header offsets and alignment for validity.
The code in ctf_bfdopen_ctfsect (which is the ultimate place where you
end up if you use ctf_open to open a CTF file and pull in the ELF string
and symbol tables) was written before it was possible to actually test
it, since the linker was not written. Now it is, it turns out that the
previous code was completely nonfunctional: it assumed that you could
load the symbol table via bfd_section_from_elf_index (...,elf_onesymtab())
and the string table via bfd_section_from_elf_index on the sh_link.
Unfortunately BFD loads neither of these sections in the conventional
fashion it uses for most others: the symbol table is immediately
converted into internal form (which is useless for our purposes, since
we also have to work in the absence of BFD for readelf, etc) and the
string table is loaded specially via bfd_elf_get_str_section which is
private to bfd/elf.c.
So make this function public, export it in elf-bfd.h, and use it from
libctf, which does something similar to what bfd_elf_sym_name and
bfd_elf_string_from_elf_section do. Similarly, load the symbol table
manually using bfd_elf_get_elf_syms and throw away the internal form
it generates for us (we never use it).
BFD allocates the strtab for us via bfd_alloc, so we can leave BFD to
deallocate it: we allocate the symbol table ourselves before calling
bfd_elf_get_elf_syms, so we still have to free it.
Also change the rules around what you are allowed to provide: It is
useful to provide a string section but no symbol table, because CTF
sections can legitimately have no function info or data object sections
while relying on the ELF strtab for some of their strings. So allow
that combination.
v4: adjust to upstream changes. ctf_bfdopen_ctfsect's first parameter
is potentially unused again (if BFD is not in use for this link
due to not supporting an ELF target).
v5: fix tabdamage.
bfd/
* elf-bfd.h (bfd_elf_get_str_section): Add.
* elf.c (bfd_elf_get_str_section): No longer static.
libctf/
* ctf-open-bfd.c: Add <assert.h>.
(ctf_bfdopen_ctfsect): Open string and symbol tables using
techniques borrowed from bfd_elf_sym_name.
(ctf_new_archive_internal): Improve comment.
* ctf-archive.c (ctf_arc_close): Do not free the ctfi_strsect.
* ctf-open.c (ctf_bufopen): Allow opening with a string section but
no symbol section, but not vice versa.
The CTF header has before now been thrown away too soon to be dumped
using the ctf_dump() machinery used by objdump and readelf: instead, a
kludge involving debugging-priority dumps of the header offsets on every
open was used.
Replace this with proper first-class dumping machinery just like
everything else in the CTF file, and have objdump and readelf use it.
(The dumper already had an enum value in ctf_sect_names_t for this
purpose, waiting to be used.)
v5: fix tabdamage.
libctf/
* ctf-impl.h (ctf_file_t): New field ctf_openflags.
* ctf-open.c (ctf_bufopen): Set it. No longer dump header offsets.
* ctf-dump.c (dump_header): New function, dump the CTF header.
(ctf_dump): Call it.
(ctf_dump_header_strfield): New function.
(ctf_dump_header_sectfield): Likewise.
binutils/
* objdump.c (dump_ctf_archive_member): Dump the CTF header.
* readelf.c (dump_section_as_ctf): Likewise.
libctf supports dynamic upgrading of the type table as file format
versions change, but before now has not supported changes to the CTF
header. Doing this is complicated by the baroque storage method used:
the CTF header is kept prepended to the rest of the CTF data, just as
when read from the file, and written out from there, and is
endian-flipped in place.
This makes accessing it needlessly hard and makes it almost impossible
to make the header larger if we add fields. The general storage
machinery around the malloced ctf pointer (the 'ctf_base') is also
overcomplicated: the pointer is sometimes malloced locally and sometimes
assigned from a parameter, so freeing it requires checking to see if
that parameter was used, needlessly coupling ctf_bufopen and
ctf_file_close together.
So split the header out into a new ctf_file_t.ctf_header, which is
written out explicitly: squeeze it out of the CTF buffer whenever we
reallocate it, and use ctf_file_t.ctf_buf to skip past the header when
we do not need to reallocate (when no upgrading or endian-flipping is
required). We now track whether the CTF base can be freed explicitly
via a new ctf_dynbase pointer which is non-NULL only when freeing is
possible.
With all this done, we can upgrade the header on the fly and add new
fields as desired, via a new upgrade_header function in ctf-open.
As with other forms of upgrading, libctf upgrades older headers
automatically to the latest supported version at open time.
For a first use of this field, we add a new string field cth_cuname, and
a corresponding setter/getter pair ctf_cuname_set and ctf_cuname: this
is used by debuggers to determine whether a CTF section's types relate
to a single compilation unit, or to all compilation units in the
program. (Types with ambiguous definitions in different CUs have only
one of these types placed in the top-level shared .ctf container: the
rest are placed in much smaller per-CU containers, which have the shared
container as their parent. Since CTF must be useful in the absence of
DWARF, we store the names of the relevant CUs ourselves, so the debugger
can look them up.)
v5: fix tabdamage.
include/
* ctf-api.h (ctf_cuname): New function.
(ctf_cuname_set): Likewise.
* ctf.h: Improve comment around upgrading, no longer
implying that v2 is the target of upgrades (it is v3 now).
(ctf_header_v2_t): New, old-format header for backward
compatibility.
(ctf_header_t): Add cth_cuname: this is the first of several
header changes in format v3.
libctf/
* ctf-impl.h (ctf_file_t): New fields ctf_header, ctf_dynbase,
ctf_cuname, ctf_dyncuname: ctf_base and ctf_buf are no longer const.
* ctf-open.c (ctf_set_base): Preserve the gap between ctf_buf and
ctf_base: do not assume that it is always sizeof (ctf_header_t).
Print out ctf_cuname: only print out ctf_parname if set.
(ctf_free_base): Removed, ctf_base is no longer freed: free
ctf_dynbase instead.
(ctf_set_version): Fix spacing.
(upgrade_header): New, in-place header upgrading.
(upgrade_types): Rename to...
(upgrade_types_v1): ... this. Free ctf_dynbase, not ctf_base. No
longer track old and new headers separately. No longer allow for
header sizes explicitly: squeeze the headers out on upgrade (they
are preserved in fp->ctf_header). Set ctf_dynbase, ctf_base and
ctf_buf explicitly. Use ctf_free, not ctf_free_base.
(upgrade_types): New, also handle ctf_parmax updating.
(flip_header): Flip ctf_cuname.
(flip_types): Flip BUF explicitly rather than deriving BUF from
BASE.
(ctf_bufopen): Store the header in fp->ctf_header. Correct minimum
required alignment of objtoff and funcoff. No longer store it in
the ctf_buf unless that buf is derived unmodified from the input.
Set ctf_dynbase where ctf_base is dynamically allocated. Drop locals
that duplicate fields in ctf_file: move allocation of ctf_file
further up instead. Call upgrade_header as needed. Move
version-specific ctf_parmax initialization into upgrade_types. More
concise error handling.
(ctf_file_close): No longer test for null pointers before freeing.
Free ctf_dyncuname, ctf_dynbase, and ctf_header. Do not call
ctf_free_base.
(ctf_cuname): New.
(ctf_cuname_set): New.
* ctf-create.c (ctf_update): Populate ctf_cuname.
(ctf_gzwrite): Write out the header explicitly. Remove obsolescent
comment.
(ctf_write): Likewise.
(ctf_compress_write): Get the header from ctf_header, not ctf_base.
Fix the compression length: fp->ctf_size never counted the CTF
header. Simplify the compress call accordingly.
With a glibc before 2.9 (such as 2.8), there's <endian.h> but no
htole64 or le64toh, so you get, compiling binutils for any target:
libtool: link: gcc -W -Wall -Wstrict-prototypes -Wmissing-prototypes \
-Wshadow -Werror -I/x/binutils/../zlib -g -O2 -o objdump \
objdump.o dwarf.o prdbg.o rddbg.o debug.o stabs.o rdcoff.o \
bucomm.o version.o filemode.o elfcomm.o ../opcodes/.libs/libopcodes.a \
../libctf/libctf.a ../bfd/.libs/libbfd.a -L/x/obj/b/zlib -lz ../libiberty/libiberty.a -ldl
../libctf/libctf.a(ctf-archive.o): In function `ctf_archive_raw_iter_internal':
/x/src/libctf/ctf-archive.c:543: undefined reference to `le64toh'
/x/src/libctf/ctf-archive.c:550: undefined reference to `le64toh'
/x/src/libctf/ctf-archive.c:551: undefined reference to `le64toh'
/x/src/libctf/ctf-archive.c:551: undefined reference to `le64toh'
/x/src/libctf/ctf-archive.c:554: undefined reference to `le64toh'
../libctf/libctf.a(ctf-archive.o):/x/src/libctf/ctf-archive.c:545: more undefined references to `le64toh' follow
(etc)
Also, I see no bswap_identity_64 *anywhere* except in libctf/swap.h
(including current glibc) and I don't think calling an "identity"-
function is better than just plain "#define foo(x) (x)" anyway.
(Where does the idea of a bytestap.h bswap_identity_64 come from?)
Speaking of that, I should mention that I instrumented the condition
to observe that the WORDS_BIGENDIAN case passes too for a presumed
big-endian target and glibc-2.8: there is a bswap_64 present for that
version. Curiously, no test-case regressed with that instrumentation.
For the record, constructing binary blobs using text source to run
tests on, can be done by linking to --oformat binary (with most ELF
targets), but I guess that's seen as unnecessary roundabout perhaps
checking in binary files in the test-suite would be ok these days.
[...]
[nca: trimmed commit log slightly, updated changelog]
v5: fix tabdamage.
libctf/
* ctf-endian.h: Don't assume htole64 and le64toh are always
present if HAVE_ENDIAN_H; also check if htole64 is defined.
[!WORDS_BIGENDIAN] (htole64, le64toh): Define as identity,
not bswap_identity_64.
The first two of these allow you to get function type info and args out
of the types section give a type ID: astonishingly, this was missing
from libctf before now: so even though types of kind CTF_K_FUNCTION were
supported, you couldn't find out anything about them. (The existing
ctf_func_info and ctf_func_args only allow you to get info about
functions in the function section, i.e. given symbol table indexes, not
type IDs.)
The second of these allows you to get the raw undecorated name out of
the CTF section (strdupped for safety) without traversing subtypes to
build a full C identifier out of it. It's useful for things that are
already tracking the type kind etc and just need an unadorned name.
include/
* ctf-api.h (ECTF_NOTFUNC): Fix description.
(ctf_func_type_info): New.
(ctf_func_type_args): Likewise.
libctf/
* ctf-types.c (ctf_type_aname_raw): New.
(ctf_func_type_info): Likewise.
(ctf_func_type_args): Likewise.
* ctf-error.c (_ctf_errlist): Fix description.
The first ctf_snapshot called after CTF file creation yields a snapshot
handle that always yields a spurious ECTF_OVERROLLBACK error ("Attempt
to roll back past a ctf_update") on ctf_rollback(), even if ctf_update
has never been called.
The fix is to start with a ctf_snapshot value higher than the zero value
that ctf_snapshot_lu ("last update CTF snapshot value") is initialized
to.
libctf/
* ctf-create.c (ctf_create): Fix off-by-one error.
ctf.h states:
> [...] the CTF string table does not contain any duplicated strings.
Unfortunately this is entirely untrue: libctf has before now made no
attempt whatsoever to deduplicate the string table. It computes the
string table's length on the fly as it adds new strings to the dynamic
CTF file, and ctf_update() just writes each string to the table and
notes the current write position as it traverses the dynamic CTF file's
data structures and builds the final CTF buffer. There is no global
view of the strings and no deduplication.
Fix this by erasing the ctf_dtvstrlen dead-reckoning length, and adding
a new dynhash table ctf_str_atoms that maps unique strings to a list
of references to those strings: a reference is a simple uint32_t * to
some value somewhere in the under-construction CTF buffer that needs
updating to note the string offset when the strtab is laid out.
Adding a string is now a simple matter of calling ctf_str_add_ref(),
which adds a new atom to the atoms table, if one doesn't already exist,
and adding the location of the reference to this atom to the refs list
attached to the atom: this works reliably as long as one takes care to
only call ctf_str_add_ref() once the final location of the offset is
known (so you can't call it on a temporary structure and then memcpy()
that structure into place in the CTF buffer, because the ref will still
point to the old location: ctf_update() changes accordingly).
Generating the CTF string table is a matter of calling
ctf_str_write_strtab(), which counts the length and number of elements
in the atoms table using the ctf_dynhash_iter() function we just added,
populating an array of pointers into the atoms table and sorting it into
order (to help compressors), then traversing this table and emitting it,
updating the refs to each atom as we go. The only complexity here is
arranging to keep the null string at offset zero, since a lot of code in
libctf depends on being able to leave strtab references at 0 to indicate
'no name'. Once the table is constructed and the refs updated, we know
how long it is, so we can realloc() the partial CTF buffer we allocated
earlier and can copy the table on to the end of it (and purge the refs
because they're not needed any more and have been invalidated by the
realloc() call in any case).
The net effect of all this is a reduction in uncompressed strtab sizes
of about 30% (perhaps a quarter to a half of all strings across the
Linux kernel are eliminated as duplicates). Of course, duplicated
strings are highly redundant, so the space saving after compression is
only about 20%: when the other non-strtab sections are factored in, CTF
sizes shrink by about 10%.
No change in externally-visible API or file format (other than the
reduction in pointless redundancy).
libctf/
* ctf-impl.h: (struct ctf_strs_writable): New, non-const version of
struct ctf_strs.
(struct ctf_dtdef): Note that dtd_data.ctt_name is unpopulated.
(struct ctf_str_atom): New, disambiguated single string.
(struct ctf_str_atom_ref): New, points to some other location that
references this string's offset.
(struct ctf_file): New members ctf_str_atoms and ctf_str_num_refs.
Remove member ctf_dtvstrlen: we no longer track the total strlen
as we add strings.
(ctf_str_create_atoms): Declare new function in ctf-string.c.
(ctf_str_free_atoms): Likewise.
(ctf_str_add): Likewise.
(ctf_str_add_ref): Likewise.
(ctf_str_purge_refs): Likewise.
(ctf_str_write_strtab): Likewise.
(ctf_realloc): Declare new function in ctf-util.c.
* ctf-open.c (ctf_bufopen): Create the atoms table.
(ctf_file_close): Destroy it.
* ctf-create.c (ctf_update): Copy-and-free it on update. No longer
special-case the position of the parname string. Construct the
strtab by calling ctf_str_add_ref and ctf_str_write_strtab after the
rest of each buffer element is constructed, not via open-coding:
realloc the CTF buffer and append the strtab to it. No longer
maintain ctf_dtvstrlen. Sort the variable entry table later, after
strtab construction.
(ctf_copy_membnames): Remove: integrated into ctf_copy_{s,l,e}members.
(ctf_copy_smembers): Drop the string offset: call ctf_str_add_ref
after buffer element construction instead.
(ctf_copy_lmembers): Likewise.
(ctf_copy_emembers): Likewise.
(ctf_create): No longer maintain the ctf_dtvstrlen.
(ctf_dtd_delete): Likewise.
(ctf_dvd_delete): Likewise.
(ctf_add_generic): Likewise.
(ctf_add_enumerator): Likewise.
(ctf_add_member_offset): Likewise.
(ctf_add_variable): Likewise.
(membadd): Likewise.
* ctf-util.c (ctf_realloc): New, wrapper around realloc that aborts
if there are active ctf_str_num_refs.
(ctf_strraw): Move to ctf-string.c.
(ctf_strptr): Likewise.
* ctf-string.c: New file, strtab manipulation.
* Makefile.am (libctf_a_SOURCES): Add it.
* Makefile.in: Regenerate.
There are two, ctf_dynhash_iter and ctf_dynhash_iter_remove: the latter
lets you return a nonzero value to remove the element being iterated
over.
Used in the next commit.
libctf/
* ctf-impl.h (ctf_hash_iter_f): New.
(ctf_dynhash_iter): New declaration.
(ctf_dynhash_iter_remove): New declaration.
* ctf-hash.c (ctf_dynhash_iter): Define.
(ctf_dynhash_iter_remove): Likewise.
(ctf_hashtab_traverse): New.
(ctf_hashtab_traverse_remove): Likewise.
(struct ctf_traverse_cb_arg): Likewise.
(struct ctf_traverse_remove_cb_arg): Likewise.
We must call htab_remove_elt with an element (in this case, a mocked-up
one with only the key populated, since no reasonable hash function will
need the other fields), not with the key alone.
libctf/
* ctf-hash.c (ctf_dynhash_remove): Call with a mocked-up element.
We were sometimes printing hex values without prefixing them with '0x',
leading to confusion about what base the numbers were actually in.
libctf/
* ctf-dump.c (ctf_dump_format_type): Prefix hex strings with 0x.
(ctf_dump_funcs): Likewise.
ctf_open (or, rather, ctf_fdopen, which underlies it) has several
endianness problems, even though it was written after the
endian-swapping code was implemented, so should have been endian-aware.
Even though the comment right above the relevant check says that it wil
check for CTF magic in any endianness, it only checks in the native
endianness, so opening raw LE CTF files on BE, or vice-versa, will fail.
It also checks the CTF version by hand, without ever endianness-swapping
the header, so that too will fail, and is entirely redundant because
ctf_simple_open does the job properly in any case. We have a similar
problem in the next if block, which checks for raw CTF archives: we are
checking in the native endianness while we should be doing a le64toh()
on it to check in little-endian form only: so opening CTF archives
created on the local machine will fail if the local machine is
big-endian.
Adding insult to injury, if ctf_simple_open then fails, we go on and try
to turn it into a single-element CTF archive regardless, throwing the
error away. Since this involves dereferencing null pointers it is not
likely to work very well.
libctf/
* ctf-open-bfd.c: Add swap.h and ctf-endian.h.
(ctf_fdopen): Check for endian-swapped raw CTF magic, and
little-endian CTF archive magic. Do not check the CTF version:
ctf_simple_open does that in endian-safe ways. Do not dereference
null pointers on open failure.
Testing of the first code to generate CTF_K_SLICEs on big-endian
revealed a bunch of new problems in this area. Most importantly, the
trick we did earlier to avoid wasting two bytes on padding in the
ctf_slice_t is best avoided: because it leads to the whole file after
that point no longer being naturally aligned, all multibyte accesses
from then on must use memmove() to avoid unaligned access on platforms
where that is fatal. In future, this is planned, but for now we are
still doing direct access in many places, so we must revert to making
ctf_slice_t properly aligned for storage in an array.
Rather than wasting bytes on padding, we boost the size of cts_offset
and cts_bits. This is still a waste of space (we cannot have offsets or
bits in bitfields > 256) but it cannot be avoided for now, and slices
are not so common that this will be a serious problem.
A possibly-worse endianness problem fixed at the same time involves
a codepath used only for foreign-endian, uncompressed CTF files, where
we were not copying the actual CTF data into the buffer, leading to
libctf reading only zeroes (or, possibly, uninitialized garbage).
Finally, when we read in a CTF file, we copy the header and work from
the copy. We were flipping the endianness of the header copy, and of
the body of the file buffer, but not of the header in the file buffer
itself: so if we write the file back out again we end up with an
unreadable frankenfile with header and body of different endiannesses.
Fix by flipping both copies of the header.
include/
* ctf.h (ctf_slice_t): Make cts_offset and cts_bits unsigned
short, so following structures are properly aligned.
libctf/
* ctf-open.c (get_vbytes_common): Return the new slice size.
(ctf_bufopen): Flip the endianness of the CTF-section header copy.
Remember to copy in the CTF data when opening an uncompressed
foreign-endian CTF file. Prune useless variable manipulation.
If we see a CTF type with a kind we do not recognize in its ctt_info
during opening, we cannot skip it and continue opening the file: if the
type kind is unknown, we do not know how long its vlen is, and we cannot
have skipped past it: so if we continue reading we will almost certainly
read in part of the vlen as if it were a new ctf_type_t.
Avoid this trouble by considering unknown type kinds to be a reason to
return ECTF_CORRUPT, just like everything else that reads in type kinds
does.
libctf/
* ctf-open.c (ctf_types): Fail when unidentified type kinds are
seen.
This is an essential first piece of info needed to debug both libctf
writing and reading problems, and we weren't recording it anywhere!
(This is a short-term fix: fairly soon, we will record all of this in a
form that outlives ctf_bufopen, and then ctf_dump() will be able to dump
it like it can everything else.)
libctf/
* ctf-open.c (ctf_bufopen): Dump header offsets into the debugging
output.
This allocator has the ostensible benefit that it lets us mprotect() the
memory used for CTF storage: but in exchange for this it adds
considerable complexity, since we have to track allocation sizes
ourselves for use at freeing time, note whether the data we are storing
was ctf_data_alloc()ed or not so we know if we can safely mprotect()
it... and while the mprotect()ing has found few bugs, it *has* been the
cause of more than one due to errors in all this tracking leading to us
mprotect()ing bits of the heap and stuff like that.
We are about to start composing CTF buffers from pieces so that we can
do usage-based optimizations on the strtab. This means we need
realloc(), which needs nonportable mremap() and *more* tracking of the
*original* allocation size, and the complexity and bureaucracy of all of
this is just too high for its negligible benefits.
Drop the whole thing and just use malloc() like everyone else. It knows
better than we do when it is safe to use mmap() under the covers,
anyway.
While we're at it, don't leak the entire buffer if ctf_compress_write()
fails to compress it.
libctf/
* ctf-subr.c (_PAGESIZE): Remove.
(ctf_data_alloc): Likewise.
(ctf_data_free): Likewise.
(ctf_data_protect): Likewise.
* ctf-impl.h: Remove declarations.
* ctf-create.c (ctf_update): No longer call ctf_data_protect: use
ctf_free, not ctf_data_free.
(ctf_compress_write): Use ctf_data_alloc, not ctf_alloc. Free
the buffer again on compression error.
* ctf-open.c (ctf_set_base): No longer track the size: call
ctf_free, not ctf_data_free.
(upgrade_types): Likewise. Call ctf_alloc, not ctf_data_alloc.
(ctf_bufopen): Likewise. No longer call ctf_data_protect.
We were missing several cases where dynhash insertion might fail, likely
due to OOM but possibly for other reasons. Pass the errors on.
libctf/
* ctf-create.c (ctf_dtd_insert): Pass on error returns from
ctf_dynhash_insert.
(ctf_dvd_insert): Likewise.
(ctf_add_generic): Likewise.
(ctf_add_variable): Likewise.
* ctf-impl.h: Adjust declarations.
Not all platforms have it. Use libiberty xstrndup() instead.
(The include of libiberty.h happens in an unusual place due to the
requirements of synchronization of most source files between this
project and another that does not use libiberty. It serves to pull
libiberty.h in for all source files in libctf/, which does the trick.)
Tested on x86_64-pc-linux-gnu, x86_64-unknown-freebsd12.0,
sparc-sun-solaris2.11, i686-pc-cygwin, i686-w64-mingw32.
libctf/
* ctf-decls.h: Include <libiberty.h>.
* ctf-lookup.c (ctf_lookup_by_name): Call xstrndup(), not strndup().
Unsigned long will always be adequate (the only cases involving an
ssize_t are cases in which no error can be generated, or in which
negative output would require a seriously corrupted file: the latter has
been rewritten on a branch in any case).
Tested on x86_64-pc-linux-gnu, x86_64-unknown-freebsd12.0,
sparc-sun-solaris2.11, i686-pc-cygwin, i686-w64-mingw32.
libctf/
* ctf-dump.c (ctf_dump_format_type): Cast size_t's used in printf()s.
(ctf_dump_objts): Likewise.
(ctf_dump_funcs): Likewise.
(ctf_dump_member): Likewise.
(ctf_dump_str): Likewise.
Tested on x86_64-pc-linux-gnu, x86_64-unknown-freebsd12.0,
sparc-sun-solaris2.11, i686-pc-cygwin, i686-w64-mingw32.
libctf/
* ctf-archive.c (arc_mmap_header): Mark fd as potentially unused.
* ctf-subr.c (ctf_data_protect): Mark both args as potentially unused.
Too many platforms don't support it, and we can always safely use %lu or
%li anyway, because the only uses are in debugging output.
libctf/
* ctf-archive.c (ctf_arc_write): Eschew %zi format specifier.
(ctf_arc_open_by_offset): Likewise.
* ctf-create.c (ctf_add_type): Likewise.
On x86-64 Fedora 29, I tried to build a mingw-hosted gdb that targets
ppc-linux. You can do this with:
../binutils-gdb/configure --host=i686-w64-mingw32 --target=ppc-linux \
--disable-{binutils,gas,gold,gprof,ld}
The build failed with these errors in libctf:
In file included from ../../binutils-gdb/libctf/ctf-create.c:20:
../../binutils-gdb/libctf/ctf-create.c: In function 'ctf_add_encoded':
../../binutils-gdb/libctf/ctf-create.c:803:59: error: 'NBBY' undeclared (first use in this function)
dtd->dtd_data.ctt_size = clp2 (P2ROUNDUP (ep->cte_bits, NBBY) / NBBY);
^~~~
../../binutils-gdb/libctf/ctf-impl.h:254:42: note: in definition of macro 'P2ROUNDUP'
#define P2ROUNDUP(x, align) (-(-(x) & -(align)))
^~~~~
../../binutils-gdb/libctf/ctf-create.c:803:59: note: each undeclared identifier is reported only once for each function it appears in
dtd->dtd_data.ctt_size = clp2 (P2ROUNDUP (ep->cte_bits, NBBY) / NBBY);
^~~~
../../binutils-gdb/libctf/ctf-impl.h:254:42: note: in definition of macro 'P2ROUNDUP'
#define P2ROUNDUP(x, align) (-(-(x) & -(align)))
^~~~~
../../binutils-gdb/libctf/ctf-create.c: In function 'ctf_add_slice':
../../binutils-gdb/libctf/ctf-create.c:862:59: error: 'NBBY' undeclared (first use in this function)
dtd->dtd_data.ctt_size = clp2 (P2ROUNDUP (ep->cte_bits, NBBY) / NBBY);
^~~~
../../binutils-gdb/libctf/ctf-impl.h:254:42: note: in definition of macro 'P2ROUNDUP'
#define P2ROUNDUP(x, align) (-(-(x) & -(align)))
^~~~~
../../binutils-gdb/libctf/ctf-create.c: In function 'ctf_add_member_offset':
../../binutils-gdb/libctf/ctf-create.c:1341:21: error: 'NBBY' undeclared (first use in this function)
off += lsize * NBBY;
^~~~
../../binutils-gdb/libctf/ctf-create.c: In function 'ctf_add_type':
../../binutils-gdb/libctf/ctf-create.c:1822:16: warning: unknown conversion type character 'z' in format [-Wformat=]
ctf_dprintf ("Conflict for type %s against ID %lx: "
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../../binutils-gdb/libctf/ctf-create.c:1823:35: note: format string is defined here
"union size differs, old %zi, new %zi\n",
^
../../binutils-gdb/libctf/ctf-create.c:1822:16: warning: unknown conversion type character 'z' in format [-Wformat=]
ctf_dprintf ("Conflict for type %s against ID %lx: "
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../../binutils-gdb/libctf/ctf-create.c:1823:44: note: format string is defined here
"union size differs, old %zi, new %zi\n",
^
../../binutils-gdb/libctf/ctf-create.c:1822:16: warning: too many arguments for format [-Wformat-extra-args]
ctf_dprintf ("Conflict for type %s against ID %lx: "
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This patch fixes the actual errors in here. I did not try to fix the
printf warnings, though I think someone ought to.
Ok?
libctf/ChangeLog
2019-06-04 Tom Tromey <tromey@adacore.com>
* ctf-create.c (ctf_add_encoded, ctf_add_slice)
(ctf_add_member_offset): Use CHAR_BIT, not NBBY.
(Not tested on any such platforms, since I don't have access to any at
the moment. Testing encouraged.)
libctf/
* configure.ac: Check for O_CLOEXEC.
* ctf-decls.h (O_CLOEXEC): Define (to 0), if need be.
* config.h.in: Regenerate.
We cannot just look for any declaration of qsort_r, because some
operating systems have a qsort_r that has a different prototype
but which still has a pair of pointers in the right places (the last two
args are interchanged): so use AC_LINK_IFELSE to check for both
known variants of qsort_r(), and swap their args into a consistent order
in a suitable inline function. (The code for this is taken almost
unchanged from gnulib.)
(Now we are not using AC_LIBOBJ any more, we can use a better name for
the qsort_r replacement as well.)
libctf/
* qsort_r.c: Rename to...
* ctf-qsort_r.c: ... this.
(_quicksort): Define to ctf_qsort_r.
* ctf-decls.h (qsort_r): Remove.
(ctf_qsort_r): Add.
(struct ctf_qsort_arg): New, transport the real ARG and COMPAR.
(ctf_qsort_compar_thunk): Rearrange the arguments to COMPAR.
* Makefile.am (libctf_a_LIBADD): Remove.
(libctf_a_SOURCES): New, add ctf-qsort_r.c.
* ctf-archive.c (ctf_arc_write): Call ctf_qsort_r, not qsort_r.
* ctf-create.c (ctf_update): Likewise.
* configure.ac: Check for BSD versus GNU qsort_r signature.
* Makefile.in: Regenerate.
* config.h.in: Likewise.
* configure: Likewise.
- Use of nonportable <endian.h>
- Use of qsort_r
- Use of zlib without appropriate magic to pull in the binutils zlib
- Use of off64_t without checking (fixed by dropping the unused fields
that need off64_t entirely)
- signedness problems due to long being too short a type on 32-bit
platforms: ctf_id_t is now 'unsigned long', and CTF_ERR must be
used only for functions that return ctf_id_t
- One lingering use of bzero() and of <sys/errno.h>
All fixed, using code from gnulib where possible.
Relatedly, set cts_size in a couple of places it was missed
(string table and symbol table loading upon ctf_bfdopen()).
binutils/
* objdump.c (make_ctfsect): Drop cts_type, cts_flags, and
cts_offset.
* readelf.c (shdr_to_ctf_sect): Likewise.
include/
* ctf-api.h (ctf_sect_t): Drop cts_type, cts_flags, and cts_offset.
(ctf_id_t): This is now an unsigned type.
(CTF_ERR): Cast it to ctf_id_t. Note that it should only be used
for ctf_id_t-returning functions.
libctf/
* Makefile.am (ZLIB): New.
(ZLIBINC): Likewise.
(AM_CFLAGS): Use them.
(libctf_a_LIBADD): New, for LIBOBJS.
* configure.ac: Check for zlib, endian.h, and qsort_r.
* ctf-endian.h: New, providing htole64 and le64toh.
* swap.h: Code style fixes.
(bswap_identity_64): New.
* qsort_r.c: New, from gnulib (with one added #include).
* ctf-decls.h: New, providing a conditional qsort_r declaration,
and unconditional definitions of MIN and MAX.
* ctf-impl.h: Use it. Do not use <sys/errno.h>.
(ctf_set_errno): Now returns unsigned long.
* ctf-util.c (ctf_set_errno): Adjust here too.
* ctf-archive.c: Use ctf-endian.h.
(ctf_arc_open_by_offset): Use memset, not bzero. Drop cts_type,
cts_flags and cts_offset.
(ctf_arc_write): Drop debugging dependent on the size of off_t.
* ctf-create.c: Provide a definition of roundup if not defined.
(ctf_create): Drop cts_type, cts_flags and cts_offset.
(ctf_add_reftype): Do not check if type IDs are below zero.
(ctf_add_slice): Likewise.
(ctf_add_typedef): Likewise.
(ctf_add_member_offset): Cast error-returning ssize_t's to size_t
when known error-free. Drop CTF_ERR usage for functions returning
int.
(ctf_add_member_encoded): Drop CTF_ERR usage for functions returning
int.
(ctf_add_variable): Likewise.
(enumcmp): Likewise.
(enumadd): Likewise.
(membcmp): Likewise.
(ctf_add_type): Likewise. Cast error-returning ssize_t's to size_t
when known error-free.
* ctf-dump.c (ctf_is_slice): Drop CTF_ERR usage for functions
returning int: use CTF_ERR for functions returning ctf_type_id.
(ctf_dump_label): Likewise.
(ctf_dump_objts): Likewise.
* ctf-labels.c (ctf_label_topmost): Likewise.
(ctf_label_iter): Likewise.
(ctf_label_info): Likewise.
* ctf-lookup.c (ctf_func_args): Likewise.
* ctf-open.c (upgrade_types): Cast to size_t where appropriate.
(ctf_bufopen): Likewise. Use zlib types as needed.
* ctf-types.c (ctf_member_iter): Drop CTF_ERR usage for functions
returning int.
(ctf_enum_iter): Likewise.
(ctf_type_size): Likewise.
(ctf_type_align): Likewise. Cast to size_t where appropriate.
(ctf_type_kind_unsliced): Likewise.
(ctf_type_kind): Likewise.
(ctf_type_encoding): Likewise.
(ctf_member_info): Likewise.
(ctf_array_info): Likewise.
(ctf_enum_value): Likewise.
(ctf_type_rvisit): Likewise.
* ctf-open-bfd.c (ctf_bfdopen): Drop cts_type, cts_flags and
cts_offset.
(ctf_simple_open): Likewise.
(ctf_bfdopen_ctfsect): Likewise. Set cts_size properly.
* Makefile.in: Regenerate.
* aclocal.m4: Likewise.
* config.h: Likewise.
* configure: Likewise.
All machinery works as on ELF, except for automatic loading of ELF
string and symbol tables in the BFD-style open machinery.
* Makefile.def (dependencies): configure-libctf depends on all-bfd
and all its deps.
* Makefile.in: Regenerated.
libctf/
* configure.in: Check for bfd_section_from_elf_index.
* configure: Regenerate.
* config.h.in [HAVE_BFD_ELF]: Likewise.
* libctf/ctf_open_bfd (ctf_bfdopen_ctfsect): Use it.
abfd is potentially unused now.
This ties libctf into the build system, and makes binutils depend on it
(used by the next commits).
* Makefile.def (host_modules): Add libctf.
* Makefile.def (dependencies): Likewise.
libctf depends on zlib, libiberty, and bfd.
* Makefile.in: Regenerated.
* configure.ac (host_libs): Add libctf.
* configure: Regenerated.
libctf/
* Makefile.am: New.
* Makefile.in: Regenerated.
* config.h.in: Likewise.
* aclocal.m4: Likewise.
* configure: Likewise.
This introduces ctf_dump(), an iterator which returns a series of
strings, each representing a debugging dump of one item from a given
section in the CTF file. The items may be multiline: a callback is
provided to allow the caller to decorate each line as they desire before
the line is returned.
libctf/
* ctf-dump.c: New.
include/
* ctf-api.h (ctf_dump_decorate_f): New.
(ctf_dump_state_t): new.
(ctf_dump): New.
This facility allows you to associate regions of type IDs with *labels*,
a labelled tiling of the type ID space. You can use these to define
CTF containers with distinct parents for distinct ranges of the ID
space, or to assist with parallelization of CTF processing, or for any
other purpose you can think of.
Notably absent from here (though declared in the API header) is any way
to define new labels: this will probably be introduced soon, as part of
the linker deduplication work. (One existed in the past, but was deeply
tied to the Solaris CTF file generator and had to be torn out.)
libctf/
* ctf-labels.c: New.
include/
* ctf-api.h (ctf_label_f): New.
(ctf_label_set): New.
(ctf_label_get): New.
(ctf_label_topmost): New.
(ctf_label_info): New.
(ctf_label_iter): New.
This old Solaris standard allows callers to specify that they are
expecting one particular API and/or CTF file format from the library.
libctf/
* ctf-impl.h (_libctf_version): New declaration.
* ctf-subr.c (_libctf_version): Define it.
(ctf_version): New.
include/
* ctf-api.h (ctf_version): New.
ctf_add_type() allows you to copy types, and all the types they depend
on, from one container to another (writable) container. This lets a
program maintaining multiple distinct containers (not in a parent-child
relationship) introduce types that depend on types in one container in
another writable one, by copying the necessary types.
libctf/
* ctf-create.c (enumcmp): New.
(enumadd): Likewise.
(membcmp): Likewise.
(membadd): Likewise.
(ctf_add_type): Likewise.
These functions allow you to look up types given a name in a simple
subset of C declarator syntax (no function pointers), to look up the
types of variables given a name, and to look up the types of data
objects and the type signatures of functions given symbol table offsets.
(Despite its name, one function in this commit, ctf_lookup_symbol_name(),
is for the internal use of libctf only, and does not appear in any
public header files.)
libctf/
* ctf-lookup.c (isqualifier): New.
(ctf_lookup_by_name): Likewise.
(struct ctf_lookup_var_key): Likewise.
(ctf_lookup_var): Likewise.
(ctf_lookup_variable): Likewise.
(ctf_lookup_symbol_name): Likewise.
(ctf_lookup_by_symbol): Likewise.
(ctf_func_info): Likewise.
(ctf_func_args): Likewise.
include/
* ctf-api.h (ctf_func_info): New.
(ctf_func_args): Likewise.
(ctf_lookup_by_symbol): Likewise.
(ctf_lookup_by_symbol): Likewise.
(ctf_lookup_variable): Likewise.
Finally we get to the functions used to actually look up and enumerate
properties of types in a container (names, sizes, members, what type a
pointer or cv-qual references, determination of whether two types are
assignment-compatible, etc).
With a very few exceptions these do not work for types newly added via
ctf_add_*(): they only work on types in read-only containers, or types
added before the most recent call to ctf_update().
This also adds support for lookup of "variables" (string -> type ID
mappings) and for generation of C type names corresponding to a type ID.
libctf/
* ctf-decl.c: New file.
* ctf-types.c: Likewise.
* ctf-impl.h: New declarations.
include/
* ctf-api.h (ctf_visit_f): New definition.
(ctf_member_f): Likewise.
(ctf_enum_f): Likewise.
(ctf_variable_f): Likewise.
(ctf_type_f): Likewise.
(ctf_type_isparent): Likewise.
(ctf_type_ischild): Likewise.
(ctf_type_resolve): Likewise.
(ctf_type_aname): Likewise.
(ctf_type_lname): Likewise.
(ctf_type_name): Likewise.
(ctf_type_sizee): Likewise.
(ctf_type_align): Likewise.
(ctf_type_kind): Likewise.
(ctf_type_reference): Likewise.
(ctf_type_pointer): Likewise.
(ctf_type_encoding): Likewise.
(ctf_type_visit): Likewise.
(ctf_type_cmp): Likewise.
(ctf_type_compat): Likewise.
(ctf_member_info): Likewise.
(ctf_array_info): Likewise.
(ctf_enum_name): Likewise.
(ctf_enum_value): Likewise.
(ctf_member_iter): Likewise.
(ctf_enum_iter): Likewise.
(ctf_type_iter): Likewise.
(ctf_variable_iter): Likewise.
These functions let you open an ELF file with a customarily-named CTF
section in it, automatically opening the CTF file or archive and
associating the symbol and string tables in the ELF file with the CTF
container, so that you can look up the types of symbols in the ELF file
via ctf_lookup_by_symbol(), and so that strings can be shared between
the ELF file and CTF container, to save space.
It uses BFD machinery to do so. This has now been lightly tested and
seems to work. In particular, if you already have a bfd you can pass
it in to ctf_bfdopen(), and if you want a bfd made for you you can
call ctf_open() or ctf_fdopen(), optionally specifying a target (or
try once without a target and then again with one if you get
ECTF_BFD_AMBIGUOUS back).
We use a forward declaration for the struct bfd in ctf-api.h, so that
ctf-api.h users are not required to pull in <bfd.h>. (This is mostly
for the sake of readelf.)
libctf/
* ctf-open-bfd.c: New file.
* ctf-open.c (ctf_close): New.
* ctf-impl.h: Include bfd.h.
(ctf_file): New members ctf_data_mmapped, ctf_data_mmapped_len.
(ctf_archive_internal): New members ctfi_abfd, ctfi_data,
ctfi_bfd_close.
(ctf_bfdopen_ctfsect): New declaration.
(_CTF_SECTION): likewise.
include/
* ctf-api.h (struct bfd): New forward.
(ctf_fdopen): New.
(ctf_bfdopen): Likewise.
(ctf_open): Likewise.
(ctf_arc_open): Likewise.
If you need to store a large number of CTF containers somewhere, this
provides a dedicated facility for doing so: an mmappable archive format
like a very simple tar or ar without all the system-dependent format
horrors or need for heavy file copying, with built-in compression of
files above a particular size threshold.
libctf automatically mmap()s uncompressed elements of these archives, or
uncompresses them, as needed. (If the platform does not support mmap(),
copying into dynamically-allocated buffers is used.)
Archive iteration operations are partitioned into raw and non-raw
forms. Raw operations pass thhe raw archive contents to the callback:
non-raw forms open each member with ctf_bufopen() and pass the resulting
ctf_file_t to the iterator instead. This lets you manipulate the raw
data in the archive, or the contents interpreted as a CTF file, as
needed.
It is not yet known whether we will store CTF archives in a linked ELF
object in one of these (akin to debugdata) or whether they'll get one
section per TU plus one parent container for types shared between them.
(In the case of ELF objects with very large numbers of TUs, an archive
of all of them would seem preferable, so we might just use an archive,
and add lzma support so you can assume that .gnu_debugdata and .ctf are
compressed using the same algorithm if both are present.)
To make usage easier, the ctf_archive_t is not the on-disk
representation but an abstraction over both ctf_file_t's and archives of
many ctf_file_t's: users see both CTF archives and raw CTF files as
ctf_archive_t's upon opening, the only difference being that a raw CTF
file has only a single "archive member", named ".ctf" (the default if a
null pointer is passed in as the name). The next commit will make use
of this facility, in addition to providing the public interface to
actually open archives. (In the future, it should be possible to have
all CTF sections in an ELF file appear as an "archive" in the same
fashion.)
This machinery is also used to allow library-internal creators of
ctf_archive_t's (such as the next commit) to stash away an ELF string
and symbol table, so that all opens of members in a given archive will
use them. This lets CTF archives exploit the ELF string and symbol
table just like raw CTF files can.
(All this leads to somewhat confusing type naming. The ctf_archive_t is
a typedef for the opaque internal type, struct ctf_archive_internal: the
non-internal "struct ctf_archive" is the on-disk structure meant for
other libraries manipulating CTF files. It is probably clearest to use
the struct name for struct ctf_archive_internal inside the program, and
the typedef names outside.)
libctf/
* ctf-archive.c: New.
* ctf-impl.h (ctf_archive_internal): New type.
(ctf_arc_open_internal): New declaration.
(ctf_arc_bufopen): Likewise.
(ctf_arc_close_internal): Likewise.
include/
* ctf.h (CTFA_MAGIC): New.
(struct ctf_archive): New.
(struct ctf_archive_modent): Likewise.
* ctf-api.h (ctf_archive_member_f): New.
(ctf_archive_raw_member_f): Likewise.
(ctf_arc_write): Likewise.
(ctf_arc_close): Likewise.
(ctf_arc_open_by_name): Likewise.
(ctf_archive_iter): Likewise.
(ctf_archive_raw_iter): Likewise.
(ctf_get_arc): Likewise.
This fills in the other half of the opening/creation puzzle: opening of
already-existing CTF files. Such files are always read-only: if you
want to add to a CTF file opened with one of the opening functions in
this file, use ctf_add_type(), in a later commit, to copy appropriate
types into a newly ctf_create()d, writable container.
The lowest-level opening functions are in here: ctf_bufopen(), which
takes ctf_sect_t structures akin to ELF section headers, and
ctf_simple_open(), which can be used if you don't have an entire ELF
section header to work from. Both will malloc() new space for the
buffers only if necessary, will mmap() directly from the file if
requested, and will mprotect() it afterwards to prevent accidental
corruption of the types. These functions are also used by ctf_update()
when converting types in a writable container into read-only types that
can be looked up using the lookup functions (in later commits).
The files are always of the native endianness of the system that created
them: at read time, the endianness of the header magic number is used to
determine whether or not the file needs byte-swapping, and the entire
thing is aggressively byte-swapped.
The agggressive nature of this swapping avoids complicating the rest of
the code with endianness conversions, while the native endianness
introduces no byte-swapping overhead in the common case. (The
endianness-independence code is also much newer than everything else in
this file, and deserves closer scrutiny.)
The accessors at the top of the file are there to transparently support
older versions of the CTF file format, allowing translation from older
formats that have different sizes for the structures in ctf.h:
currently, these older formats are intermingled with the newer ones in
ctf.h: they will probably migrate to a compatibility header in time, to
ease readability. The ctf_set_base() function is split out for the same
reason: when conversion code to a newer format is written, it would need
to malloc() new storage for the entire ctf_file_t if a file format
change causes it to grow, and for that we need ctf_set_base() to be a
separate function.
One pair of linked data structures supported by this file has no
creation code in libctf yet: the data and function object sections read
by init_symtab(). These will probably arrive soon, when the linker comes
to need them. (init_symtab() has hardly been changed since 2009, but if
any code in libctf has rotted over time, this will.)
A few simple accessors are also present that can even be called on
read-only containers because they don't actually modify them, since the
relevant things are not stored in the container but merely change its
operation: ctf_setmodel(), which lets you specify whether a container is
LP64 or not (used to statically determine the sizes of a few types),
ctf_import(), which is the only way to associate a parent container with
a child container, and ctf_setspecific(), which lets the caller
associate an arbitrary pointer with the CTF container for any use. If
the user doesn't call these functions correctly, libctf will misbehave:
this is particularly important for ctf_import(), since a container built
against a given parent container will not be able to resolve types that
depend on types in the parent unless it is ctf_import()ed with a parent
container with the same set of types at the same IDs, or a superset.
Possible future extensions (also noted in the ctf-hash.c file) include
storing a count of things so that we don't need to do one pass over the
CTF file counting everything, and computing a perfect hash at CTF
creation time in some compact form, storing it in the CTF file, and
using it to hash things so we don't need to do a second pass over the
entire CTF file to set up the hashes used to go from names to type IDs.
(There are multiple such hashes, one for each C type namespace: types,
enums, structs, and unions.)
libctf/
* ctf-open.c: New file.
* swap.h: Likewise.
include/
* ctf-api.h (ctf_file_close): New declaration.
(ctf_getdatasect): Likewise.
(ctf_parent_file): Likewise.
(ctf_parent_name): Likewise.
(ctf_parent_name_set): Likewise.
(ctf_import): Likewise.
(ctf_setmodel): Likewise.
(ctf_getmodel): Likewise.
(ctf_setspecific): Likewise.
(ctf_getspecific): Likewise.
The CTF creation process looks roughly like (error handling elided):
int err;
ctf_file_t *foo = ctf_create (&err);
ctf_id_t type = ctf_add_THING (foo, ...);
ctf_update (foo);
ctf_*write (...);
Some ctf_add_THING functions accept other type IDs as arguments,
depending on the type: cv-quals, pointers, and structure and union
members all take other types as arguments. So do 'slices', which
let you take an existing integral type and recast it as a type
with a different bitness or offset within a byte, for bitfields.
One class of THING is not a type: "variables", which are mappings
of names (in the internal string table) to types. These are mostly
useful when encoding variables that do not appear in a symbol table
but which some external user has some other way to figure out the
address of at runtime (dynamic symbol lookup or querying a VM
interpreter or something).
You can snapshot the creation process at any point: rolling back to a
snapshot deletes all types and variables added since that point.
You can make arbitrary type queries on the CTF container during the
creation process, but you must call ctf_update() first, which
translates the growing dynamic container into a static one (this uses
the CTF opening machinery, added in a later commit), which is quite
expensive. This function must also be called after adding types
and before writing the container out.
Because addition of types involves looking up existing types, we add a
little of the type lookup machinery here, as well: only enough to
look up types in dynamic containers under construction.
libctf/
* ctf-create.c: New file.
* ctf-lookup.c: New file.
include/
* ctf-api.h (zlib.h): New include.
(ctf_sect_t): New.
(ctf_sect_names_t): Likewise.
(ctf_encoding_t): Likewise.
(ctf_membinfo_t): Likewise.
(ctf_arinfo_t): Likewise.
(ctf_funcinfo_t): Likewise.
(ctf_lblinfo_t): Likewise.
(ctf_snapshot_id_t): Likewise.
(CTF_FUNC_VARARG): Likewise.
(ctf_simple_open): Likewise.
(ctf_bufopen): Likewise.
(ctf_create): Likewise.
(ctf_add_array): Likewise.
(ctf_add_const): Likewise.
(ctf_add_enum_encoded): Likewise.
(ctf_add_enum): Likewise.
(ctf_add_float): Likewise.
(ctf_add_forward): Likewise.
(ctf_add_function): Likewise.
(ctf_add_integer): Likewise.
(ctf_add_slice): Likewise.
(ctf_add_pointer): Likewise.
(ctf_add_type): Likewise.
(ctf_add_typedef): Likewise.
(ctf_add_restrict): Likewise.
(ctf_add_struct): Likewise.
(ctf_add_union): Likewise.
(ctf_add_struct_sized): Likewise.
(ctf_add_union_sized): Likewise.
(ctf_add_volatile): Likewise.
(ctf_add_enumerator): Likewise.
(ctf_add_member): Likewise.
(ctf_add_member_offset): Likewise.
(ctf_add_member_encoded): Likewise.
(ctf_add_variable): Likewise.
(ctf_set_array): Likewise.
(ctf_update): Likewise.
(ctf_snapshot): Likewise.
(ctf_rollback): Likewise.
(ctf_discard): Likewise.
(ctf_write): Likewise.
(ctf_gzwrite): Likewise.
(ctf_compress_write): Likewise.
We now enter a series of commits that are sufficiently tangled that
avoiding forward definitions is almost impossible: no attempt is made to
make individual commits compilable (which is why the build system does
not reference any of them yet): the only important thing is that they
should form something like conceptual groups.
But first, some definitions, including the core ctf_file_t itself. Uses
of these definitions will be introduced in later commits.
libctf/
* ctf-impl.h: New definitions and declarations for type creation
and lookup.
libctf maintains two distinct hash ADTs, one (ctf_dynhash) for wrapping
dynamically-generated unknown-sized hashes during CTF file construction,
one (ctf_hash) for wrapping unchanging hashes whose size is known at
creation time for reading CTF files that were previously created.
In the binutils implementation, these are both fairly thin wrappers
around libiberty hashtab.
Unusually, this code is not kept synchronized with libdtrace-ctf,
due to its dependence on libiberty hashtab.
libctf/
* ctf-hash.c: New file.
* ctf-impl.h: New declarations.
CTF functions return zero on success or an extended errno value which
can be translated into a string via the functions in this commit.
The errno numbers start at -CTF_BASE.
libctf/
* ctf-error.c: New file.
include/
* ctf-api.h (ctf_errno): New declaration.
(ctf_errmsg): Likewise.
These utilities are a bit of a ragbag of small things needed by more
than one TU: list manipulation, ELF32->64 translators, routines to look
up strings in string tables, dynamically-allocated string appenders, and
routines to set the specialized errno values previously committed in
<ctf-api.h>.
We do still need to dig around in raw ELF symbol tables in places,
because libctf allows the caller to pass in the contents of string and
symbol sections without telling it where they come from, so we cannot
use BFD to get the symbols (BFD reasonably demands the entire file). So
extract minimal ELF definitions from glibc into a private header named
libctf/elf.h: later, we use those to get symbols. (The start-of-
copyright range on elf.h reflects this glibc heritage.)
libctf/
* ctf-util.c: New file.
* elf.h: Likewise.
* ctf-impl.h: Include it, and add declarations.
The memory-allocation wrappers are simple things to allow malloc
interposition: they are only used inconsistently at present, usually
where malloc debugging was required in the past.
These provide a default implementation that is environment-variable
triggered (initialized on the first call to the libctf creation and
file-opening functions, the first functions people will use), and
a ctf_setdebug()/ctf_getdebug() pair that allows the caller to
explicitly turn debugging off and on. If ctf_setdebug() is called,
the automatic setting from an environment variable is skipped.
libctf/
* ctf-impl.h: New file.
* ctf-subr.c: New file.
include/
* ctf-api.h (ctf_setdebug): New.
(ctf_getdebug): Likewise.