--- /dev/null
+*.rej
+*.orig
+*~
+/*.patch
+
+*.o
+*.o[ls]
+
+/src/libaio.a
+/src/libaio.so*
--- /dev/null
+
+ GNU LESSER GENERAL PUBLIC LICENSE
+ Version 2.1, February 1999
+
+ Copyright (C) 1991, 1999 Free Software Foundation, Inc.
+ 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+[This is the first released version of the Lesser GPL. It also counts
+ as the successor of the GNU Library Public License, version 2, hence
+ the version number 2.1.]
+
+ Preamble
+
+ The licenses for most software are designed to take away your
+freedom to share and change it. By contrast, the GNU General Public
+Licenses are intended to guarantee your freedom to share and change
+free software--to make sure the software is free for all its users.
+
+ This license, the Lesser General Public License, applies to some
+specially designated software packages--typically libraries--of the
+Free Software Foundation and other authors who decide to use it. You
+can use it too, but we suggest you first think carefully about whether
+this license or the ordinary General Public License is the better
+strategy to use in any particular case, based on the explanations
+below.
+
+ When we speak of free software, we are referring to freedom of use,
+not price. Our General Public Licenses are designed to make sure that
+you have the freedom to distribute copies of free software (and charge
+for this service if you wish); that you receive source code or can get
+it if you want it; that you can change the software and use pieces of
+it in new free programs; and that you are informed that you can do
+these things.
+
+ To protect your rights, we need to make restrictions that forbid
+distributors to deny you these rights or to ask you to surrender these
+rights. These restrictions translate to certain responsibilities for
+you if you distribute copies of the library or if you modify it.
+
+ For example, if you distribute copies of the library, whether gratis
+or for a fee, you must give the recipients all the rights that we gave
+you. You must make sure that they, too, receive or can get the source
+code. If you link other code with the library, you must provide
+complete object files to the recipients, so that they can relink them
+with the library after making changes to the library and recompiling
+it. And you must show them these terms so they know their rights.
+
+ We protect your rights with a two-step method: (1) we copyright the
+library, and (2) we offer you this license, which gives you legal
+permission to copy, distribute and/or modify the library.
+
+ To protect each distributor, we want to make it very clear that
+there is no warranty for the free library. Also, if the library is
+modified by someone else and passed on, the recipients should know
+that what they have is not the original version, so that the original
+author's reputation will not be affected by problems that might be
+introduced by others.
+^L
+ Finally, software patents pose a constant threat to the existence of
+any free program. We wish to make sure that a company cannot
+effectively restrict the users of a free program by obtaining a
+restrictive license from a patent holder. Therefore, we insist that
+any patent license obtained for a version of the library must be
+consistent with the full freedom of use specified in this license.
+
+ Most GNU software, including some libraries, is covered by the
+ordinary GNU General Public License. This license, the GNU Lesser
+General Public License, applies to certain designated libraries, and
+is quite different from the ordinary General Public License. We use
+this license for certain libraries in order to permit linking those
+libraries into non-free programs.
+
+ When a program is linked with a library, whether statically or using
+a shared library, the combination of the two is legally speaking a
+combined work, a derivative of the original library. The ordinary
+General Public License therefore permits such linking only if the
+entire combination fits its criteria of freedom. The Lesser General
+Public License permits more lax criteria for linking other code with
+the library.
+
+ We call this license the "Lesser" General Public License because it
+does Less to protect the user's freedom than the ordinary General
+Public License. It also provides other free software developers Less
+of an advantage over competing non-free programs. These disadvantages
+are the reason we use the ordinary General Public License for many
+libraries. However, the Lesser license provides advantages in certain
+special circumstances.
+
+ For example, on rare occasions, there may be a special need to
+encourage the widest possible use of a certain library, so that it
+becomes
+a de-facto standard. To achieve this, non-free programs must be
+allowed to use the library. A more frequent case is that a free
+library does the same job as widely used non-free libraries. In this
+case, there is little to gain by limiting the free library to free
+software only, so we use the Lesser General Public License.
+
+ In other cases, permission to use a particular library in non-free
+programs enables a greater number of people to use a large body of
+free software. For example, permission to use the GNU C Library in
+non-free programs enables many more people to use the whole GNU
+operating system, as well as its variant, the GNU/Linux operating
+system.
+
+ Although the Lesser General Public License is Less protective of the
+users' freedom, it does ensure that the user of a program that is
+linked with the Library has the freedom and the wherewithal to run
+that program using a modified version of the Library.
+
+ The precise terms and conditions for copying, distribution and
+modification follow. Pay close attention to the difference between a
+"work based on the library" and a "work that uses the library". The
+former contains code derived from the library, whereas the latter must
+be combined with the library in order to run.
+^L
+ GNU LESSER GENERAL PUBLIC LICENSE
+ TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+ 0. This License Agreement applies to any software library or other
+program which contains a notice placed by the copyright holder or
+other authorized party saying it may be distributed under the terms of
+this Lesser General Public License (also called "this License").
+Each licensee is addressed as "you".
+
+ A "library" means a collection of software functions and/or data
+prepared so as to be conveniently linked with application programs
+(which use some of those functions and data) to form executables.
+
+ The "Library", below, refers to any such software library or work
+which has been distributed under these terms. A "work based on the
+Library" means either the Library or any derivative work under
+copyright law: that is to say, a work containing the Library or a
+portion of it, either verbatim or with modifications and/or translated
+straightforwardly into another language. (Hereinafter, translation is
+included without limitation in the term "modification".)
+
+ "Source code" for a work means the preferred form of the work for
+making modifications to it. For a library, complete source code means
+all the source code for all modules it contains, plus any associated
+interface definition files, plus the scripts used to control
+compilation
+and installation of the library.
+
+ Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope. The act of
+running a program using the Library is not restricted, and output from
+such a program is covered only if its contents constitute a work based
+on the Library (independent of the use of the Library in a tool for
+writing it). Whether that is true depends on what the Library does
+and what the program that uses the Library does.
+
+ 1. You may copy and distribute verbatim copies of the Library's
+complete source code as you receive it, in any medium, provided that
+you conspicuously and appropriately publish on each copy an
+appropriate copyright notice and disclaimer of warranty; keep intact
+all the notices that refer to this License and to the absence of any
+warranty; and distribute a copy of this License along with the
+Library.
+
+ You may charge a fee for the physical act of transferring a copy,
+and you may at your option offer warranty protection in exchange for a
+fee.
+\f
+ 2. You may modify your copy or copies of the Library or any portion
+of it, thus forming a work based on the Library, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+ a) The modified work must itself be a software library.
+
+ b) You must cause the files modified to carry prominent notices
+ stating that you changed the files and the date of any change.
+
+ c) You must cause the whole of the work to be licensed at no
+ charge to all third parties under the terms of this License.
+
+ d) If a facility in the modified Library refers to a function or a
+ table of data to be supplied by an application program that uses
+ the facility, other than as an argument passed when the facility
+ is invoked, then you must make a good faith effort to ensure that,
+ in the event an application does not supply such function or
+ table, the facility still operates, and performs whatever part of
+ its purpose remains meaningful.
+
+ (For example, a function in a library to compute square roots has
+ a purpose that is entirely well-defined independent of the
+ application. Therefore, Subsection 2d requires that any
+ application-supplied function or table used by this function must
+ be optional: if the application does not supply it, the square
+ root function must still compute square roots.)
+
+These requirements apply to the modified work as a whole. If
+identifiable sections of that work are not derived from the Library,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works. But when you
+distribute the same sections as part of a whole which is a work based
+on the Library, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote
+it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Library.
+
+In addition, mere aggregation of another work not based on the Library
+with the Library (or with a work based on the Library) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+ 3. You may opt to apply the terms of the ordinary GNU General Public
+License instead of this License to a given copy of the Library. To do
+this, you must alter all the notices that refer to this License, so
+that they refer to the ordinary GNU General Public License, version 2,
+instead of to this License. (If a newer version than version 2 of the
+ordinary GNU General Public License has appeared, then you can specify
+that version instead if you wish.) Do not make any other change in
+these notices.
+^L
+ Once this change is made in a given copy, it is irreversible for
+that copy, so the ordinary GNU General Public License applies to all
+subsequent copies and derivative works made from that copy.
+
+ This option is useful when you wish to copy part of the code of
+the Library into a program that is not a library.
+
+ 4. You may copy and distribute the Library (or a portion or
+derivative of it, under Section 2) in object code or executable form
+under the terms of Sections 1 and 2 above provided that you accompany
+it with the complete corresponding machine-readable source code, which
+must be distributed under the terms of Sections 1 and 2 above on a
+medium customarily used for software interchange.
+
+ If distribution of object code is made by offering access to copy
+from a designated place, then offering equivalent access to copy the
+source code from the same place satisfies the requirement to
+distribute the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+ 5. A program that contains no derivative of any portion of the
+Library, but is designed to work with the Library by being compiled or
+linked with it, is called a "work that uses the Library". Such a
+work, in isolation, is not a derivative work of the Library, and
+therefore falls outside the scope of this License.
+
+ However, linking a "work that uses the Library" with the Library
+creates an executable that is a derivative of the Library (because it
+contains portions of the Library), rather than a "work that uses the
+library". The executable is therefore covered by this License.
+Section 6 states terms for distribution of such executables.
+
+ When a "work that uses the Library" uses material from a header file
+that is part of the Library, the object code for the work may be a
+derivative work of the Library even though the source code is not.
+Whether this is true is especially significant if the work can be
+linked without the Library, or if the work is itself a library. The
+threshold for this to be true is not precisely defined by law.
+
+ If such an object file uses only numerical parameters, data
+structure layouts and accessors, and small macros and small inline
+functions (ten lines or less in length), then the use of the object
+file is unrestricted, regardless of whether it is legally a derivative
+work. (Executables containing this object code plus portions of the
+Library will still fall under Section 6.)
+
+ Otherwise, if the work is a derivative of the Library, you may
+distribute the object code for the work under the terms of Section 6.
+Any executables containing that work also fall under Section 6,
+whether or not they are linked directly with the Library itself.
+^L
+ 6. As an exception to the Sections above, you may also combine or
+link a "work that uses the Library" with the Library to produce a
+work containing portions of the Library, and distribute that work
+under terms of your choice, provided that the terms permit
+modification of the work for the customer's own use and reverse
+engineering for debugging such modifications.
+
+ You must give prominent notice with each copy of the work that the
+Library is used in it and that the Library and its use are covered by
+this License. You must supply a copy of this License. If the work
+during execution displays copyright notices, you must include the
+copyright notice for the Library among them, as well as a reference
+directing the user to the copy of this License. Also, you must do one
+of these things:
+
+ a) Accompany the work with the complete corresponding
+ machine-readable source code for the Library including whatever
+ changes were used in the work (which must be distributed under
+ Sections 1 and 2 above); and, if the work is an executable linked
+ with the Library, with the complete machine-readable "work that
+ uses the Library", as object code and/or source code, so that the
+ user can modify the Library and then relink to produce a modified
+ executable containing the modified Library. (It is understood
+ that the user who changes the contents of definitions files in the
+ Library will not necessarily be able to recompile the application
+ to use the modified definitions.)
+
+ b) Use a suitable shared library mechanism for linking with the
+ Library. A suitable mechanism is one that (1) uses at run time a
+ copy of the library already present on the user's computer system,
+ rather than copying library functions into the executable, and (2)
+ will operate properly with a modified version of the library, if
+ the user installs one, as long as the modified version is
+ interface-compatible with the version that the work was made with.
+
+ c) Accompany the work with a written offer, valid for at
+ least three years, to give the same user the materials
+ specified in Subsection 6a, above, for a charge no more
+ than the cost of performing this distribution.
+
+ d) If distribution of the work is made by offering access to copy
+ from a designated place, offer equivalent access to copy the above
+ specified materials from the same place.
+
+ e) Verify that the user has already received a copy of these
+ materials or that you have already sent this user a copy.
+
+ For an executable, the required form of the "work that uses the
+Library" must include any data and utility programs needed for
+reproducing the executable from it. However, as a special exception,
+the materials to be distributed need not include anything that is
+normally distributed (in either source or binary form) with the major
+components (compiler, kernel, and so on) of the operating system on
+which the executable runs, unless that component itself accompanies
+the executable.
+
+ It may happen that this requirement contradicts the license
+restrictions of other proprietary libraries that do not normally
+accompany the operating system. Such a contradiction means you cannot
+use both them and the Library together in an executable that you
+distribute.
+^L
+ 7. You may place library facilities that are a work based on the
+Library side-by-side in a single library together with other library
+facilities not covered by this License, and distribute such a combined
+library, provided that the separate distribution of the work based on
+the Library and of the other library facilities is otherwise
+permitted, and provided that you do these two things:
+
+ a) Accompany the combined library with a copy of the same work
+ based on the Library, uncombined with any other library
+ facilities. This must be distributed under the terms of the
+ Sections above.
+
+ b) Give prominent notice with the combined library of the fact
+ that part of it is a work based on the Library, and explaining
+ where to find the accompanying uncombined form of the same work.
+
+ 8. You may not copy, modify, sublicense, link with, or distribute
+the Library except as expressly provided under this License. Any
+attempt otherwise to copy, modify, sublicense, link with, or
+distribute the Library is void, and will automatically terminate your
+rights under this License. However, parties who have received copies,
+or rights, from you under this License will not have their licenses
+terminated so long as such parties remain in full compliance.
+
+ 9. You are not required to accept this License, since you have not
+signed it. However, nothing else grants you permission to modify or
+distribute the Library or its derivative works. These actions are
+prohibited by law if you do not accept this License. Therefore, by
+modifying or distributing the Library (or any work based on the
+Library), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Library or works based on it.
+
+ 10. Each time you redistribute the Library (or any work based on the
+Library), the recipient automatically receives a license from the
+original licensor to copy, distribute, link with or modify the Library
+subject to these terms and conditions. You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties with
+this License.
+^L
+ 11. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License. If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Library at all. For example, if a patent
+license would not permit royalty-free redistribution of the Library by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Library.
+
+If any portion of this section is held invalid or unenforceable under
+any particular circumstance, the balance of the section is intended to
+apply, and the section as a whole is intended to apply in other
+circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system which is
+implemented by public license practices. Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+ 12. If the distribution and/or use of the Library is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Library under this License
+may add an explicit geographical distribution limitation excluding those
+countries, so that distribution is permitted only in or among
+countries not thus excluded. In such case, this License incorporates
+the limitation as if written in the body of this License.
+
+ 13. The Free Software Foundation may publish revised and/or new
+versions of the Lesser General Public License from time to time.
+Such new versions will be similar in spirit to the present version,
+but may differ in detail to address new problems or concerns.
+
+Each version is given a distinguishing version number. If the Library
+specifies a version number of this License which applies to it and
+"any later version", you have the option of following the terms and
+conditions either of that version or of any later version published by
+the Free Software Foundation. If the Library does not specify a
+license version number, you may choose any version ever published by
+the Free Software Foundation.
+^L
+ 14. If you wish to incorporate parts of the Library into other free
+programs whose distribution conditions are incompatible with these,
+write to the author to ask for permission. For software which is
+copyrighted by the Free Software Foundation, write to the Free
+Software Foundation; we sometimes make exceptions for this. Our
+decision will be guided by the two goals of preserving the free status
+of all derivatives of our free software and of promoting the sharing
+and reuse of software generally.
+
+ NO WARRANTY
+
+ 15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO
+WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW.
+EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR
+OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY
+KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE
+LIBRARY IS WITH YOU. SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME
+THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+ 16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
+WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY
+AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU
+FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR
+CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE
+LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING
+RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A
+FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF
+SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
+DAMAGES.
+
+ END OF TERMS AND CONDITIONS
+^L
+ How to Apply These Terms to Your New Libraries
+
+ If you develop a new library, and you want it to be of the greatest
+possible use to the public, we recommend making it free software that
+everyone can redistribute and change. You can do so by permitting
+redistribution under these terms (or, alternatively, under the terms
+of the ordinary General Public License).
+
+ To apply these terms, attach the following notices to the library.
+It is safest to attach them to the start of each source file to most
+effectively convey the exclusion of warranty; and each file should
+have at least the "copyright" line and a pointer to where the full
+notice is found.
+
+
+ <one line to give the library's name and a brief idea of what it
+does.>
+ Copyright (C) <year> <name of author>
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+
+Also add information on how to contact you by electronic and paper
+mail.
+
+You should also get your employer (if you work as a programmer) or
+your
+school, if any, to sign a "copyright disclaimer" for the library, if
+necessary. Here is a sample; alter the names:
+
+ Yoyodyne, Inc., hereby disclaims all copyright interest in the
+ library `Frob' (a library for tweaking knobs) written by James
+Random Hacker.
+
+ <signature of Ty Coon>, 1 April 1990
+ Ty Coon, President of Vice
+
+That's all there is to it!
+
+
--- /dev/null
+0.3.107
+ - Make tests compile again on modern systems (warnings + -Werror)
+ - Add 'make partcheck' and don't require manual setup for testing.
+ - Change test harness to compile against this dir, not global install
+ - Fix 5.t for archs where PROT_WRITE mappings are readable.
+ - Allow sending of SIGXFSZ on aio over limits
+ - Explicitly specify bash for runtests.sh
+ - Put deprecating comments on never-merged io_prep_poll
+ - Add io_prep_preadv and io_prep_pwritev
+ - Add eventfd support (io_set_eventfd).
+
+0.4.0
+ - remove libredhat-kernel
+ - add rough outline for man pages
+ - make the compiled io_getevents() add the extra parameter and
+ pass the timeout for updating as per 2.5
+ - fixes for ia64, now works
+ - fixes for x86-64
+ - powerpc support from Gianni Tedesco <gianni@ecsc.co.uk>
+ - disable the NULL check in harness/cases/4.t on ia64: ia64
+ maps the 0 page and causes this check to fail.
+
+0.3.15
+ - use real syscall interface, but don't break source compatibility
+ yet (that will happen with 0.4.0)
+
+0.3.13
+ - add test cases
+
+0.3.11
+ - use library versioning of libredhat-kernel to always provide a
+ fallback
+
+0.3.9
+ - add io_queue_release function
+
+0.3.8
+ - make clean deletes libredhat-kernel.so.1
+ - const struct timespec *
+ - add make srpm target
+
+0.3.7
+ - fix assembly function .types
+ - export io_getevents
+ - fix io_submit function prototype to match the kernel
+ - provide /usr/lib/libredhat-kernel.so link for compilation
+ (do NOT link against libredhat-kernel.so directly)
+ - fix soname to libaio.so.1
+ - fix dummy libredhat-kernel's soname
+ - work around nfs bug
+ - provide and install libredhat-kernel.so.1 stub
+ - Makefile improvements
+ - make sure dummy libredhat-kernel.so only returns -ENOSYS
+
--- /dev/null
+To install the library, execute the command:
+
+ make prefix=`pwd`/usr install
+
+which will install the binaries and header files into the directory
+usr. Set prefix=/usr to get them installed into the main system.
+
+Please note: Do not attempt to install on the system the
+"libredhat-kernel.so" file. It is a dummy shared library
+provided only for the purpose of being able to bootstrap
+this facility while running on systems without the correct
+libredhat-kernel.so built. The contents of the included
+libredhat-kernel.so are only stubs; this library is NOT
+functional for anything except the internal purpose of
+linking libaio.so against the provided stubs. At runtime,
+libaio.so requires a real libredhat-kernel.so library; this
+is provided by the Red Hat kernel RPM packages with async
+I/O functionality.
--- /dev/null
+NAME=libaio
+SPECFILE=$(NAME).spec
+VERSION=$(shell awk '/Version:/ { print $$2 }' $(SPECFILE))
+TAG = $(NAME)-$(VERSION)
+RPMBUILD=$(shell `which rpmbuild >&/dev/null` && echo "rpmbuild" || echo "rpm")
+
+prefix=/usr
+includedir=$(prefix)/include
+libdir=$(prefix)/lib
+
+default: all
+
+all:
+ @$(MAKE) -C src
+
+install:
+ @$(MAKE) -C src install prefix=$(DESTDIR)$(prefix) includedir=$(DESTDIR)$(includedir) libdir=$(DESTDIR)$(libdir)
+
+check:
+ @$(MAKE) -C harness check
+
+partcheck: all
+ @$(MAKE) -C harness partcheck
+
+clean:
+ @$(MAKE) -C src clean
+ @$(MAKE) -C harness clean
+
+tag-archive:
+ @git tag $(TAG)
+
+create-archive:
+ @git archive --prefix=$(NAME)-$(VERSION)/ -o $(NAME)-$(VERSION).tar.gz $(TAG)
+ @echo "The final archive is ./$(NAME)-$(VERSION).tar.gz."
+
+archive: clean tag-archive create-archive
+
+srpm: create-archive
+ $(RPMBUILD) --define "_sourcedir `pwd`" --define "_srcrpmdir `pwd`" --nodeps -bs $(SPECFILE)
--- /dev/null
+# libaio
+
+The Linux-native asynchronous I/O facility ("async I/O", or "aio") has a
+richer API and capability set than the simple POSIX async I/O facility.
+This library, libaio, provides the Linux-native API for async I/O.
+The POSIX async I/O facility requires this library in order to provide
+kernel-accelerated async I/O capabilities, as do applications which
+require the Linux-native async I/O API.
+
+## Contributing
+
+Patches should be submitted to linux-aio@kvack.org.
--- /dev/null
+- Write man pages.
+- Make -static links against libaio work.
+- Fallback on userspace if the kernel calls return -ENOSYS.
+
--- /dev/null
+# foo.
+TEST_SRCS:=$(shell find cases/ -name \*.t | sort -n -t/ -k2)
+EXTRAPROGS:=cases/8.p cases/10.p
+PARTPROGS:=$(filter-out $(EXTRAPROGS), $(patsubst %.t,%.p,$(TEST_SRCS)))
+PROGS:=$(PARTPROGS) $(EXTRAPROGS)
+HARNESS_SRCS:=main.c
+# io_queue.c
+
+CFLAGS+=-Wall -Werror -I../src -g -O2
+#-lpthread -lrt
+
+# gcc-11 does not like the test case in 3.t that
+# passes an invalid pointer (-1) to the kernel, so
+# tell the compiler we do not need a warning here
+cases/3.p: CFLAGS+=-Wno-stringop-overflow
+
+# Change this on the build line to run tests against the installed libraries:
+# make LIBAIO=-laio partcheck
+LIBAIO?=../src/libaio.a
+
+all: $(PROGS)
+
+$(PROGS): %.p: %.t $(HARNESS_SRCS)
+ $(CC) $(CFLAGS) -DTEST_NAME=\"$<\" -o $@ main.c $(LIBAIO) -lpthread
+
+clean:
+ rm -f $(PROGS) *.o runtests.out rofile wofile rwfile
+
+.PHONY:
+
+testdir/rofile: testdir .PHONY
+ rm -f $@
+ echo "test" >$@
+ chmod 400 $@
+
+testdir/wofile: testdir .PHONY
+ rm -f $@
+ echo "test" >$@
+ chmod 200 $@
+
+testdir/rwfile: testdir .PHONY
+ rm -f $@
+ echo "test" >$@
+ chmod 600 $@
+
+testdir testdir.enospc testdir.ext2:
+ mkdir $@
+
+root: .PHONY
+ @if [ `id -u` -ne 0 ]; then echo Need root for check, try partcheck >&2; exit 1; fi
+
+partcheck: $(PARTPROGS) testdir/rofile testdir/rwfile testdir/wofile
+ ./runtests.sh $(PARTPROGS)
+
+ext2.img:
+ dd if=/dev/zero bs=1M count=10 of=$@
+ mke2fs -F -b 4096 $@
+
+extracheck: $(EXTRAPROGS) root testdir.ext2 testdir.enospc ext2.img
+ mount -o loop -t ext2 ext2-enospc.img testdir.enospc
+ ./runtests.sh cases/10.p; ret=$$?; umount testdir.enospc; exit $$ret
+ mount -o loop -t ext2 ext2.img testdir.ext2
+ ./runtests.sh cases/8.p; ret=$$?; umount testdir.ext2; exit $$ret
+
+check: partcheck extracheck
--- /dev/null
+Notes on running this test suite:
+
+To run the test suite, run "make check". All test cases should pass
+and there should be 0 fails.
+
+Several of the test cases require a directory on the filesystem under
+test for the creation of test files, as well as the generation of
+error conditions. The test cases assume the directories (or symlinks
+to directories) are as follows:
+
+ testdir/
+ - used for general read/write test cases. Must have at
+ least as much free space as the machine has RAM (up
+ to 768MB).
+ testdir.enospc/
+ - a filesystem that has space for writing 8KB out, but
+ fails with -ENOSPC beyond 8KB.
+ testdir.ext2/
+ - must be an ext2 filesystem.
--- /dev/null
+/* 0.t
+ Test harness check: okay.
+*/
+int test_main(void)
+{
+ printf("test_main: okay\n");
+ return 0;
+}
+
--- /dev/null
+/* 1.t
+ Test harness check: fail.
+*/
+int test_main(void)
+{
+ printf("test_main: fail\n");
+ return 1;
+}
+
--- /dev/null
+/* 10.t - uses testdir.enospc/rwfile
+- Check results on out-of-space and out-of-quota. (10.t)
+ - write that fills filesystem but does not go over should succeed
+ - write that fills filesystem and goes over should be partial
+ - write to full filesystem should return -ENOSPC
+ - read beyond end of file after ENOSPC should return 0
+*/
+#include "aio_setup.h"
+
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <unistd.h>
+
+int test_main(void)
+{
+/* Note: changing either of these requires updating the ext2-enospc.img
+ * filesystem image. Also, if SIZE is less than PAGE_SIZE, problems
+ * crop up due to ext2's preallocation.
+ */
+#define LIMIT 65536
+#define SIZE 65536
+ char *buf;
+ int rwfd;
+ int status = 0, res;
+
+ rwfd = open("testdir.enospc/rwfile", O_RDWR|O_CREAT|O_TRUNC, 0600);
+ assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ buf = malloc(SIZE); assert(buf != NULL);
+ memset(buf, 0, SIZE);
+
+
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT-SIZE, WRITE, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT-SIZE, READ, SIZE);
+
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT, WRITE, -ENOSPC);
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT, READ, 0);
+
+ res = ftruncate(rwfd, 0); assert(res == 0);
+
+ status |= attempt_rw(rwfd, buf, SIZE, 1+LIMIT-SIZE, WRITE, SIZE-1);
+ status |= attempt_rw(rwfd, buf, SIZE, 1+LIMIT-SIZE, READ, SIZE-1);
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT, READ, 0);
+
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT, WRITE, -ENOSPC);
+ status |= attempt_rw(rwfd, buf, SIZE, LIMIT, READ, 0);
+ status |= attempt_rw(rwfd, buf, 0, LIMIT, WRITE, 0);
+
+ res = close(rwfd); assert(res == 0);
+ res = unlink("testdir.enospc/rwfile"); assert(res == 0);
+ return status;
+}
+
--- /dev/null
+/* 11.t - uses testdir/rwfile
+- repeated read / write of same page (to check accounting) (11.t)
+*/
+#include "aio_setup.h"
+
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <unistd.h>
+
+int test_main(void)
+{
+#define COUNT 1000000
+#define SIZE 256
+ char *buf;
+ int rwfd;
+ int status = 0;
+ int i;
+
+ rwfd = open("testdir/rwfile", O_RDWR|O_CREAT|O_TRUNC, 0600);
+ assert(rwfd != -1);
+ buf = malloc(SIZE); assert(buf != NULL);
+ memset(buf, 0, SIZE);
+
+ for (i=0; i<COUNT; i++) {
+ status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE_SILENT, SIZE);
+ if (status)
+ break;
+ }
+ printf("completed %d out of %d writes\n", i, COUNT);
+ for (i=0; i<COUNT; i++) {
+ status |= attempt_rw(rwfd, buf, SIZE, 0, READ_SILENT, SIZE);
+ if (status)
+ break;
+ }
+ printf("completed %d out of %d reads\n", i, COUNT);
+
+ return status;
+}
+
--- /dev/null
+/* 12.t
+- ioctx access across fork() (12.t)
+ */
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <signal.h>
+
+#include "aio_setup.h"
+
+void test_child(void)
+{
+ int res;
+ res = attempt_io_submit(io_ctx, 0, NULL, -EINVAL);
+ fflush(stdout);
+ _exit(res);
+}
+
+int test_main(void)
+{
+ int res, status;
+ pid_t pid;
+ sigset_t set;
+
+ if (attempt_io_submit(io_ctx, 0, NULL, 0))
+ return 1;
+
+ sigemptyset(&set);
+ sigaddset(&set, SIGCHLD);
+ sigprocmask(SIG_BLOCK, &set, NULL);
+
+ fflush(NULL);
+ pid = fork(); assert(pid != -1);
+
+ if (pid == 0)
+ test_child();
+
+ res = waitpid(pid, &status, 0);
+ if (res < 0) {
+ printf("waitpid error\n");
+ return res;
+ }
+
+ if (WIFEXITED(status)) {
+ int failed = (WEXITSTATUS(status) != 0);
+ printf("child exited with status %d%s\n", WEXITSTATUS(status),
+ failed ? " -- FAILED" : "");
+ return failed;
+ }
+
+ /* anything else: failed */
+ if (WIFSIGNALED(status))
+ printf("child killed by signal %d -- FAILED.\n",
+ WTERMSIG(status));
+
+ return 1;
+}
--- /dev/null
+/* 13.t - uses testdir/rwfile
+- Submit multiple writes larger than aio-max-size (deadlocks on older
+ aio code)
+*/
+#include "aio_setup.h"
+
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <unistd.h>
+
+int test_main(void)
+{
+#define SIZE (1024 * 1024)
+#define IOS 8
+ struct iocb iocbs[IOS];
+ struct iocb *iocb_list[IOS];
+ char *bufs[IOS];
+ int rwfd;
+ int status = 0, res;
+ int i;
+
+ rwfd = open("testdir/rwfile", O_RDWR|O_CREAT|O_TRUNC, 0600);
+ assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+
+ for (i=0; i<IOS; i++) {
+ bufs[i] = malloc(SIZE);
+ assert(bufs[i] != NULL);
+ memset(bufs[i], 0, SIZE);
+
+ io_prep_pwrite(&iocbs[i], rwfd, bufs[i], SIZE, i * SIZE);
+ iocb_list[i] = &iocbs[i];
+ }
+
+ status |= attempt_io_submit(io_ctx, IOS, iocb_list, IOS);
+
+ for (i=0; i<IOS; i++) {
+ struct timespec ts = { tv_sec: 30, tv_nsec: 0 };
+ struct io_event event;
+ struct iocb *iocb;
+
+ res = io_getevents(io_ctx, 0, 1, &event, &ts);
+ if (res != 1) {
+ status |= 1;
+ printf("io_getevents failed [%d] with res=%d [%s]\n",
+ i, res, (res < 0) ? strerror(-res) : "okay");
+ break;
+ }
+
+ if (event.res != SIZE)
+ status |= 1;
+
+ iocb = (void *)event.obj;
+ printf("event[%d]: write[%d] %s, returned: %ld [%s]\n",
+ i, (int)(iocb - &iocbs[0]),
+ (event.res != SIZE) ? "failed" : "okay",
+ (long)event.res,
+ (event.res < 0) ? strerror(-event.res) : "okay"
+ );
+ }
+
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ res = close(rwfd); assert(res == 0);
+ return status;
+}
+
--- /dev/null
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <signal.h>
+
+#include "aio_setup.h"
+#include <sys/mman.h>
+
+#define SIZE 768*1024*1024
+
+//just submit an I/O
+
+int test_child(void)
+{
+ char *buf;
+ int rwfd;
+ int res;
+ long size;
+ struct iocb iocb;
+ struct iocb *iocbs[] = { &iocb };
+ int loop = 10;
+ int i;
+
+ aio_setup(1024);
+
+ size = SIZE;
+
+ printf("size = %ld\n", size);
+
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd !=
+-1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ buf = malloc(size); assert(buf !=
+NULL);
+
+ for(i=0;i<loop;i++) {
+
+ switch(i%2) {
+ case 0:
+ io_prep_pwrite(&iocb, rwfd, buf, size, 0);
+ break;
+ case 1:
+ io_prep_pread(&iocb, rwfd, buf, size, 0);
+ }
+
+ res = io_submit(io_ctx, 1, iocbs);
+ if (res != 1) {
+ printf("child: submit: io_submit res=%d [%s]\n", res,
+strerror(-res));
+ _exit(1);
+ }
+ }
+
+ res = ftruncate(rwfd, 0); assert(res == 0);
+
+ _exit(0);
+}
+
+/* from 12.t */
+int test_main(void)
+{
+ int res, status;
+ pid_t pid;
+ sigset_t set;
+
+ if (attempt_io_submit(io_ctx, 0, NULL, 0))
+ return 1;
+
+ sigemptyset(&set);
+ sigaddset(&set, SIGCHLD);
+ sigprocmask(SIG_BLOCK, &set, NULL);
+ fflush(NULL);
+ pid = fork(); assert(pid != -1);
+
+ if (pid == 0)
+ test_child();
+
+ res = waitpid(pid, &status, 0);
+ if (res < 0) {
+ printf("waitpid error\n");
+ return res;
+ }
+
+ if (WIFEXITED(status)) {
+ int failed = (WEXITSTATUS(status) != 0);
+ printf("child exited with status %d%s\n", WEXITSTATUS(status),
+ failed ? " -- FAILED" : "");
+ return failed;
+ }
+
+ /* anything else: failed */
+ if (WIFSIGNALED(status))
+ printf("child killed by signal %d -- FAILED.\n",
+ WTERMSIG(status));
+
+ return 1;
+}
--- /dev/null
+/* 15.t
+- pwritev and preadv tests.
+*/
+#include "aio_setup.h"
+#include <sys/mman.h>
+#include <sys/uio.h>
+#include <errno.h>
+
+int test_main(void)
+{
+#define SIZE 512
+#define NUM_IOV 10
+ char buf[SIZE*NUM_IOV];
+ struct iovec iov[NUM_IOV];
+ int rwfd;
+ int status = 0, res, i;
+
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd != -1);
+ res = ftruncate(rwfd, sizeof(buf)); assert(res == 0);
+
+ for (i = 0; i < NUM_IOV; i++) {
+ iov[i].iov_base = buf + i*SIZE;
+ iov[i].iov_len = SIZE;
+ memset(iov[i].iov_base, i, SIZE);
+ }
+ status |= attempt_rw(rwfd, iov, NUM_IOV, 0, WRITEV, SIZE*NUM_IOV);
+ res = pread(rwfd, buf, sizeof(buf), 0); assert(res == sizeof(buf));
+ for (i = 0; i < NUM_IOV; i++) {
+ unsigned int j;
+ for (j = 0; j < SIZE; j++) {
+ if (buf[i*SIZE + j] != i) {
+ printf("Unexpected value after writev at %i\n",
+ i*SIZE + j);
+ status |= 1;
+ break;
+ }
+ }
+ }
+ if (!status)
+ printf("Checking memory: [Success]\n");
+
+ memset(buf, 0, sizeof(buf));
+ status |= attempt_rw(rwfd, iov, NUM_IOV, 0, READV, SIZE*NUM_IOV);
+ for (i = 0; i < NUM_IOV; i++) {
+ unsigned int j;
+ for (j = 0; j < SIZE; j++) {
+ if (buf[i*SIZE + j] != i) {
+ printf("Unexpected value after readv at %i\n",
+ i*SIZE + j);
+ status |= 1;
+ break;
+ }
+ }
+ }
+
+ /* Check that offset works. */
+ status |= attempt_rw(rwfd, iov+1, NUM_IOV-1, SIZE, WRITEV,
+ SIZE*(NUM_IOV-1));
+ memset(buf, 0, sizeof(buf));
+ res = pread(rwfd, buf, sizeof(buf), 0); assert(res == sizeof(buf));
+ for (i = 1; i < NUM_IOV; i++) {
+ unsigned int j;
+ for (j = 0; j < SIZE; j++) {
+ if (buf[i*SIZE + j] != i) {
+ printf("Unexpected value after offset writev at %i\n",
+ i*SIZE + j);
+ status |= 1;
+ break;
+ }
+ }
+ }
+ if (!status)
+ printf("Checking memory: [Success]\n");
+
+ memset(buf, 0, sizeof(buf));
+ status |= attempt_rw(rwfd, iov+1, NUM_IOV-1, SIZE, READV,
+ SIZE*(NUM_IOV-1));
+ for (i = 1; i < NUM_IOV; i++) {
+ unsigned int j;
+ for (j = 0; j < SIZE; j++) {
+ if (buf[i*SIZE + j] != i) {
+ printf("Unexpected value after offset readv at %i\n",
+ i*SIZE + j);
+ status |= 1;
+ break;
+ }
+ }
+ }
+ if (!status)
+ printf("Checking memory: [Success]\n");
+
+ return status;
+}
+
--- /dev/null
+/* 16.t
+- eventfd tests.
+*/
+#include <stdint.h>
+#include <err.h>
+#include <sys/syscall.h> /* For SYS_xxx definitions */
+
+#ifndef SYS_eventfd
+#if defined(__i386__)
+#define SYS_eventfd 323
+#elif defined(__x86_64__)
+#define SYS_eventfd 284
+#elif defined(__ia64__)
+#define SYS_eventfd 1309
+#elif defined(__PPC__)
+#define SYS_eventfd 307
+#elif defined(__s390__)
+#define SYS_eventfd 318
+#elif defined(__alpha__)
+#define SYS_eventfd 478
+#elif defined(__aarch64__) || defined(__loongarch__) || defined(__riscv)
+/* these recent architectures do not implement eventfd, only eventfd2 */
+#define USE_EVENTFD2
+#ifndef SYS_eventfd2
+#define SYS_eventfd2 19
+#endif /* __aarch64__ || __riscv */
+#else
+#error define SYS_eventfd for your arch!
+#endif
+#endif
+
+int test_main(void)
+{
+ /* 10 MB takes long enough that we would fail if eventfd
+ * returned immediately. */
+#define SIZE 10000000
+ char *buf;
+ struct io_event io_event;
+ struct iocb iocb;
+ struct iocb *iocbs[] = { &iocb };
+ int rwfd, efd;
+ int res;
+ io_context_t io_ctx;
+ uint64_t event;
+ struct timespec notime = { .tv_sec = 0, .tv_nsec = 0 };
+
+ buf = malloc(SIZE); assert(buf);
+#ifndef USE_EVENTFD2
+ efd = syscall(SYS_eventfd, 0);
+#else
+ efd = syscall(SYS_eventfd2, 0, 0);
+#endif
+ if (efd < 0) {
+ if (errno == ENOSYS) {
+ printf("No eventfd support. [SKIPPING]\n");
+ exit(0);
+ }
+ err(1, "Failed to get eventfd");
+ }
+
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ memset(buf, 0x42, SIZE);
+
+ /* Write test. */
+ res = io_queue_init(1024, &io_ctx); assert(res == 0);
+ io_prep_pwrite(&iocb, rwfd, buf, SIZE, 0);
+ io_set_eventfd(&iocb, efd);
+ res = io_submit(io_ctx, 1, iocbs); assert(res == 1);
+
+ alarm(30);
+ res = read(efd, &event, sizeof(event)); assert(res == sizeof(event));
+ assert(event == 1);
+
+ /* This should now be ready. */
+ res = io_getevents(io_ctx, 0, 1, &io_event, ¬ime);
+ if (res != 1)
+ err(1, "io_getevents did not return 1 event after eventfd");
+ assert(io_event.res == SIZE);
+ printf("eventfd write test [SUCCESS]\n");
+
+ /* Read test. */
+ memset(buf, 0, SIZE);
+ io_prep_pread(&iocb, rwfd, buf, SIZE, 0);
+ io_set_eventfd(&iocb, efd);
+ res = io_submit(io_ctx, 1, iocbs); assert(res == 1);
+
+ alarm(30);
+ res = read(efd, &event, sizeof(event)); assert(res == sizeof(event));
+ assert(event == 1);
+
+ /* This should now be ready. */
+ res = io_getevents(io_ctx, 0, 1, &io_event, ¬ime);
+ if (res != 1)
+ err(1, "io_getevents did not return 1 event after eventfd");
+ assert(io_event.res == SIZE);
+
+ for (res = 0; res < SIZE; res++)
+ assert(buf[res] == 0x42);
+ printf("eventfd read test [SUCCESS]\n");
+
+ return 0;
+}
+
--- /dev/null
+/*
+ * Copyright (C) 2014, Dan Aloni, Kernelim Ltd.
+ * Copyright (C) 2014, Benjamin LaHaise <bcrl@kvack.org>.
+ * Copyright (C) 2014, Red Hat, Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * 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.
+ */
+/*
+ * Description:
+ * This regression test ensures that submitting more events than can
+ * fit in the completion ring will not result in a hung task.
+ */
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE 1
+#endif
+#include <assert.h>
+#include <errno.h>
+#include <libaio.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/param.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/wait.h>
+
+const int max_events = 32;
+const int io_size = 0x1000;
+struct iocb *io;
+struct iocb **iops;
+struct iovec *iovecs;
+struct io_event *events;
+char *data;
+
+long submitted = 0;
+long completed = 0;
+long pending = 0;
+
+#define SYS_IO_GETEVENTS 0
+#define USER_GETEVENTS 1
+
+static volatile sig_atomic_t done = 0;
+
+struct aio_ring {
+ unsigned id; /* kernel internal index number */
+ unsigned nr; /* number of io_events */
+ volatile unsigned head;
+ volatile unsigned tail;
+
+ unsigned magic;
+ unsigned compat_features;
+ unsigned incompat_features;
+ unsigned header_length; /* size of aio_ring */
+
+ struct io_event io_events[0];
+};
+
+int get_ring_size(int nr_events)
+{
+ io_context_t ctx;
+ int ret, ring_size;
+ struct aio_ring *ring;
+
+ memset(&ctx, 0, sizeof(ctx));
+ ret = io_setup(nr_events, &ctx);
+ assert(!ret);
+
+ ring = (void *)ctx;
+ ring_size = ring->nr;
+
+ ret = io_destroy(ctx);
+ assert(!ret);
+
+ return ring_size;
+}
+
+int user_getevents(io_context_t ctx, int nr_events, struct io_event *event)
+{
+ struct aio_ring *ring = (void *)ctx;
+ int completed = 0;
+ while ((completed < nr_events) && (ring->head != ring->tail)) {
+ unsigned new_head = ring->head;
+ *event = ring->io_events[new_head];
+ new_head += 1;
+ new_head %= ring->nr;
+ ring->head = new_head;
+ completed++;
+ }
+ return completed;
+}
+
+void prune(io_context_t io_ctx, int max_ios, int getevents_type)
+{
+ int ret;
+
+ if (getevents_type == USER_GETEVENTS)
+ ret = user_getevents(io_ctx, max_ios, events);
+ else
+ ret = io_getevents(io_ctx, pending, max_ios, events, NULL);
+ if (ret > 0) {
+ printf("Completed: %d\n", ret);
+ completed += ret;
+ pending -= ret;
+ }
+}
+
+void run_test(int max_ios, int getevents_type)
+{
+ int fd, ret, flags;
+ long i, to_submit;
+ struct iocb **iocb_sub;
+ io_context_t io_ctx;
+ const char *filename = "testfile";
+
+ printf("MAX_IOS: %d, %s\n", max_ios, getevents_type == USER_GETEVENTS ?
+ "USER_GETEVENTS" : "IO_GETEVENTS");
+ memset(&io_ctx, 0, sizeof(io_ctx));
+ ret = io_setup(max_events, &io_ctx);
+ assert(!ret);
+
+ io = calloc(max_ios, sizeof(*io));
+ iops = calloc(max_ios, sizeof(*iops));
+ iovecs = calloc(max_ios, sizeof(*iovecs));
+ events = calloc(max_ios, sizeof(*events));
+
+ unlink(filename);
+ fd = open(filename, O_CREAT | O_RDWR, 0644);
+ assert(fd >= 0);
+
+ /*
+ * Use O_DIRECT if it's available. If it's not, the test code
+ * will still operate correctly, just potentially slower.
+ */
+ flags = fcntl(fd, F_GETFL, 0);
+ fcntl(fd, F_SETFL, flags | O_DIRECT);
+
+ ret = ftruncate(fd, max_ios * io_size);
+ assert(!ret);
+
+ data = mmap(NULL, io_size * max_ios, PROT_READ | PROT_WRITE,
+ MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+ assert(data != MAP_FAILED);
+
+ for (i = 0; i < max_ios; i++) {
+ iops[i] = &io[i];
+ io[i].data = io;
+ iovecs[i].iov_base = &data[io_size * i];
+ iovecs[i].iov_len = io_size;
+ io_prep_preadv(&io[i], fd, &iovecs[i], 1, 0);
+ }
+
+ submitted = completed = pending = 0;
+
+ to_submit = max_ios;
+ iocb_sub = iops;
+
+ while (submitted < max_ios) {
+ printf("Submitting: %ld\n", to_submit);
+
+ ret = io_submit(io_ctx, to_submit, iocb_sub);
+ if (ret >= 0) {
+ printf("Submitted: %d\n", ret);
+ submitted += ret;
+ iocb_sub += ret;
+ pending += ret;
+ to_submit -= ret;
+ } else {
+ if (ret == -EAGAIN) {
+ printf("Submitted too much, that's okay\n");
+ prune(io_ctx, max_ios, getevents_type);
+ }
+ }
+ }
+
+ prune(io_ctx, max_ios, getevents_type);
+ io_destroy(io_ctx);
+ close(fd);
+ ret = munmap(data, io_size * max_ios);
+ assert(!ret);
+
+ printf("Verifying...\n");
+
+ assert(completed == submitted);
+
+ printf("OK\n");
+}
+
+void run_child(void)
+{
+ int ring_size;
+
+ ring_size = get_ring_size(max_events);
+
+ printf("aio ring size: %d\n", ring_size);
+
+ run_test(ring_size-1, SYS_IO_GETEVENTS);
+ run_test(ring_size, SYS_IO_GETEVENTS);
+ run_test(ring_size+1, SYS_IO_GETEVENTS);
+ run_test(ring_size*2, SYS_IO_GETEVENTS);
+ run_test(ring_size*4, SYS_IO_GETEVENTS);
+
+ run_test(ring_size-1, USER_GETEVENTS);
+ run_test(ring_size, USER_GETEVENTS);
+ run_test(ring_size+1, USER_GETEVENTS);
+ run_test(ring_size*2, USER_GETEVENTS);
+ run_test(ring_size*4, USER_GETEVENTS);
+
+ exit(0);
+}
+
+void sighandler(int signo)
+{
+ assert(signo == SIGCHLD);
+ done = 1;
+}
+
+int test_main(void)
+{
+ unsigned int ret;
+ sighandler_t oldhandler;
+ pid_t child;
+
+ switch (child = fork()) {
+ case 0: /* child */
+ run_child();
+ break;
+ case -1:
+ perror("fork");
+ exit(1);
+ default:
+ oldhandler = signal(SIGCHLD, sighandler);
+ assert(oldhandler != SIG_ERR);
+ break;
+ }
+
+ ret = sleep(10);
+ if (ret != 0) {
+ pid_t pid;
+ int status;
+
+ assert(done);
+
+ pid = wait(&status);
+ if (pid != child) {
+ perror("wait");
+ exit(1);
+ }
+
+ return WEXITSTATUS(status);
+ }
+
+ return 1; /* failed */
+}
+/*
+ * Local Variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * 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.
+ */
+
+/*
+ * Author: Anatol Pomozov <anatol.pomozov@gmail.com>
+ *
+ * Description: This code tests to make sure that when io_destroy
+ * returns, all outstanding I/Os have been completed. It does this by
+ * issuing one I/O and then calling io_destroy (without calling
+ * io_getevents). After the call to io_destroy, the buffer is checked
+ * to ensure that the data was retrieved. This is done simultaneously
+ * from 100 threads.
+ */
+
+#define _GNU_SOURCE
+
+#include <pthread.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <assert.h>
+#include <unistd.h>
+#include <string.h>
+#include <libaio.h>
+#include <stdbool.h>
+
+#define FILENAME "tempfile"
+#define FILEPATTERN '1'
+#define DESTROY_PATTERN '2'
+
+#define THREADS_NUM 100
+
+static size_t page_size;
+
+void
+aio_worker(void *ptr)
+{
+ int i, j, fd, ret;
+ char *buffer = NULL;
+
+ ret = posix_memalign((void **)&buffer, page_size, page_size);
+ assert(ret == 0);
+ assert(buffer != NULL);
+
+ fd = open(FILENAME, O_DIRECT|O_RDONLY);
+ if (fd < 0 && errno == EINVAL)
+ exit(3); /* skip this test, O_DIRECT is unavailable */
+ assert(fd >= 0);
+
+ for (i = 0; i < 1000; i++) {
+ io_context_t ctx;
+ struct iocb cb;
+ struct iocb *cbs[1];
+
+ assert(!io_queue_init(1, &ctx));
+ io_prep_pread(&cb, fd, buffer, page_size, 0);
+ cbs[0] = &cb;
+
+ memset(buffer, '0', page_size);
+ assert(io_submit(ctx, 1, &cbs[0]) == 1);
+ // wait random time (0-500ms) ?
+
+ io_destroy(ctx);
+ memset(buffer, DESTROY_PATTERN, page_size);
+ // wait random for (0-500ms) ?
+
+ // check it is still DESTROY_PATTERN
+ for (j = 0; j < page_size; j++) {
+ if (buffer[j] != DESTROY_PATTERN) {
+ fprintf(stderr,
+ "Buffer has unexpected character: %c\n",
+ buffer[j]);
+ exit(EXIT_FAILURE);
+ }
+ }
+ }
+
+ free(buffer);
+ close(fd);
+}
+
+int
+test_main(void)
+{
+ int i, fd, ret;
+ char *buffer = NULL;
+ pthread_t threads[THREADS_NUM];
+
+ page_size = sysconf(_SC_PAGESIZE);
+ assert(page_size >= 1);
+
+ ret = posix_memalign((void **)&buffer, page_size, page_size);
+ assert(ret == 0);
+ assert(buffer != NULL);
+
+ fd = open(FILENAME, O_CREAT|O_TRUNC|O_APPEND|O_RDWR, S_IRUSR|S_IWUSR);
+ assert(fd != -1);
+
+ memset(buffer, FILEPATTERN, page_size);
+ ret = write(fd, buffer, page_size);
+ assert(ret == page_size);
+ close(fd);
+
+ for (i = 0; i < THREADS_NUM; i++) {
+ ret = pthread_create(&threads[i], NULL,
+ (void *)&aio_worker, NULL);
+ assert(ret == 0);
+ }
+ for (i = 0; i < THREADS_NUM; i++) {
+ ret = pthread_join(threads[i], NULL);
+ assert(ret == 0);
+ }
+
+ return EXIT_SUCCESS;
+}
+/*
+ * Local Variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+/*
+ * Copyright 2015, Red Hat, Inc.
+ *
+ * This test remaps the aio ring buffer and ensures that I/O completions
+ * can still be reaped from userspace.
+ *
+ * Author: Jeff Moyer <jmoyer@redhat.com>
+ */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <signal.h>
+#include <sched.h>
+#include <libaio.h>
+
+#define BUFLEN 4096
+#define TEMPLATE "19.XXXXXX"
+
+volatile sig_atomic_t timed_out = 0;
+
+struct aio_ring {
+ unsigned id; /* kernel internal index number */
+ unsigned nr; /* number of io_events */
+ volatile unsigned head;
+ volatile unsigned tail;
+
+ unsigned magic;
+ unsigned compat_features;
+ unsigned incompat_features;
+ unsigned header_length; /* size of aio_ring */
+
+ struct io_event io_events[0];
+};
+
+int
+open_temp_file(void)
+{
+ int fd, flags;
+ char template[sizeof(TEMPLATE)];
+
+ strncpy(template, TEMPLATE, sizeof(template));
+ fd = mkstemp(template);
+ if (fd < 0) {
+ perror("mkstemp");
+ exit(1);
+ }
+ /*
+ * O_DIRECT is desirable, but not required for this test.
+ */
+ flags = fcntl(F_GETFL, 0);
+ fcntl(F_SETFL, flags | O_DIRECT);
+
+ unlink(template);
+ return fd;
+}
+
+/*
+ * mmap will do the address space search for us. when remapping the ring,
+ * the use of MREMAP_FIXED will cause this mapping to be unmapped.
+ *
+ * len - length in bytes
+ *
+ * Returns the available virtual address, or MAP_FAILED on error.
+ */
+void *
+find_unused_va(size_t len)
+{
+ return mmap(0, len, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+}
+
+void
+alarm_handler(int __attribute__((unused))signo)
+{
+ timed_out = 1;
+}
+
+int
+user_getevents(io_context_t ctx, int nr_events, struct io_event *event)
+{
+ struct aio_ring *ring = (void *)ctx;
+ int completed = 0;
+
+ timed_out = 0;
+ signal(SIGALRM, alarm_handler);
+ alarm(30);
+
+ while ((completed < nr_events) && !timed_out) {
+ unsigned new_head;
+
+ if (ring->head == ring->tail) {
+ sched_yield();
+ continue;
+ }
+
+ new_head = ring->head;
+ *event = ring->io_events[new_head];
+ new_head += 1;
+ new_head %= ring->nr;
+ ring->head = new_head;
+ completed++;
+ }
+
+ alarm(0);
+ signal(SIGALRM, SIG_DFL);
+
+ return completed;
+}
+
+struct aio_ring *
+remap_ring(struct aio_ring *ring)
+{
+ struct aio_ring *new_ring;
+ size_t ring_size;
+
+ /*
+ * No need to round up to page size as ring->nr was adjusted
+ * already to fill the last page in the ring.
+ */
+ ring_size = sizeof(struct aio_ring) + ring->nr * sizeof(struct io_event);
+
+ /*
+ * Remap the ring.
+ */
+ new_ring = find_unused_va(ring_size);
+ if (new_ring == MAP_FAILED) {
+ fprintf(stderr, "Unable to find suitable va for ring\n");
+ return NULL;
+ }
+
+ new_ring = mremap(ring, ring_size, ring_size,
+ MREMAP_FIXED|MREMAP_MAYMOVE, new_ring);
+ if (new_ring == MAP_FAILED || new_ring == ring) {
+ perror("mremap");
+ return NULL;
+ }
+
+ return new_ring;
+}
+
+io_context_t
+remap_io_context(io_context_t ctx)
+{
+ struct aio_ring *ring, *new_ring;
+
+ ring = (void *)ctx;
+ new_ring = remap_ring(ring);
+ if (!new_ring)
+ return NULL;
+
+ ctx = (io_context_t)new_ring;
+ return ctx;
+}
+
+int
+do_io(io_context_t ctx, struct iocb *iocbp, int fd)
+{
+ int ret;
+ char buf[BUFLEN];
+
+ io_prep_pwrite(iocbp, fd, buf, BUFLEN, 0);
+ ret = io_submit(ctx, 1, &iocbp);
+ if (ret != 1) {
+ fprintf(stderr, "io_submit failed with %d\n", ret);
+ return 1;
+ }
+ return 0;
+}
+
+int
+check_completion(io_context_t ctx, struct iocb *iocbp)
+{
+ int ret;
+ struct io_event event;
+
+ ret = user_getevents(ctx, 1, &event);
+ if (ret != 1) {
+ fprintf(stderr, "user_getevents timed out.\n");
+ return 1;
+ }
+
+ if (event.obj != iocbp) {
+ fprintf(stderr,
+ "Error: event->opj (%p) does not match iocbp (%p)\n",
+ event.obj, iocbp);
+ return 1;
+ }
+
+ return 0;
+}
+
+int
+test_main()
+{
+ int fd;
+ int ret;
+ io_context_t ctx;
+ struct iocb iocb;
+
+ memset(&ctx, 0, sizeof(ctx));
+ fd = open_temp_file();
+
+ ret = io_setup(1, &ctx);
+ if (ret != 0) {
+ fprintf(stderr, "io_setup failed with %d\n", ret);
+ return 1;
+ }
+
+ /*
+ * First, try remapping the ring buffer in-between io_setup and
+ * io_submit.
+ */
+ ctx = remap_io_context(ctx);
+ if (ctx == NULL)
+ return 1;
+
+ ret = do_io(ctx, &iocb, fd);
+ if (ret != 0)
+ return 1;
+
+ ret = check_completion(ctx, &iocb);
+ if (ret != 0)
+ return 1;
+
+ /*
+ * Now remap the ring in between io_submit and getevents.
+ */
+ ret = do_io(ctx, &iocb, fd);
+ if (ret != 0)
+ return 1;
+
+ ctx = remap_io_context(ctx);
+ if (ctx == NULL)
+ return 1;
+
+ ret = check_completion(ctx, &iocb);
+ if (ret != 0)
+ return 1;
+
+ /*
+ * Success, clean up.
+ */
+ ret = io_destroy(ctx);
+ if (ret != 0) {
+ fprintf(stderr, "io_destroy failed with %d\n", ret);
+ return 1;
+ }
+ close(fd);
+
+ return 0;
+}
+/*
+ * Local variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+/* 2.t
+- io_setup (#2)
+ - with invalid context pointer
+ - with maxevents <= 0
+ - with an already initialized ctxp
+*/
+
+int attempt(int n, io_context_t *ctxp, int expect)
+{
+ int res;
+
+ printf("expect %3d: io_setup(%5d, %p) = ", expect, n, ctxp);
+ fflush(stdout);
+ res = io_setup(n, ctxp);
+ printf("%3d [%s]%s\n", res, strerror(-res),
+ (res != expect) ? " -- FAILED" : "");
+ if (res != expect)
+ return 1;
+
+ return 0;
+}
+
+int test_main(void)
+{
+ io_context_t ctx;
+ int status = 0;
+
+ ctx = NULL;
+ status |= attempt(-1000, KERNEL_RW_POINTER, -EFAULT);
+ status |= attempt( 1000, KERNEL_RW_POINTER, -EFAULT);
+ status |= attempt( 0, KERNEL_RW_POINTER, -EFAULT);
+ status |= attempt(-1000, &ctx, -EINVAL);
+ status |= attempt( -1, &ctx, -EINVAL);
+ status |= attempt( 0, &ctx, -EINVAL);
+ assert(ctx == NULL);
+ status |= attempt( 1, &ctx, 0);
+ status |= attempt( 1, &ctx, -EINVAL);
+
+ return status;
+}
+
--- /dev/null
+/*
+ * Copyright 2017, Red Hat, Inc.
+ * Author: Jeff Moyer <jmoyer@redhat.com>
+ * Based on test code from Mauricio Faria de Oliveira
+ * <mauricfo@linux.vnet.ibm.com>
+ * License: GPLv2
+ *
+ * Description: Ensure that aio-max-nr requests can be allocated, or,
+ * if not, that the reason is not faulty accounting.
+ */
+#include <libaio.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+
+#define FAIL 1
+
+#define AIO_MAX_NR "/proc/sys/fs/aio-max-nr"
+#define AIO_NR "/proc/sys/fs/aio-nr"
+
+static unsigned aio_max_nr;
+
+int
+read_proc_val(const char *path, unsigned *val)
+{
+ FILE *fp;
+ int ret;
+
+ fp = fopen(path, "r");
+ if (!fp) {
+ fprintf(stderr, "Unable to open proc file \"%s\" for reading\n",
+ path);
+ return FAIL;
+ }
+
+ ret = fscanf(fp, "%u\n", val);
+ fclose(fp);
+
+ if (ret == EOF) {
+ fprintf(stderr, "Failed to read from proc file \"%s\"\n",
+ path);
+ return FAIL;
+ }
+
+ return 0;
+}
+
+/*
+ * Create as many ioctx-s with nr_events each as possible (up to aio_max_nr).
+ * Report any failures of -EAGAIN.
+ */
+int
+do_alloc_ioctxs(int nr_events)
+{
+ long ret;
+ unsigned i, avail, aio_nr, nr_ctxs;
+ io_context_t *ioctx;
+
+ ret = read_proc_val(AIO_NR, &aio_nr);
+ if (ret)
+ return FAIL;
+
+ avail = aio_max_nr - aio_nr;
+ nr_ctxs = avail / nr_events;
+ ioctx = calloc(nr_ctxs, sizeof(*ioctx));
+ if (!ioctx) {
+ fprintf(stderr, "allocating %u ioctx-s failed with %d\n",
+ nr_ctxs, errno);
+ return FAIL;
+ }
+
+ fprintf(stderr, "Creating %u ioctx-s with %u events each...\n", nr_ctxs,
+ nr_events);
+ fflush(stderr);
+ for (i = 0; i < nr_ctxs; i++) {
+ ret = io_setup(nr_events, &ioctx[i]);
+ if (ret) {
+ /*
+ * EAGAIN is the only failure case we're interested
+ * in. -ENOMEM, for example, is expected in this
+ * test.
+ */
+ if (ret != -EAGAIN)
+ break;
+
+ fprintf(stderr,"io_setup(%u) failed on iteration %d.\n",
+ nr_events, i);
+ fprintf(stderr, "allocated %u of %u possible events.\n",
+ nr_events * i, aio_max_nr);
+ ret = read_proc_val(AIO_NR, &aio_nr);
+ if (ret == 0)
+ fprintf(stderr, "aio_nr is currently at %u\n",
+ aio_nr);
+
+ free(ioctx);
+ return FAIL;
+ }
+ }
+ fprintf(stderr, "Successfully created %u io_context-s\n", i);
+ if (i < nr_ctxs - 1)
+ fprintf(stderr, "Last io_setup call returned %ld (%s)\n", ret,
+ strerror(-ret));
+ fflush(stderr);
+
+ return 0;
+}
+
+/*
+ * We fork off a child to do the actual work. The reason is that each
+ * io_destroy will incur an rcu grace period to complete. That really
+ * adds up, depending on the number of io_contexts created. We take
+ * advantage of an optimization in the kernel that waits for all
+ * contexts to be torn down in one grace period. In other words, just
+ * exiting the process without tearing down the ioctx-s using
+ * io_destroy is way faster.
+ */
+int
+alloc_ioctxs(int nr_events)
+{
+ pid_t child;
+ int ret, status;
+
+ child = fork();
+ switch (child) {
+ case 0: /* child */
+ ret = do_alloc_ioctxs(nr_events);
+ exit(ret);
+ case -1:
+ fprintf(stderr, "fork() failed with %d\n", errno);
+ return FAIL;
+ default:
+ break;
+ }
+ if (waitpid(child, &status, 0) < 0)
+ return FAIL;
+ if (WIFEXITED(status) && WEXITSTATUS(status) == 0)
+ return 0;
+
+ return FAIL;
+}
+
+int
+test_main()
+{
+ int ret;
+ unsigned nr_events;
+
+ ret = read_proc_val(AIO_MAX_NR, &aio_max_nr);
+ if (ret)
+ return FAIL;
+
+ fprintf(stderr, "aio_max_nr: %u\n", aio_max_nr);
+
+ nr_events = 1;
+ while (1) {
+ ret = alloc_ioctxs(nr_events);
+ if (ret)
+ return FAIL;
+
+ if (nr_events == aio_max_nr)
+ break;
+
+ nr_events *= 2;
+ if (nr_events > aio_max_nr)
+ nr_events = aio_max_nr;
+ }
+
+ return 0;
+}
+/*
+ * Local variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+/*
+ * Copyright 2017, Red Hat, Inc.
+ *
+ * Test RWF_NOWAIT.
+ *
+ * RWF_NOWAIT will cause -EAGAIN to be returned in the io_event for
+ * any I/O that cannot be serviced without blocking the submission
+ * thread. Instances covered by the kernel at the time this test was
+ * written include:
+ * - O_DIRECT I/O to a file offset that has populated page cache pages
+ * - the submission context cannot obtain the inode lock
+ * - space allocation is necessary
+ * - we need to wait for other I/O (e.g. in the misaligned I/O case)
+ * - ...
+ *
+
+ * The easiest of these to test is that a direct I/O is writing to a
+ * file offset with populated page cache. We also test to ensure that
+ * we can perform I/O in the absence of the above conditions.
+ *
+ * Author: Jeff Moyer <jmoyer@redhat.com>
+ */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <signal.h>
+#include <sched.h>
+#include <libaio.h>
+
+#define TEMPLATE "21.XXXXXX"
+#define BUFLEN 4096
+
+#ifndef RWF_NOWAIT
+#define RWF_NOWAIT 0x00000008
+#endif
+
+int
+open_temp_file()
+{
+ int fd;
+ char temp_file[sizeof(TEMPLATE)];
+
+ strncpy(temp_file, TEMPLATE, sizeof(temp_file));
+ fd = mkstemp(temp_file);
+ if (fd < 0) {
+ perror("mkstemp");
+ return -1;
+ }
+ unlink(temp_file);
+ return fd;
+}
+
+int
+test_main()
+{
+ int fd, flags;
+ int ret;
+ io_context_t ctx;
+ struct iocb iocb, *iocbp = &iocb;
+ struct io_event event;
+ char buf[BUFLEN] __attribute__((aligned (4096)));
+ struct iovec iov;
+
+ fd = open_temp_file();
+ if (fd < 0)
+ return 1;
+
+ memset(&ctx, 0, sizeof(ctx));
+ ret = io_setup(1, &ctx);
+ if (ret != 0) {
+ fprintf(stderr, "io_setup failed with %d\n", ret);
+ return 1;
+ }
+
+ /*
+ * Perform a buffered write to a file. This instantiates the
+ * block and adds the page to the page cache.
+ */
+ memset(buf, 0xa, BUFLEN);
+ ret = write(fd, buf, BUFLEN);
+ if (ret != BUFLEN) {
+ perror("write");
+ return 1;
+ }
+
+ /*
+ * Now attempt an aio/dio pwritev2 with the RWF_NONBLOCK flag
+ * set.
+ */
+ flags = fcntl(fd, F_GETFL);
+ ret = fcntl(fd, F_SETFL, flags | O_DIRECT);
+ if (ret < 0) {
+ /* SKIP this test if O_DIRECT is not available on this fs */
+ if (errno == EINVAL)
+ return 3;
+ perror("fcntl");
+ return 1;
+ }
+
+ memset(buf, 0, BUFLEN);
+ iov.iov_base = buf;
+ iov.iov_len = BUFLEN;
+ io_prep_preadv2(&iocb, fd, &iov, 1, 0, RWF_NOWAIT);
+
+ ret = io_submit(ctx, 1, &iocbp);
+
+ /*
+ * io_submit will return -EINVAL if RWF_NOWAIT is not supported by
+ * the kernel, and EOPNOTSUPP if it's not supported by the fs.
+ */
+ if (ret != 1) {
+ if (ret == -EINVAL || ret == -ENOTSUP) {
+ fprintf(stderr, "RWF_NOWAIT not supported by %s.\n",
+ ret == -EINVAL ? "kernel" : "file system");
+ /* skip this test */
+ return 3;
+ }
+ errno = -ret;
+ perror("io_submit");
+ return 1;
+ }
+
+ ret = io_getevents(ctx, 1, 1, &event, NULL);
+ if (ret != 1) {
+ errno = -ret;
+ perror("io_getevents");
+ return 1;
+ }
+
+ /*
+ * We expect -EAGAIN due to the existence of a page cache page
+ * for the file system block we are writing.
+ */
+ if (event.res != -EAGAIN) {
+ fprintf(stderr, "Expected -EAGAIN, got %lu\n", event.res);
+ return 1;
+ }
+
+ /*
+ * An O_DIRECT write to the page will force the page out of the
+ * page cache, allowing the subsequent RWF_NOWAIT I/O to complete.
+ */
+ ret = pwrite(fd, buf, BUFLEN, 0);
+ if (ret != BUFLEN) {
+ perror("write");
+ return 1;
+ }
+
+ /*
+ * Now retry the RWF_NOWAIT I/O. This should succeed.
+ */
+ ret = io_submit(ctx, 1, &iocbp);
+ if (ret != 1) {
+ errno = -ret;
+ perror("io_submit");
+ return 1;
+ }
+
+ ret = io_getevents(ctx, 1, 1, &event, NULL);
+ if (ret != 1) {
+ errno = -ret;
+ perror("io_getevents");
+ return 1;
+ }
+
+ if (event.res != BUFLEN) {
+ fprintf(stderr, "Expected %d, got %lu\n", BUFLEN, event.res);
+ return 1;
+ }
+
+ return 0;
+}
+/*
+ * Local variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+/*
+ * Copyright (C) 2006-2018 Free Software Foundation, Inc.
+ * Copyright (C) 2018 Christoph Hellwig.
+ * License: LGPLv2.1 or later.
+ *
+ * Description: test aio poll and io_pgetevents signal handling.
+ *
+ * Very roughly based on glibc tst-pselect.c.
+ */
+#include <errno.h>
+#include <signal.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/poll.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+
+static volatile int handler_called;
+
+static void
+handler(int sig)
+{
+ handler_called = 1;
+}
+
+int test_main(void)
+{
+ struct timespec to = { .tv_sec = 0, .tv_nsec = 500000000 };
+ pid_t parent = getpid(), p;
+ int pipe1[2], pipe2[2];
+ struct sigaction sa = { .sa_flags = 0 };
+ sigset_t sigmask;
+ struct io_context *ctx = NULL;
+ struct io_event ev;
+ struct iocb iocb;
+ struct iocb *iocbs[] = { &iocb };
+ int ret;
+
+ sigemptyset(&sa.sa_mask);
+
+ sa.sa_handler = handler;
+ if (sigaction(SIGUSR1, &sa, NULL) != 0) {
+ printf("sigaction(1) failed\n");
+ return 1;
+ }
+
+ sa.sa_handler = SIG_IGN;
+ if (sigaction(SIGCHLD, &sa, NULL) != 0) {
+ printf("sigaction(2) failed\n");
+ return 1;
+ }
+
+ sigemptyset(&sigmask);
+ sigaddset(&sigmask, SIGUSR1);
+ if (sigprocmask(SIG_BLOCK, &sigmask, NULL) != 0) {
+ printf("sigprocmask failed\n");
+ return 1;
+ }
+
+ if (pipe(pipe1) != 0 || pipe(pipe2) != 0) {
+ printf("pipe failed\n");
+ return 1;
+ }
+
+ sigprocmask(SIG_SETMASK, NULL, &sigmask);
+ sigdelset(&sigmask, SIGUSR1);
+
+ p = fork();
+ switch (p) {
+ case -1:
+ printf("fork failed\n");
+ exit(2);
+ case 0:
+ close(pipe1[1]);
+ close(pipe2[0]);
+
+ ret = io_setup(1, &ctx);
+ if (ret) {
+ printf("child: io_setup failed: %s\n", strerror(-ret));
+ return 1;
+ }
+
+ io_prep_poll(&iocb, pipe1[0], POLLIN);
+ ret = io_submit(ctx, 1, iocbs);
+ if (ret != 1) {
+ /* if poll isn't supported, skip the test */
+ if (ret == -EINVAL)
+ return 3;
+ printf("child: io_submit failed: %s\n", strerror(-ret));
+ return 1;
+ }
+
+ do {
+ if (getppid() != parent) {
+ printf("parent died\n");
+ exit(2);
+ }
+ ret = io_pgetevents(ctx, 1, 1, &ev, &to, &sigmask);
+ } while (ret == 0);
+
+ /* SKIP if the syscall has not been implemented. */
+ if (ret == -ENOSYS)
+ return 3;
+
+ if (ret != -EINTR) {
+ printf("child: io_pgetevents did not set errno to "
+ "EINTR: %s\n", strerror(-ret));
+ return 1;
+ }
+
+ do {
+ errno = 0;
+ ret = write(pipe2[1], "foo", 3);
+ } while (ret == -1 && errno == EINTR);
+
+ exit(0);
+ default:
+ close(pipe1[0]);
+ close(pipe2[1]);
+
+ io_prep_poll(&iocb, pipe2[0], POLLIN);
+
+ ret = io_setup(1, &ctx);
+ if (ret) {
+ printf("parent: io_setup failed: %s\n", strerror(-ret));
+ return 1;
+ }
+
+ ret = io_submit(ctx, 1, iocbs);
+ if (ret != 1) {
+ /* if poll isn't supported, skip the test */
+ if (ret == -EINVAL)
+ return 3;
+ printf("parent: io_submit failed with %d: %s\n",
+ ret, strerror(-ret));
+ return 1;
+ }
+
+ kill(p, SIGUSR1);
+
+ ret = io_pgetevents(ctx, 1, 1, &ev, NULL, &sigmask);
+ /* SKIP if the syscall has not been implemented. */
+ if (ret == -ENOSYS)
+ return 3;
+ if (ret < 0) {
+ printf("parent: io_pgetevents failed: %s\n",
+ strerror(-ret));
+ return 1;
+ }
+ if (ret != 1) {
+ printf("parent: io_pgetevents did not report event\n");
+ return 1;
+ }
+ if (ev.obj != &iocb) {
+ printf("parent: io_pgetevents reports wrong fd\n");
+ return 1;
+ }
+ if ((ev.res & POLLIN) != POLLIN) {
+ printf("parent: io_pgetevents did not report readable fd\n");
+ return 1;
+ }
+
+ return 0;
+ }
+}
+/*
+ * Local variables:
+ * mode: c
+ * c-basic-offset: 8
+ * End:
+ */
--- /dev/null
+// SPDX-License-Identifier: LGPL-2.1+
+/*
+ * Copyright 2022 Google LLC
+ */
+
+/*
+ * Verify that aio poll doesn't miss any events. This is a regression test for
+ * kernel commit 363bee27e258 ("aio: keep poll requests on waitqueue until
+ * completed").
+ *
+ * This test repeatedly does the following operations in parallel:
+ *
+ * Thread 1: Aio-poll the read end of a pipe, then wait for it to complete.
+ * Thread 2: Splice a byte to the pipe.
+ * Thread 3: Read from the pipe, then splice another byte to it.
+ *
+ * The pipe will never be empty at the end, so the poll should always complete.
+ * With the bug, that didn't always happen, as the second splice sometimes
+ * didn't cause a wakeup due to the waitqueue entry being temporarily removed,
+ * as per the following buggy sequence of events:
+ *
+ * 1. Thread 1 adds the poll waitqueue entry.
+ * 2. Thread 2 does the first splice. This calls the poll wakeup function,
+ * which deletes the waitqueue entry [BUG!], then schedules the async
+ * completion work.
+ * 3. Thread 3 reads from the pipe.
+ * 4. Async work sees that the pipe isn't ready.
+ * 5. Thread 3 splices some data to the pipe, but doesn't call the poll
+ * wakeup function because the waitqueue is currently empty.
+ * 6. Async work re-adds the waitqueue entry, as the pipe wasn't ready.
+ *
+ * The reason we use splice() rather than write() is because in order for step
+ * (2) to not complete the poll inline, the kernel must not pass an event mask
+ * to the wakeup function. This is the case for splice() to a pipe, but not
+ * write() to a pipe (at least in the kernel versions with the bug).
+ */
+#include <poll.h>
+#include <pthread.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <sys/wait.h>
+
+static int tmpfd;
+static int pipefds[2];
+static pthread_barrier_t barrier;
+static bool exiting;
+
+static void fail(const char *format, ...)
+{
+ va_list va;
+
+ va_start(va, format);
+ vfprintf(stderr, format, va);
+ putc('\n', stderr);
+ va_end(va);
+ exit(1);
+}
+
+static void fail_errno(const char *format, ...)
+{
+ va_list va;
+
+ va_start(va, format);
+ vfprintf(stderr, format, va);
+ fprintf(stderr, ": %s", strerror(errno));
+ putc('\n', stderr);
+ va_end(va);
+ exit(1);
+}
+
+/* Thread 2 */
+static void *thrproc2(void *arg)
+{
+ for (;;) {
+ off_t offset = 0;
+
+ pthread_barrier_wait(&barrier);
+ if (exiting)
+ break;
+
+ /* Splice a byte to the pipe. */
+ if (splice(tmpfd, &offset, pipefds[1], NULL, 1, 0) != 1)
+ fail_errno("splice failed in thread 2");
+
+ pthread_barrier_wait(&barrier);
+ }
+ return NULL;
+}
+
+/* Thread 3 */
+static void *thrproc3(void *arg)
+{
+ for (;;) {
+ char c;
+ off_t offset = 0;
+
+ pthread_barrier_wait(&barrier);
+ if (exiting)
+ break;
+
+ /* Read a byte from the pipe. */
+ if (read(pipefds[0], &c, 1) != 1)
+ fail_errno("read failed in thread 3");
+
+ /* Splice a byte to the pipe. */
+ if (splice(tmpfd, &offset, pipefds[1], NULL, 1, 0) != 1)
+ fail_errno("splice failed in thread 3");
+
+ pthread_barrier_wait(&barrier);
+ }
+ return NULL;
+}
+
+static int child_process(void)
+{
+ int ret;
+ io_context_t ctx = NULL;
+ pthread_t t2, t3;
+ int i;
+ struct iocb iocb;
+ struct iocb *iocbs[] = { &iocb };
+ struct io_event event;
+ struct timespec timeout = { .tv_sec = 5 };
+ char c;
+
+ ret = io_setup(1, &ctx);
+ if (ret != 0) {
+ errno = -ret;
+ fail_errno("io_setup failed");
+ }
+
+ if (pipe(pipefds) != 0)
+ fail_errno("pipe failed");
+
+ errno = pthread_barrier_init(&barrier, NULL, 3);
+ if (errno)
+ fail_errno("pthread_barrier_init failed");
+
+ errno = pthread_create(&t2, NULL, thrproc2, NULL);
+ if (errno)
+ fail_errno("pthread_create failed");
+
+ errno = pthread_create(&t3, NULL, thrproc3, NULL);
+ if (errno)
+ fail_errno("pthread_create failed");
+
+ for (i = 0; i < 5000; i++) {
+ /* Thread 1 */
+
+ /* Submit a poll request. */
+ io_prep_poll(&iocb, pipefds[0], POLLIN);
+ ret = io_submit(ctx, 1, iocbs);
+ if (ret != 1) {
+ /* If aio poll isn't supported, skip the test. */
+ errno = -ret;
+ if (errno == EINVAL) {
+ printf("aio poll is not supported\n");
+ return 3;
+ }
+ fail_errno("io_submit failed");
+ }
+
+ /* Tell the other threads to continue. */
+ pthread_barrier_wait(&barrier);
+
+ /*
+ * Wait for the poll to complete. Wait at most 5 seconds, in
+ * case we hit the bug so the poll wouldn't otherwise complete.
+ */
+ ret = io_getevents(ctx, 1, 1, &event, &timeout);
+ if (ret < 0) {
+ errno = -ret;
+ fail_errno("io_getevents failed");
+ }
+ if (ret == 0) {
+ /*
+ * The poll eventually timed out rather than giving us
+ * an event, so we've detected the bug.
+ */
+ fail("FAIL: poll missed an event!");
+ }
+
+ /* Wait for the other threads to finish their iteration. */
+ pthread_barrier_wait(&barrier);
+
+ /* The pipe has 1 byte; read it to get back to initial state. */
+ if (read(pipefds[0], &c, 1) != 1)
+ fail_errno("read failed in thread 1");
+ }
+ exiting = true;
+ pthread_barrier_wait(&barrier);
+ errno = pthread_join(t2, NULL);
+ if (errno)
+ fail_errno("pthread_join failed");
+ errno = pthread_join(t3, NULL);
+ if (errno)
+ fail_errno("pthread_join failed");
+ return 0;
+}
+
+int test_main(void)
+{
+ char tmpfile_name[] = "/tmp/aio.XXXXXX";
+ int ret;
+ int ncpus;
+ int i;
+ int overall_status = 0;
+ int status;
+
+ /* Create a 1-byte temporary file. */
+ tmpfd = mkstemp(tmpfile_name);
+ if (tmpfd < 0) {
+ fprintf(stderr, "failed to create temporary file: %s\n",
+ strerror(errno));
+ return 1;
+ }
+ if (pwrite(tmpfd, "X", 1, 0) != 1) {
+ fprintf(stderr, "failed to write to temporary file: %s\n",
+ strerror(errno));
+ return 1;
+ }
+
+ /*
+ * Run the test in multiple processes to increase the chance of hitting
+ * the race condition.
+ */
+ ret = sysconf(_SC_NPROCESSORS_CONF);
+ if (ret < 1) {
+ fprintf(stderr, "failed to get number of CPUs: %s\n",
+ strerror(errno));
+ return 1;
+ }
+ ncpus = ret;
+ for (i = 0; i < ncpus; i++) {
+ ret = fork();
+ if (ret < 0) {
+ fprintf(stderr, "fork failed: %s\n", strerror(errno));
+ return 1;
+ }
+ if (ret == 0)
+ exit(child_process());
+ }
+ for (i = 0; i < ncpus; i++) {
+ if (wait(&status) < 0) {
+ fprintf(stderr, "wait failed: %s\n", strerror(errno));
+ return 1;
+ }
+ if (!WIFEXITED(status))
+ overall_status = 1;
+ else if (overall_status == 0)
+ overall_status = WEXITSTATUS(status);
+ }
+ unlink(tmpfile_name);
+ return overall_status;
+}
--- /dev/null
+/* 3.t
+- io_submit/io_getevents with invalid addresses (3.t)
+
+*/
+#include "aio_setup.h"
+
+int test_main(void)
+{
+ struct iocb a, b;
+ struct iocb *good_ios[] = { &a, &b };
+ struct iocb *bad1_ios[] = { NULL, &b };
+ struct iocb *bad2_ios[] = { KERNEL_RW_POINTER, &a };
+ int status = 0;
+
+ status |= attempt_io_submit(BAD_CTX, 1, good_ios, -EINVAL);
+ status |= attempt_io_submit( io_ctx, 0, good_ios, 0);
+ status |= attempt_io_submit( io_ctx, 1, NULL, -EFAULT);
+ status |= attempt_io_submit( io_ctx, 1, (void *)-1, -EFAULT);
+ status |= attempt_io_submit( io_ctx, 2, bad1_ios, -EFAULT);
+ status |= attempt_io_submit( io_ctx, 2, bad2_ios, -EFAULT);
+ status |= attempt_io_submit( io_ctx, -1, good_ios, -EINVAL);
+
+ return status;
+}
+
--- /dev/null
+/* 4.t
+- read of descriptor without read permission (4.t)
+- write to descriptor without write permission (4.t)
+- check that O_APPEND writes actually append
+
+*/
+#include "aio_setup.h"
+
+#define SIZE 512
+#define READ 'r'
+#define WRITE 'w'
+int attempt(int fd, void *buf, int count, long long pos, int rw, int expect)
+{
+ struct iocb iocb;
+ int res;
+
+ switch(rw) {
+ case READ: io_prep_pread (&iocb, fd, buf, count, pos); break;
+ case WRITE: io_prep_pwrite(&iocb, fd, buf, count, pos); break;
+ }
+
+ printf("expect %3d: (%c), res = ", expect, rw);
+ fflush(stdout);
+ res = sync_submit(&iocb);
+ printf("%3d [%s]%s\n", res, (res <= 0) ? strerror(-res) : "Success",
+ (res != expect) ? " -- FAILED" : "");
+ if (res != expect)
+ return 1;
+
+ return 0;
+}
+
+int test_main(void)
+{
+ char buf[SIZE];
+ int rofd, wofd, rwfd;
+ int status = 0, res;
+
+ memset(buf, 0, SIZE);
+
+ rofd = open("testdir/rofile", O_RDONLY); assert(rofd != -1);
+ wofd = open("testdir/wofile", O_WRONLY); assert(wofd != -1);
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd != -1);
+
+ status |= attempt(rofd, buf, SIZE, 0, WRITE, -EBADF);
+ status |= attempt(wofd, buf, SIZE, 0, READ, -EBADF);
+ status |= attempt(rwfd, buf, SIZE, 0, WRITE, SIZE);
+ status |= attempt(rwfd, buf, SIZE, 0, READ, SIZE);
+ status |= attempt(rwfd, buf, SIZE, -1, READ, -EINVAL);
+ status |= attempt(rwfd, buf, SIZE, -1, WRITE, -EINVAL);
+
+ rwfd = open("testdir/rwfile", O_RDWR|O_APPEND); assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ status |= attempt(rwfd, buf, SIZE, 0, READ, 0);
+ status |= attempt(rwfd, "1234", 4, 0, WRITE, 4);
+ status |= attempt(rwfd, "5678", 4, 0, WRITE, 4);
+ memset(buf, 0, SIZE);
+ status |= attempt(rwfd, buf, SIZE, 0, READ, 8);
+ printf("read after append: [%s]\n", buf);
+ assert(memcmp(buf, "12345678", 8) == 0);
+
+ status |= attempt(rwfd, KERNEL_RW_POINTER, SIZE, 0, READ, -EFAULT);
+ status |= attempt(rwfd, KERNEL_RW_POINTER, SIZE, 0, WRITE, -EFAULT);
+
+ /* Some architectures map the 0 page. Ugh. */
+#if !defined(__ia64__)
+ status |= attempt(rwfd, NULL, SIZE, 0, WRITE, -EFAULT);
+#endif
+
+ return status;
+}
+
--- /dev/null
+/* 5.t
+- Write from a mmap() of the same file. (5.t)
+*/
+#include "aio_setup.h"
+#include <sys/mman.h>
+#include <errno.h>
+
+int test_main(void)
+{
+ int page_size = getpagesize();
+#define SIZE 512
+ char *buf;
+ int rwfd;
+ int status = 0, res;
+
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd != -1);
+ res = ftruncate(rwfd, 512); assert(res == 0);
+
+ buf = mmap(0, page_size, PROT_READ|PROT_WRITE, MAP_SHARED, rwfd, 0);
+ assert(buf != (char *)-1);
+
+ status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, 0, READ, SIZE);
+
+ res = munmap(buf, page_size); assert(res == 0);
+ buf = mmap(0, page_size, PROT_READ|PROT_WRITE, MAP_SHARED, rwfd, 0);
+ assert(buf != (char *)-1);
+
+ status |= attempt_rw(rwfd, buf, SIZE, 0, READ, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE, SIZE);
+
+ res = munmap(buf, page_size); assert(res == 0);
+ buf = mmap(0, page_size, PROT_READ, MAP_SHARED, rwfd, 0);
+ assert(buf != (char *)-1);
+
+ status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, 0, READ, -EFAULT);
+
+ res = munmap(buf, page_size); assert(res == 0);
+ buf = mmap(0, page_size, PROT_WRITE, MAP_SHARED, rwfd, 0);
+ assert(buf != (char *)-1);
+
+ /* Whether PROT_WRITE is readable is arch-dependent. So compare
+ * against write() result (to make the kernel read from buf). */
+ res = write(rwfd, buf, SIZE);
+ if (res < 0)
+ res = -errno;
+ status |= attempt_rw(rwfd, buf, SIZE, 0, READ, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE, res);
+
+ return status;
+}
+
--- /dev/null
+/* 6.t
+- huge reads (pinned pages) (6.t)
+- huge writes (6.t)
+*/
+#include "aio_setup.h"
+#include <sys/mman.h>
+
+long getmemsize(void)
+{
+ FILE *f = fopen("/proc/meminfo", "r");
+ long size;
+ int gotit = 0;
+ char str[256];
+
+ assert(f != NULL);
+ while (NULL != fgets(str, 255, f)) {
+ str[255] = 0;
+ if (0 == memcmp(str, "MemTotal:", 9)) {
+ if (1 == sscanf(str + 9, "%ld", &size)) {
+ gotit = 1;
+ break;
+ }
+ }
+ }
+ fclose(f);
+
+ assert(gotit != 0);
+ return size;
+}
+
+int test_main(void)
+{
+ char *buf;
+ int rwfd;
+ int status = 0, res;
+ long size;
+
+ size = getmemsize();
+ printf("size = %ld\n", size);
+ assert(size >= (16 * 1024));
+ if (size > (768 * 1024))
+ size = 768 * 1024;
+ size *= 1024;
+
+ rwfd = open("testdir/rwfile", O_RDWR); assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ buf = malloc(size); assert(buf != NULL);
+
+ //memset(buf, 0, size);
+ status |= attempt_rw(rwfd, buf, size, 0, WRITE, size);
+ status |= attempt_rw(rwfd, buf, size, 0, READ, size);
+
+ //res = ftruncate(rwfd, 0); assert(res == 0);
+
+ return status;
+}
+
--- /dev/null
+/* 7.t
+- Write overlapping the file size rlimit boundary: should be a short
+ write. (7.t)
+- Write at the file size rlimit boundary: should give EFBIG. (I think
+ the spec requires that you do NOT deliver SIGXFSZ in this case, where
+ you would do so for sync IO.) (7.t)
+- Special case: a write of zero bytes at or beyond the file size rlimit
+ boundary must return success. (7.t)
+*/
+
+#include <sys/resource.h>
+#include <signal.h>
+
+void SET_RLIMIT(long long limit)
+{
+ struct rlimit rlim;
+ int res;
+
+ /* Seems that we do send SIGXFSZ, but hard to fix... */
+ signal(SIGXFSZ, SIG_IGN);
+ rlim.rlim_cur = limit; assert(rlim.rlim_cur == limit);
+ rlim.rlim_max = limit; assert(rlim.rlim_max == limit);
+
+ res = setrlimit(RLIMIT_FSIZE, &rlim); assert(res == 0);
+}
+
+#define LIMIT 8192
+#define FILENAME "testdir/rwfile"
+
+#include "common-7-8.h"
--- /dev/null
+/* 8.t
+- Ditto for the above three tests at the offset maximum (largest
+ possible ext2/3 file size.) (8.t)
+ */
+#include <sys/types.h>
+#include <unistd.h>
+
+long long get_fs_limit(int fd)
+{
+ long long min = 0, max = 9223372036854775807LL;
+ char c = 0;
+ int ret;
+
+ while (max - min > 1) {
+ if (pwrite64(fd, &c, 1, (min + max) / 2) == -1)
+ max = (min + max) / 2;
+ else {
+ ret = ftruncate(fd, 0);
+ assert(ret == 0);
+ min = (min + max) / 2;
+ }
+ }
+ return max;
+}
+
+#define SET_RLIMIT(x) do ; while (0)
+#define LIMIT get_fs_limit(rwfd)
+#define FILENAME "testdir.ext2/rwfile"
+
+#include "common-7-8.h"
--- /dev/null
+#include <time.h>
+io_context_t io_ctx;
+#define BAD_CTX ((io_context_t)-1)
+
+void aio_setup(int n)
+{
+ int res = io_queue_init(n, &io_ctx);
+ if (res != 0) {
+ printf("io_queue_setup(%d) returned %d (%s)\n",
+ n, res, strerror(-res));
+ exit(3);
+ }
+}
+
+int attempt_io_submit(io_context_t ctx, long nr, struct iocb *ios[], int expect)
+{
+ int res;
+
+ printf("expect %3d: io_submit(%10p, %3ld, %10p) = ", expect, ctx, nr, ios);
+ fflush(stdout);
+ res = io_submit(ctx, nr, ios);
+ printf("%3d [%s]%s\n", res, (res <= 0) ? strerror(-res) : "",
+ (res != expect) ? " -- FAILED" : "");
+ if (res != expect)
+ return 1;
+
+ return 0;
+}
+
+int sync_submit(struct iocb *iocb)
+{
+ struct io_event event;
+ struct iocb *iocbs[] = { iocb };
+ int res;
+
+ /* 30 second timeout should be enough */
+ struct timespec ts;
+ ts.tv_sec = 30;
+ ts.tv_nsec = 0;
+
+ res = io_submit(io_ctx, 1, iocbs);
+ if (res != 1) {
+ printf("sync_submit: io_submit res=%d [%s]\n", res, strerror(-res));
+ return res;
+ }
+
+ res = io_getevents(io_ctx, 0, 1, &event, &ts);
+ if (res != 1) {
+ printf("sync_submit: io_getevents res=%d [%s]\n", res, strerror(-res));
+ return res;
+ }
+ return event.res;
+}
+
+#define SETUP aio_setup(1024)
+
+
+#define READ 'r'
+#define WRITE 'w'
+#define READ_SILENT 'R'
+#define WRITE_SILENT 'W'
+#define READV '<'
+#define WRITEV '>'
+
+int attempt_rw(int fd, void *buf, int count, long long pos, int rw, int expect)
+{
+ struct iocb iocb;
+ int res;
+ int silent = 0;
+
+ switch(rw) {
+ case READ_SILENT:
+ silent = 1;
+ case READ:
+ io_prep_pread (&iocb, fd, buf, count, pos);
+ break;
+ case WRITE_SILENT:
+ silent = 1;
+ case WRITE:
+ io_prep_pwrite(&iocb, fd, buf, count, pos);
+ break;
+ case WRITEV:
+ io_prep_pwritev(&iocb, fd, buf, count, pos);
+ break;
+ case READV:
+ io_prep_preadv(&iocb, fd, buf, count, pos);
+ break;
+ }
+
+ if (!silent) {
+ printf("expect %5d: (%c), res = ", expect, rw);
+ fflush(stdout);
+ }
+ res = sync_submit(&iocb);
+ if (!silent || res != expect) {
+ if (silent)
+ printf("expect %5d: (%c), res = ", expect, rw);
+ printf("%5d [%s]%s\n", res,
+ (res <= 0) ? strerror(-res) : "Success",
+ (res != expect) ? " -- FAILED" : "");
+ }
+
+ if (res != expect)
+ return 1;
+
+ return 0;
+}
+
--- /dev/null
+/* common-7-8.h
+*/
+#include "aio_setup.h"
+
+#include <errno.h>
+#include <unistd.h>
+
+#define SIZE 512
+
+int test_main(void)
+{
+ char *buf;
+ int rwfd;
+ int status = 0, res;
+ long long limit;
+
+ rwfd = open(FILENAME, O_RDWR|O_CREAT, 0600); assert(rwfd != -1);
+ res = ftruncate(rwfd, 0); assert(res == 0);
+ buf = malloc(SIZE); assert(buf != NULL);
+ memset(buf, 0, SIZE);
+
+ limit = LIMIT;
+
+ SET_RLIMIT(limit);
+
+ status |= attempt_rw(rwfd, buf, SIZE, limit-SIZE, WRITE, SIZE);
+ status |= attempt_rw(rwfd, buf, SIZE, limit-SIZE, READ, SIZE);
+
+ status |= attempt_rw(rwfd, buf, SIZE, 1+limit-SIZE, WRITE, SIZE-1);
+ status |= attempt_rw(rwfd, buf, SIZE, 1+limit-SIZE, READ, SIZE-1);
+
+ status |= attempt_rw(rwfd, buf, SIZE, limit, WRITE, -EFBIG);
+ status |= attempt_rw(rwfd, buf, SIZE, limit, READ, 0);
+ status |= attempt_rw(rwfd, buf, 0, limit, WRITE, 0);
+
+ return status;
+}
+
--- /dev/null
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdlib.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+#include <libaio.h>
+
+#if __LP64__ == 0
+#define KERNEL_RW_POINTER ((void *)0xc0010000)
+#else
+//#warning Not really sure where kernel memory is. Guessing.
+#define KERNEL_RW_POINTER ((void *)0xffffffff81000000)
+#endif
+
+
+char test_name[] = TEST_NAME;
+
+#include TEST_NAME
+
+int main(void)
+{
+ int res;
+ const char *test_result;
+
+#if defined(SETUP)
+ SETUP;
+#endif
+
+ res = test_main();
+ switch(res) {
+ case 0:
+ test_result = "PASSED";
+ break;
+ case 3:
+ test_result = "SKIPPED";
+ break;
+ default:
+ test_result = "FAILED";
+ res = 1;
+ break;
+ }
+
+ printf("test %s completed %s.\n", test_name, test_result);
+ fflush(stdout);
+ return res;
+}
--- /dev/null
+#!/bin/bash
+
+passes=0
+fails=0
+skips=0
+
+echo "Test run starting at" `date`
+
+while [ $# -ge 1 ] ; do
+ this_test=$1
+ shift
+ echo "Starting $this_test"
+ $this_test 2>&1
+ res=$?
+ if [ $res -eq 0 ]; then
+ str="";
+ passes=$((passes + 1));
+ elif [ $res -eq 3 ]; then
+ str=" -- SKIPPED";
+ skips=$((skips + 1));
+ else
+ str=" -- FAILED"
+ fails=$((fails + 1));
+ fi
+ echo "Completed $this_test with $res$str".
+done
+
+echo "Pass: $passes Fail: $fails Skip: $skips"
+echo "Test run complete at" `date`
+
+exit $fails
--- /dev/null
+Name: libaio
+Version: 0.3.113
+Release: 1
+Summary: Linux-native asynchronous I/O access library
+License: LGPL
+Group: System Environment/Libraries
+Source: %{name}-%{version}.tar.gz
+BuildRoot: %{_tmppath}/%{name}-root
+# Fix ExclusiveArch as we implement this functionality on more architectures
+ExclusiveArch: i386 x86_64 ia64 s390 s390x ppc ppc64 ppc64pseries ppc64iseries alpha alphaev6 %{arm} loongarch64
+
+%description
+The Linux-native asynchronous I/O facility ("async I/O", or "aio") has a
+richer API and capability set than the simple POSIX async I/O facility.
+This library, libaio, provides the Linux-native API for async I/O.
+The POSIX async I/O facility requires this library in order to provide
+kernel-accelerated async I/O capabilities, as do applications which
+require the Linux-native async I/O API.
+
+%package devel
+Summary: Development files for Linux-native asynchronous I/O access
+Group: Development/System
+Requires: libaio
+Provides: libaio.so.1
+
+%description devel
+This package provides header files to include and libraries to link with
+for the Linux-native asynchronous I/O facility ("async I/O", or "aio").
+
+%prep
+%setup
+
+%build
+make
+
+%install
+[ "$RPM_BUILD_ROOT" != "/" ] && rm -rf $RPM_BUILD_ROOT
+
+make install DESTDIR=$RPM_BUILD_ROOT prefix=/usr libdir=/%{_libdir}
+
+%clean
+[ "$RPM_BUILD_ROOT" != "/" ] && rm -rf $RPM_BUILD_ROOT
+
+%post -p /sbin/ldconfig
+
+%postun -p /sbin/ldconfig
+
+%files
+%defattr(-,root,root)
+%attr(0755,root,root) %{_libdir}/libaio.so.*
+%doc COPYING TODO
+
+%files devel
+%defattr(-,root,root)
+%attr(0644,root,root) %{_includedir}/*
+%attr(0755,root,root) %{_libdir}/libaio.so
+%attr(0644,root,root) %{_libdir}/libaio.a
+
+%changelog
+* Mon Mar 28 2022 Jeff Moyer <jmoyer@redhat.com> - 0.3.113-1
+- Fix padding in vector and sockaddr structures on 32 bit (Brett Holman)
+- Add a test case for missed poll events (Eric Biggers)
+- Add loongarch support (Yehui Ren)
+- man page cleanup and fixes (Guillem Jover)
+- assorted test harness fixess (Guillem Jover, Jeff Moyer)
+
+* Mon Oct 22 2018 Jeff Moyer <jmoyer@redhat.com> - 0.3.112-1
+- Add async poll support (Christoph Hellwig)
+- Use canonical DESTDIR= environment variable (Thomas Petazzoni)
+- Add ability to disable building the shared library (Thomas Petazzoni)
+
+* Tue Mar 6 2018 Jeff Moyer <jmoyer@redhat.com> - 0.3.111-1
+- Add two new tests to the test harness (Jeff Moyer)
+- Generic arch dectection for padding defines (Nathan Rossi)
+- harness: don't hardcode page size (Jeff Moyer)
+- harness: add a test case for mremap (Jeff Moyer)
+- libaio: harness: fix build errors due to attribute warn_unused_result (Mauricio Faria de Oliveira)
+- libaio: harness: fix build error due to linker search order (Mauricio Faria de Oliveira)
+- harness: add test for allocating aio-max-nr ioctxs (Jeff Moyer)
+- Add support for preadv2/pwritev2 (Jeff Moyer)
+- syscall-generic: don't overwrite errno (Jeff Moyer)
+- syscall: get rid of custom syscall implementation (Jeff Moyer)
+- Change syscall-arm64.h to syscall-generic.h (Icenowy Zheng)
+- Use generic syscall number schema for RISC-V (Icenowy Zheng)
+- Add endian detection (LE) and bit width detection (32/64) for RISC-V (Icenowy Zheng)
+- Makefile: convert tag and archive targets to git (Jeff Moyer)
+
+* Fri Jul 5 2013 Jeff Moyer <jmoyer@redhat.com> - 0.3.110-1
+- Add suport for sparc and arm64 (Mike Frysinger and Jeff Moyer)
+- Add generic syscall fallbacks (Mike Frysinger)
+- Update man pages (Jeff Moyer and Cyril Hrubis)
+- Build system fixes (Mike Frysinger)
+
+* Tue Jun 9 2009 Jeff Moyer <jmoyer@redhat.com> - 0.3.108-1
+- add ARM architecture support (grabbed from Debian arches tree)
+- replace check of __i386__ with __LP64__ in test harness (Jeff Moyer)
+- change the .spec's copyright tag to license
+
+* Wed Jan 9 2008 Jeff Moyer <jmoyer@redhat.com> - 0.3.107-1
+- Fix the test harness (Rusty Russell)
+- Add eventfd support (Rusty Russell)
+
+* Tue Jan 3 2006 Jeff Moyer <jmoyer@redhat.com> - 0.3.106-1
+- Add a .proc directive for the ia64_aio_raw_syscall macro. This sounds a lot
+ like the previous entry, but that one fixed the __ia64_raw_syscall macro,
+ located in syscall-ia64.h. This macro is in raw_syscall.c, which pretty much
+ only exists for ia64. This bug prevented the package from building with
+ newer version of gcc.
+
+* Mon Aug 1 2005 Jeff Moyer <jmoyer@redhat.com> - 0.3.105-1
+- Add a .proc directive for the ia64 raw syscall macro.
+
+* Fri Apr 1 2005 Jeff Moyer <jmoyer@redhat.com> - 0.3.104-1
+- Add Alpha architecture support. (Sergey Tikhonov <tsv@solvo.ru>)
+
+* Tue Jan 25 2005 Jeff Moyer <jmoyer@redhat.com> - 0.3.103-1
+- Fix SONAME breakage. In changing file names around, I also changed the
+ SONAME, which is a no no.
+
+* Thu Oct 14 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.102-1
+- S390 asm had a bug; I forgot to update the clobber list. Lucky for me,
+ newer compilers complain about such things.
+- Also update the s390 asm to look more like the new kernel variants.
+
+* Wed Oct 13 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.101-1
+- Revert syscall return values to be -ERRNO. This was an inadvertant bug
+ introduced when clobber lists changed.
+- add ppc64pseries and ppc64iseries to exclusivearch
+
+* Tue Sep 14 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.100-1
+- Switch around the tests for _PPC_ and _powerpc64_ so that the ppc64
+ platforms get the right padding.
+
+* Wed Jul 14 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.99-4
+- Ok, there was a race in moving the cvs module. Someone rebuild from
+ the old cvs into fc3. *sigh* bumping rev.
+
+* Wed Jul 14 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.99-3
+- Actually provide libaio.so.1.
+
+* Tue Mar 30 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.99-2
+- Apparently the 0.3.93 patch was not meant for 0.3.96. Backed it out.
+
+* Tue Mar 30 2004 Jeff Moyer <jmoyer@redhat.com> - 0.3.99-1
+- Fix compat calls.
+- make library .so.1.0.0 and make symlinks properly.
+- Fix header file for inclusion in c++ code.
+
+* Thu Feb 26 2004 Jeff Moyer <jmoyer@redhat.com> 0.3.98-2
+- bah. fix version nr in changelog.
+
+* Thu Feb 26 2004 Jeff Moyer <jmoyer@redhat.com> 0.3.98-1
+- fix compiler warnings.
+
+* Thu Feb 26 2004 Jeff Moyer <jmoyer@redhat.com> 0.3.97-2
+- make srpm was using rpm to do a build. changed that to use rpmbuild if
+ it exists, and fallback to rpm if it doesn't.
+
+* Tue Feb 24 2004 Jeff Moyer <jmoyer@redhat.com> 0.3.97-1
+- Use libc syscall(2) instead of rolling our own calling mechanism. This
+ change is inspired due to a failure to build with newer gcc, since clobber
+ lists were wrong.
+- Add -fpic to the CFLAGS for all architectures. Should address bz #109457.
+- change a #include from <linux/types.h> to <sys/types.h>. Fixes a build
+ issue on s390.
+
+* Wed Jul 7 2003 Bill Nottingham <notting@redhat.com> 0.3.96-3
+- fix paths on lib64 arches
+
+* Wed Jun 18 2003 Michael K. Johnson <johnsonm@redhat.com> 0.3.96-2
+- optimization in io_getevents from Arjan van de Ven in 0.3.96-1
+- deal with ia64 in 0.3.96-2
+
+* Wed May 28 2003 Michael K. Johnson <johnsonm@redhat.com> 0.3.95-1
+- ppc bugfix from Julie DeWandel
+
+* Tue May 20 2003 Michael K. Johnson <johnsonm@redhat.com> 0.3.94-1
+- symbol versioning fix from Ulrich Drepper
+
+* Mon Jan 27 2003 Benjamin LaHaise <bcrl@redhat.com>
+- bump to 0.3.93-3 for rebuild.
+
+* Mon Dec 16 2002 Benjamin LaHaise <bcrl@redhat.com>
+- libaio 0.3.93 test release
+- add powerpc support from Gianni Tedesco <gianni@ecsc.co.uk>
+- add s/390 support from Arnd Bergmann <arnd@bergmann-dalldorf.de>
+
+* Fri Sep 12 2002 Benjamin LaHaise <bcrl@redhat.com>
+- libaio 0.3.92 test release
+- build on x86-64
+
+* Thu Sep 12 2002 Benjamin LaHaise <bcrl@redhat.com>
+- libaio 0.3.91 test release
+- build on ia64
+- remove libredhat-kernel from the .spec file
+
+* Thu Sep 5 2002 Benjamin LaHaise <bcrl@redhat.com>
+- libaio 0.3.90 test release
+
+* Mon Apr 29 2002 Benjamin LaHaise <bcrl@redhat.com>
+- add requires initscripts >= 6.47-1 to get boot time libredhat-kernel
+ linkage correct.
+- typo fix
+
+* Thu Apr 25 2002 Benjamin LaHaise <bcrl@redhat.com>
+- make /usr/lib/libredhat-kernel.so point to /lib/libredhat-kernel.so.1.0.0
+
+* Mon Apr 15 2002 Tim Powers <timp@redhat.com>
+- make the post scriptlet not use /bin/sh
+
+* Sat Apr 12 2002 Benjamin LaHaise <bcrl@redhat.com>
+- add /lib/libredhat-kernel* to %files.
+
+* Fri Apr 12 2002 Benjamin LaHaise <bcrl@redhat.com>
+- make the dummy install as /lib/libredhat-kernel.so.1.0.0 so
+ that ldconfig will link against it if no other is installed.
+
+* Tue Jan 22 2002 Benjamin LaHaise <bcrl@redhat.com>
+- add io_getevents
+
+* Tue Jan 22 2002 Michael K. Johnson <johnsonm@redhat.com>
+- Make linker happy with /usr/lib symlink for libredhat-kernel.so
+
+* Mon Jan 21 2002 Michael K. Johnson <johnsonm@redhat.com>
+- Added stub library
+
+* Sun Jan 20 2002 Michael K. Johnson <johnsonm@redhat.com>
+- Initial packaging
--- /dev/null
+.TH io 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io \- Asynchronous IO
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.sp
+.br
+.B #include <libaio.h>
+.sp
+.fi
+.SH DESCRIPTION
+The
+.B libaio
+library defines a new set of I/O operations which can
+significantly reduce the time an application spends waiting at I/O. The
+new functions allow a program to initiate one or more I/O operations and
+then immediately resume normal work while the I/O operations are
+executed in parallel.
+
+These functions are part of the library with realtime functions named
+.BR libaio .
+They are not actually part of the
+.B libc
+binary.
+The implementation of these functions can be done using support in the
+kernel.
+
+All IO operations operate on files which were opened previously. There
+might be arbitrarily many operations running for one file. The
+asynchronous I/O operations are controlled using a data structure named
+.B struct iocb
+It is defined in
+.I libaio.h
+as follows.
+
+.nf
+
+typedef struct io_context *io_context_t;
+
+typedef enum io_iocb_cmd {
+ IO_CMD_PREAD = 0,
+ IO_CMD_PWRITE = 1,
+
+ IO_CMD_FSYNC = 2,
+ IO_CMD_FDSYNC = 3,
+
+ IO_CMD_POLL = 5,
+ IO_CMD_NOOP = 6,
+} io_iocb_cmd_t;
+
+struct io_iocb_common {
+ void *buf;
+ unsigned __pad1;
+ long nbytes;
+ unsigned __pad2;
+ long long offset;
+ long long __pad3, __pad4;
+}; /* result code is the amount read or -'ve errno */
+
+
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+ union {
+ struct io_iocb_common c;
+ struct io_iocb_vector v;
+ struct io_iocb_poll poll;
+ struct io_iocb_sockaddr saddr;
+ } u;
+};
+
+
+.fi
+.TP
+.BI int " aio_fildes"
+This element specifies the file descriptor to be used for the
+operation. It must be a legal descriptor, otherwise the operation will
+fail.
+
+The device on which the file is opened must allow the seek operation.
+I.e., it is not possible to use any of the IO operations on devices
+like terminals where an
+.BR lseek (2)
+call would lead to an error.
+.TP
+.BI long " u.c.offset"
+This element specifies the offset in the file at which the operation (input
+or output) is performed. Since the operations are carried out in arbitrary
+order and more than one operation for one file descriptor can be
+started, one cannot expect a current read/write position of the file
+descriptor.
+.TP
+.BI "void *" buf
+This is a pointer to the buffer with the data to be written or the place
+where the read data is stored.
+.TP
+.BI long " u.c.nbytes"
+This element specifies the length of the buffer pointed to by
+.IR io_buf .
+.TP
+.BI int " aio_reqprio"
+Is not currently used.
+.TP
+.B "IO_CMD_PREAD"
+Start a read operation. Read from the file at position
+.I u.c.offset
+and store the next
+.I u.c.nbytes
+bytes in the
+buffer pointed to by
+.IR buf .
+.TP
+.B "IO_CMD_PWRITE"
+Start a write operation. Write
+.I u.c.nbytes
+bytes starting at
+.I buf
+into the file starting at position
+.IR u.c.offset .
+.TP
+.B "IO_CMD_NOP"
+Do nothing for this control block. This value is useful sometimes when
+an array of
+.B struct iocb
+values contains holes, i.e., some of the
+values must not be handled although the whole array is presented to the
+.BR io_submit (3)
+function.
+.TP
+.B "IO_CMD_FSYNC"
+.TQ
+.B "IO_CMD_POLL"
+This is experimental.
+.SH EXAMPLE
+.nf
+/*
+ * Simplistic version of copy command using async i/o
+ *
+ * From: Stephen Hemminger <shemminger@osdl.org>
+ * Copy file by using a async I/O state machine.
+ * 1. Start read request
+ * 2. When read completes turn it into a write request
+ * 3. When write completes decrement counter and free resources
+ *
+ *
+ * Usage: aiocp file(s) desination
+ */
+
+#include <unistd.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/param.h>
+#include <fcntl.h>
+#include <errno.h>
+
+#include <libaio.h>
+
+#define AIO_BLKSIZE (64*1024)
+#define AIO_MAXIO 32
+
+static int busy = 0; // # of I/O's in flight
+static int tocopy = 0; // # of blocks left to copy
+static int dstfd = -1; // destination file descriptor
+static const char *dstname = NULL;
+static const char *srcname = NULL;
+
+
+/* Fatal error handler */
+static void io_error(const char *func, int rc)
+{
+ if (rc == -ENOSYS)
+ fprintf(stderr, "AIO not in this kernel\\n");
+ else if (rc < 0 && -rc < sys_nerr)
+ fprintf(stderr, "%s: %s\\n", func, sys_errlist[-rc]);
+ else
+ fprintf(stderr, "%s: error %d\\n", func, rc);
+
+ if (dstfd > 0)
+ close(dstfd);
+ if (dstname)
+ unlink(dstname);
+ exit(1);
+}
+
+/*
+ * Write complete callback.
+ * Adjust counts and free resources
+ */
+static void wr_done(io_context_t ctx, struct iocb *iocb, long res, long res2)
+{
+ if (res2 != 0) {
+ io_error("aio write", res2);
+ }
+ if (res != iocb->u.c.nbytes) {
+ fprintf(stderr, "write missed bytes expect %d got %d\\n", iocb->u.c.nbytes, res2);
+ exit(1);
+ }
+ --tocopy;
+ --busy;
+ free(iocb->u.c.buf);
+
+ memset(iocb, 0xff, sizeof(iocb)); // paranoia
+ free(iocb);
+ write(2, "w", 1);
+}
+
+/*
+ * Read complete callback.
+ * Change read iocb into a write iocb and start it.
+ */
+static void rd_done(io_context_t ctx, struct iocb *iocb, long res, long res2)
+{
+ /* library needs accessors to look at iocb? */
+ int iosize = iocb->u.c.nbytes;
+ char *buf = iocb->u.c.buf;
+ off_t offset = iocb->u.c.offset;
+
+ if (res2 != 0)
+ io_error("aio read", res2);
+ if (res != iosize) {
+ fprintf(stderr, "read missing bytes expect %d got %d\\n", iocb->u.c.nbytes, res);
+ exit(1);
+ }
+
+
+ /* turn read into write */
+ io_prep_pwrite(iocb, dstfd, buf, iosize, offset);
+ io_set_callback(iocb, wr_done);
+ if (1 != (res = io_submit(ctx, 1, &iocb)))
+ io_error("io_submit write", res);
+ write(2, "r", 1);
+}
+
+
+int main(int argc, char *const *argv)
+{
+ int srcfd;
+ struct stat st;
+ off_t length = 0, offset = 0;
+ io_context_t myctx;
+
+ if (argc != 3 || argv[1][0] == '-') {
+ fprintf(stderr, "Usage: aiocp SOURCE DEST");
+ exit(1);
+ }
+ if ((srcfd = open(srcname = argv[1], O_RDONLY)) < 0) {
+ perror(srcname);
+ exit(1);
+ }
+ if (fstat(srcfd, &st) < 0) {
+ perror("fstat");
+ exit(1);
+ }
+ length = st.st_size;
+
+ if ((dstfd = open(dstname = argv[2], O_WRONLY | O_CREAT, 0666)) < 0) {
+ close(srcfd);
+ perror(dstname);
+ exit(1);
+ }
+
+ /* initialize state machine */
+ memset(&myctx, 0, sizeof(myctx));
+ io_queue_init(AIO_MAXIO, &myctx);
+ tocopy = howmany(length, AIO_BLKSIZE);
+
+ while (tocopy > 0) {
+ int i, rc;
+ /* Submit as many reads as once as possible upto AIO_MAXIO */
+ int n = MIN(MIN(AIO_MAXIO - busy, AIO_MAXIO / 2),
+ howmany(length - offset, AIO_BLKSIZE));
+ if (n > 0) {
+ struct iocb *ioq[n];
+
+ for (i = 0; i < n; i++) {
+ struct iocb *io = (struct iocb *) malloc(sizeof(struct iocb));
+ int iosize = MIN(length - offset, AIO_BLKSIZE);
+ char *buf = (char *) malloc(iosize);
+
+ if (NULL == buf || NULL == io) {
+ fprintf(stderr, "out of memory\\n");
+ exit(1);
+ }
+
+ io_prep_pread(io, srcfd, buf, iosize, offset);
+ io_set_callback(io, rd_done);
+ ioq[i] = io;
+ offset += iosize;
+ }
+
+ rc = io_submit(myctx, n, ioq);
+ if (rc < 0)
+ io_error("io_submit", rc);
+
+ busy += n;
+ }
+
+ // Handle IO's that have completed
+ rc = io_queue_run(myctx);
+ if (rc < 0)
+ io_error("io_queue_run", rc);
+
+ // if we have maximum number of i/o's in flight
+ // then wait for one to complete
+ if (busy == AIO_MAXIO) {
+ rc = io_queue_wait(myctx, NULL);
+ if (rc < 0)
+ io_error("io_queue_wait", rc);
+ }
+
+ }
+
+ close(srcfd);
+ close(dstfd);
+ exit(0);
+}
+
+/*
+ * Results look like:
+ * [alanm@toolbox ~/MOT3]$ ../taio kernel-source-2.4.8-0.4g.ppc.rpm abc
+ * rrrrrrrrrrrrrrrwwwrwrrwwrrwrwwrrwrwrwwrrwrwrrrrwwrwwwrrwrrrwwwwwwwwwwwwwwwww
+ * rrrrrrrrrrrrrrwwwrrwrwrwrwrrwwwwwwwwwwwwwwrrrrrrrrrrrrrrrrrrwwwwrwrwwrwrwrwr
+ * wrrrrrrrwwwwwwwwwwwwwrrrwrrrwrrwrwwwwwwwwwwrrrrwwrwrrrrrrrrrrrwwwwwwwwwwwrww
+ * wwwrrrrrrrrwwrrrwwrwrwrwwwrrrrrrrwwwrrwwwrrwrwwwwwwwwrrrrrrrwwwrrrrrrrwwwwww
+ * wwwwwwwrwrrrrrrrrwrrwrrwrrwrwrrrwrrrwrrrwrwwwwwwwwwwwwwwwwwwrrrwwwrrrrrrrrrr
+ * rrwrrrrrrwrrwwwwwwwwwwwwwwwwrwwwrrwrwwrrrrrrrrrrrrrrrrrrrwwwwwwwwwwwwwwwwwww
+ * rrrrrwrrwrwrwrrwrrrwwwwwwwwrrrrwrrrwrwwrwrrrwrrwrrrrwwwwwwwrwrwwwwrwwrrrwrrr
+ * rrrwwwwwwwrrrrwwrrrrrrrrrrrrwrwrrrrwwwwwwwwwwwwwwrwrrrrwwwwrwrrrrwrwwwrrrwww
+ * rwwrrrrrrrwrrrrrrrrrrrrwwwwrrrwwwrwrrwwwwwwwwwwwwwwwwwwwwwrrrrrrrwwwwwwwrw
+ */
+.fi
+.SH "SEE ALSO"
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.TH io_cancel 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_cancel \- Cancel io requests
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.sp
+.br
+.B #include <libaio.h>
+.sp
+.br
+.BI "int io_cancel(io_context_t " ctx ", struct iocb *" iocb ");"
+.br
+.sp
+struct iocb {
+ void *data; /* Return in the io completion event */
+ unsigned key; /* For use in identifying io requests */
+ short aio_lio_opcode;
+ short aio_reqprio; /* Not used */
+ int aio_fildes;
+};
+.fi
+.SH DESCRIPTION
+Attempts to cancel an
+.I iocb
+previously passed to
+.BR io_submit (3).
+If the operation is successfully cancelled, the resulting event is
+copied into the memory pointed to by result without being placed
+into the completion queue.
+.PP
+When one or more requests are asynchronously processed, it might be
+useful in some situations to cancel a selected operation, e.g., if it
+becomes obvious that the written data is no longer accurate and would
+have to be overwritten soon. As an example, assume an application, which
+writes data in files in a situation where new incoming data would have
+to be written in a file which will be updated by an enqueued request.
+.SH "RETURN VALUES"
+\fI0\fP is returned on success, otherwise returns \fIerrno\fP.
+.SH ERRORS
+.TP
+.B EFAULT
+If any of the data structures pointed to are invalid.
+.TP
+.B EINVAL
+If
+.I aio_context
+specified by
+.I ctx
+is invalid.
+.TP
+.B EAGAIN
+If the
+.I iocb
+specified was not cancelled.
+.TP
+.B ENOSYS
+If not implemented.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\" static inline int io_fsync(io_context_t ctx, struct iocb *iocb, io_callback_t cb, int fd)
+.\" {
+.\" io_prep_fsync(iocb, fd);
+.\" io_set_callback(iocb, cb);
+.\" return io_submit(ctx, 1, &iocb);
+.\" }
+.TH io_fsync 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_fsync \- Synchronize a file's complete in-core state with that on disk
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.sp
+.br
+.B #include <libaio.h>
+.sp
+.br
+.BI "int io_fsync(io_context_t " ctx ", struct iocb *" iocb ", io_callback_t " cb ", int " fd ");"
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.sp
+typedef void (*io_callback_t)(io_context_t ctx, struct iocb *iocb, long res, long res2);
+.sp
+.fi
+.SH DESCRIPTION
+When dealing with asynchronous operations it is sometimes necessary to
+get into a consistent state. This would mean for AIO that one wants to
+know whether a certain request or a group of request were processed.
+This could be done by waiting for the notification sent by the system
+after the operation terminated, but this sometimes would mean wasting
+resources (mainly computation time).
+.PP
+Calling this function forces all I/O operations operating queued at the
+time of the function call operating on the file descriptor
+.IR "iocb->io_fildes"
+into the synchronized I/O completion state. The
+.BR io_fsync ()
+function returns
+immediately but the notification through the method described in
+.IR "io_callback"
+will happen only after all requests for this
+file descriptor have terminated and the file is synchronized. This also
+means that requests for this very same file descriptor which are queued
+after the synchronization request are not affected.
+.SH "RETURN VALUES"
+Returns
+.BR 0 ,
+otherwise returns
+.IR errno .
+.SH ERRORS
+.TP
+.B EFAULT
+.I iocbs
+referenced data outside of the program's accessible address space.
+.TP
+.B EINVAL
+.I ctx
+refers to an uninitialized aio context, the iocb pointed to by
+.I iocbs
+contains an improperly initialized iocb,
+.TP
+.B EBADF
+The iocb contains a file descriptor that does not exist.
+.TP
+.B EINVAL
+The file specified in the iocb does not support the given io operation.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_getevents (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\"/* io_getevents:
+.\" * Attempts to read at least min_nr events and up to nr events from
+.\" * the completion queue for the aio_context specified by ctx. May
+.\" * fail with -EINVAL if ctx is invalid, if min_nr is out of range,
+.\" * if nr is out of range, if when is out of range. May fail with
+.\" * -EFAULT if any of the memory specified to is invalid. May return
+.\" * 0 or < min_nr if no events are available and the timeout specified
+.\" * by when has elapsed, where when == NULL specifies an infinite
+.\" * timeout. Note that the timeout pointed to by when is relative and
+.\" * will be updated if not NULL and the operation blocks. Will fail
+.\" * with -ENOSYS if not implemented.
+.\" */
+.\"asmlinkage long sys_io_getevents(io_context_t ctx,
+.\" long min_nr,
+.\" long nr,
+.\" struct io_event *events,
+.\" struct timespec *timeout)
+.\"
+.TH io_getevents 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_getevents, aio_pgetevents \- Read resulting events from io requests
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.sp
+.br
+.B #include <libaio.h>
+.br
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.sp
+struct io_event {
+ unsigned PADDED(data, __pad1);
+ unsigned PADDED(obj, __pad2);
+ unsigned PADDED(res, __pad3);
+ unsigned PADDED(res2, __pad4);
+};
+.sp
+.BI "int io_getevents(io_context_t " ctx ", long " nr ", struct io_event *" events "[], struct timespec *" timeout ");"
+.BI "int io_pgetevents(io_context_t " ctx ", long " nr ", struct io_event *" events "[], struct timespec *" timeout ", sigset_t *" sigmask ");"
+.fi
+.SH DESCRIPTION
+Attempts to read up to
+.I nr
+events from the completion queue for the aio_context specified by
+.IR ctx .
+.SH "RETURN VALUES"
+May return
+.B 0
+if no events are available and the timeout specified
+by when has elapsed, where
+.I when
+== NULL specifies an infinite
+timeout. Note that the timeout pointed to by when is relative and
+will be updated if not NULL and the operation blocks. Will fail with
+.B ENOSYS
+if not implemented.
+.SS io_pgetevents()
+The relationship between
+.BR io_getevents ()
+and
+.BR io_pgetevents ()
+is analogous to the relationship between
+.BR select (2)
+and
+.BR pselect (2):
+similar to
+.BR pselect (2),
+.BR pgetevents ()
+allows an application to safely wait until either an aio completion
+event happens or until a signal is caught.
+.PP
+The following
+.BR io_pgetevents ()
+call:
+.PP
+.in +4n
+.EX
+ret = io_pgetevents(ctx, min_nr, nr, events, timeout, sigmask);
+.EE
+.in
+.PP
+is equivalent to
+.I atomically
+executing the following calls:
+.PP
+.in +4n
+.EX
+sigset_t origmask;
+
+pthread_sigmask(SIG_SETMASK, &sigmask, &origmask);
+ret = io_getevents(ctx, min_nr, nr, events, timeout);
+pthread_sigmask(SIG_SETMASK, &origmask, NULL);
+.EE
+.in
+.PP
+See the description of
+.BR pselect (2)
+for an explanation of why
+.BR io_pgetevents ()
+is necessary.
+.PP
+If the
+.I sigmask
+argument is specified as NULL, then no signal mask manipulation is
+performed (and thus
+.BR io_pgetevents ()
+behaves the same as
+.BR io_getevents() ).
+.SH ERRORS
+.TP
+.B EINVAL
+If
+.I ctx
+is invalid, if
+.I min_nr
+is out of range, if
+.I nr
+is out of range, if
+.I when
+is out of range.
+.TP
+.B EFAULT
+If any of the memory specified to is invalid.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3),
+.BR pselect (2).
--- /dev/null
+.\" static inline void io_prep_fsync(struct iocb *iocb, int fd)
+.\" {
+.\" memset(iocb, 0, sizeof(*iocb));
+.\" iocb->aio_fildes = fd;
+.\" iocb->aio_lio_opcode = IO_CMD_FSYNC;
+.\" iocb->aio_reqprio = 0;
+.\" }
+.TH io_prep_fsync 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_prep_fsync \- Synchronize a file's complete in-core state with that on disk
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "static inline void io_prep_fsync(struct iocb *" iocb ", int " fd ");"
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.sp
+.fi
+.SH DESCRIPTION
+This is an inline convenience function for setting up an
+.I iocbv
+for a
+.B FSYNC
+request.
+.
+.PP
+The file for which
+.nf
+.IR "iocb->aio_fildes = fd"
+.fi
+is a descriptor is set up with the command
+.nf
+.IR "iocb->aio_lio_opcode = IO_CMD_FSYNC"
+.fi
+.
+.PP
+The
+.BR io_prep_fsync ()
+function shall set up an
+.B IO_CMD_FSYNC
+operation to asynchronously force all I/O
+operations associated with the file indicated by the file
+descriptor
+.I aio_fildes
+member of the
+.I iocb
+structure referenced by
+the iocb argument and queued at the time of the call to
+.BR io_submit ()
+to the synchronized I/O completion state. The function
+call shall return when the synchronization request has been
+initiated or queued to the file or device (even when the data
+cannot be synchronized immediately).
+
+All currently queued I/O operations shall be completed as if by a call
+to
+.BR fsync (2);
+that is, as defined for synchronized I/O file
+integrity completion. If the
+operation queued by
+.BR io_prep_fsync ()
+fails, then, as for
+.BR fsync (2),
+outstanding I/O operations are not guaranteed to have
+been completed.
+
+If
+.BR io_prep_fsync ()
+succeeds, then it is only the I/O that was queued
+at the time of the call to
+.BR io_submit (3)
+that is guaranteed to be
+forced to the relevant completion state. The completion of
+subsequent I/O on the file descriptor is not guaranteed to be
+completed in a synchronized fashion.
+.PP
+This function returns immediately. To schedule the operation, the
+function
+.BR io_submit (3)
+must be called.
+.PP
+Simultaneous asynchronous operations using the same
+.I iocb
+produce undefined results.
+.SH "RETURN VALUES"
+None.
+.SH ERRORS
+None.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\" static inline void io_prep_pread(struct iocb *iocb, int fd, void *buf, size_t count, long long offset)
+.\" {
+.\" memset(iocb, 0, sizeof(*iocb));
+.\" iocb->aio_fildes = fd;
+.\" iocb->aio_lio_opcode = IO_CMD_PREAD;
+.\" iocb->aio_reqprio = 0;
+.\" iocb->u.c.buf = buf;
+.\" iocb->u.c.nbytes = count;
+.\" iocb->u.c.offset = offset;
+.\" }
+.TH io_prep_pread 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_prep_pread \- Set up asynchronous read
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.sp
+.br
+.B #include <libaio.h>
+.br
+.sp
+.BI "inline void io_prep_pread(struct iocb *" iocb ", int " fd ", void *" buf ", size_t " count ", long long " offset ");"
+.
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.fi
+.SH DESCRIPTION
+.BR io_prep_pread ()
+is an inline convenience function designed to facilitate the initialization of
+the iocb for an asynchronous read operation.
+
+The first
+.I iocb->u.c.nbytes = count
+bytes of the file for which
+.I iocb->aio_fildes = fd
+is a descriptor are written to the buffer
+starting at
+.IR "iocb->u.c.buf = buf" .
+Reading starts at the absolute position
+.I ioc->u.c.offset = offset
+in the file.
+.PP
+This function returns immediately. To schedule the operation, the
+function
+.BR io_submit (3)
+must be called.
+.PP
+Simultaneous asynchronous operations using the same \fIiocb\fP produce
+undefined results.
+.SH "RETURN VALUES"
+None.
+.SH ERRORS
+None.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\" static inline void io_prep_pwrite(struct iocb *iocb, int fd, void *buf, size_t count, long long offset)
+.\" {
+.\" memset(iocb, 0, sizeof(*iocb));
+.\" iocb->aio_fildes = fd;
+.\" iocb->aio_lio_opcode = IO_CMD_PWRITE;
+.\" iocb->aio_reqprio = 0;
+.\" iocb->u.c.buf = buf;
+.\" iocb->u.c.nbytes = count;
+.\" iocb->u.c.offset = offset;
+.\" }
+.TH io_prep_pwrite 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_prep_pwrite \- Set up iocb for asynchronous writes
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "inline void io_prep_pwrite(struct iocb *" iocb ", int " fd ", void *" buf ", size_t " count ", long long " offset ");"
+.
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.fi
+.SH DESCRIPTION
+.BR io_prep_write ()
+is a convenience function for setting up parallel writes.
+
+The first
+.IR "iocb->u.c.nbytes = count"
+bytes of the file for which
+.IR "iocb->aio_fildes = fd"
+is a descriptor are written from the buffer
+starting at
+.IR "iocb->u.c.buf = buf" .
+Writing starts at the absolute position
+.IR "ioc->u.c.offset = offset"
+in the file.
+.PP
+This function returns immediately. To schedule the operation, the
+function
+.BR io_submit (3)
+must be called.
+.PP
+Simultaneous asynchronous operations using the same iocb produce
+undefined results.
+.SH "RETURN VALUES"
+None.
+.SH ERRORS
+None.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.TH io_queue_init 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_queue_init \- Initialize asynchronous io state machine
+
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "int io_queue_init(int " maxevents ", io_context_t *" ctx ");"
+.sp
+.fi
+.SH DESCRIPTION
+.BR io_queue_init ()
+attempts to create an aio context capable of receiving at least
+.I maxevents
+events.
+.I ctx
+must point to an aio context that already exists and must be initialized
+to
+.B 0
+before the call.
+If the operation is successful,
+.I *cxtp
+is filled with the resulting handle.
+.SH "RETURN VALUES"
+On success,
+.BR io_queue_init ()
+returns
+.BR 0 .
+Otherwise, -error is return, where
+error is one of the Exxx values defined in the
+.B ERRORS
+section.
+.SH ERRORS
+.TP
+.B EFAULT
+.I iocbs
+referenced data outside of the program's accessible address space.
+.TP
+.B EINVAL
+.I maxevents
+is <=
+.B 0
+or
+.IR ctx
+is an invalid memory location.
+.TP
+.B ENOSYS
+Not implemented.
+.TP
+.B EAGAIN
+.IR "maxevents > max_aio_reqs"
+where
+.I max_aio_reqs
+is a tunable value.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.TH io_queue_release 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_queue_release \- Release the context associated with the userspace handle
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.B #include <libaio.h>
+.br
+.sp
+.BI "int io_queue_release(io_context_t " ctx ");"
+.sp
+.SH DESCRIPTION
+.B io_queue_release
+destroys the context associated with the userspace handle. May cancel any outstanding
+AIOs and block on completion.
+.
+.SH "RETURN VALUES"
+On success,
+.BR io_queue_release ()
+returns
+.BR 0 .
+Otherwise, -error is return, where
+error is one of the Exxx values defined in the Errors section.
+.SH ERRORS
+.TP
+.B EINVAL
+.I ctx
+refers to an uninitialized aio context, the
+.I iocb
+pointed to by
+.I iocbs
+contains an improperly initialized
+.IR iocb .
+.TP
+.B ENOSYS
+Not implemented.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_run (3),
+.BR io_queue_wait(3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.TH io_queue_run 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_queue_run \- Handle completed io requests
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "int io_queue_run(io_context_t " ctx ");"
+.sp
+.fi
+.SH DESCRIPTION
+.BR io_queue_run ()
+attempts to read all the events from
+the completion queue for the aio_context specified by
+.IR ctx .
+.SH "RETURN VALUES"
+May return
+.B 0
+if no events are available.
+Will fail with -\fBENOSYS\fP if not implemented.
+.SH ERRORS
+.TP
+.B EFAULT
+.I iocbs
+referenced data outside of the program's accessible address space.
+.TP
+.B EINVAL
+.I ctx
+refers to an uninitialized aio context, the
+.I iocb
+pointed to by
+.I iocbs
+contains an improperly initialized iocb.
+.TP
+.B ENOSYS
+Not implemented.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.TH io_queue_wait 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_queue_wait \- Wait for io requests to complete
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "int io_queue_wait(io_context_t " ctx ", const struct timespec *" timeout ");"
+.fi
+.SH DESCRIPTION
+.BR io_queue_wait ()
+attempts to read an event from
+the completion queue for the aio_context specified by
+.IR ctx .
+.SH "RETURN VALUES"
+May return
+.B 0
+if no events are available and the timeout specified
+by when has elapsed, where when == NULL specifies an infinite
+\fItimeout\fP. Note that the \fItimeout\fP pointed to by when is relative and
+will be updated if not NULL and the operation blocks. Will fail
+with -\fBENOSYS\fP if not implemented.
+.SH "RETURN VALUES"
+On success,
+.BR io_queue_wait ()
+returns
+.BR 0 .
+Otherwise, -error is return, where
+error is one of the Exxx values defined in the Errors section.
+.SH ERRORS
+.TP
+.B EFAULT
+.I iocbs
+referenced data outside of the program's accessible address space.
+.TP
+.B EINVAL
+.I ctx
+refers to an uninitialized aio context, the
+.I iocb
+pointed to by
+.I iocbs
+contains an improperly initialized iocb.
+.TP
+.B ENOSYS
+Not implemented.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_set_callback (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\"static inline void io_set_callback(struct iocb *iocb, io_callback_t cb)
+.TH io_set_callback 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_set_callback \- Set up io completion callback function
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "static inline void io_set_callback(struct iocb *" iocb ", io_callback_t " cb ");"
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.sp
+typedef void (*io_callback_t)(io_context_t ctx, struct iocb *iocb, long res, long res2);
+.sp
+.fi
+.SH DESCRIPTION
+The callback is not done if the caller uses raw events from
+.BR io_getevents (3),
+only with the library helpers.
+.SH "RETURN VALUES"
+None.
+.SH ERRORS
+None.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_submit (3),
+.BR errno (3).
--- /dev/null
+.\"/* sys_io_submit:
+.\" * Queue the nr iocbs pointed to by iocbpp for processing. Returns
+.\" * the number of iocbs queued. May return -EINVAL if the aio_context
+.\" * specified by ctx is invalid, if nr is < 0, if the iocb at
+.\" * *iocbpp[0] is not properly initialized, if the operation specified
+.\" * is invalid for the file descriptor in the iocb. May fail with
+.\" * -EFAULT if any of the data structures point to invalid data. May
+.\" * fail with -EBADF if the file descriptor specified in the first
+.\" * iocb is invalid. May fail with -EAGAIN if insufficient resources
+.\" * are available to queue any iocbs. Will return 0 if nr is 0. Will
+.\" * fail with -ENOSYS if not implemented.
+.\" */
+.TH io_submit 3 2019-07-23 "Linux" "Linux AIO"
+.SH NAME
+io_submit \- Submit io requests
+.SH SYNOPSIS
+.nf
+.B #include <errno.h>
+.br
+.sp
+.B #include <libaio.h>
+.br
+.sp
+.BI "int io_submit(io_context_t " ctx ", long " nr ", struct iocb *" iocbs "[]);"
+.sp
+struct iocb {
+ void *data;
+ unsigned key;
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+};
+.fi
+.SH DESCRIPTION
+.BR io_submit ()
+submits
+.I nr
+iocbs for processing for a given io context ctx.
+
+The
+.BR io_submit ()
+function can be used to enqueue an arbitrary
+number of read and write requests at one time. The requests can all be
+meant for the same file, all for different files or every solution in
+between.
+
+.BR io_submit ()
+gets the
+.I nr
+requests from the array pointed to by
+.IR "iocbs" .
+The operation to be performed is determined by the
+.IR "aio_lio_opcode"
+member in each element of
+.IR "iocbs" .
+If this field is
+.B "IO_CMD_PREAD"
+a read operation is enqueued, similar to a call of
+.IR "io_prep_pread"
+for this element of the array (except that the way
+the termination is signalled is different, as we will see below). If
+the
+.IR "aio_lio_opcode"
+member is
+.B "IO_CMD_PWRITE"
+a write operation
+is enqueued. Otherwise the
+.IR "aio_lio_opcode"
+must be
+.B "IO_CMD_NOP"
+in which case this element of
+.IR "iocbs"
+is simply ignored. This
+``operation'' is useful in situations where one has a fixed array of
+.B struct iocb
+elements from which only a few need to be handled at
+a time. Another situation is where the
+.BR io_submit (3)
+call was
+canceled before all requests are processed and the remaining requests have to be reissued.
+
+The other members of each element of the array pointed to by
+.IR "iocbs"
+must have values suitable for the operation as described in
+the documentation for
+.BR io_prep_pread (3)
+and
+.BR io_prep_pwrite (3)
+above.
+
+The function returns immediately after
+having enqueued all the requests.
+On success,
+.BR io_submit ()
+returns the number of iocbs submitted successfully. Otherwise, -error is return, where
+error is one of the Exxx values defined in the Errors section.
+.PP
+If an error is detected, then the behavior is undefined.
+.PP
+Simultaneous asynchronous operations using the same iocb produce
+undefined results.
+.SH ERRORS
+.TP
+.B EFAULT
+.I iocbs
+referenced data outside of the program's accessible address space.
+.TP
+.B EINVAL
+.I ctx
+refers to an uninitialized aio context, the iocb pointed to by
+.I iocbs
+contains an improperly initialized iocb,
+.TP
+.B EBADF
+The
+.I iocb
+contains a file descriptor that does not exist.
+.TP
+.B EINVAL
+The file specified in the
+.I iocb
+does not support the given io operation.
+.SH "SEE ALSO"
+.BR io (3),
+.BR io_cancel (3),
+.BR io_fsync (3),
+.BR io_getevents (3),
+.BR io_prep_fsync (3),
+.BR io_prep_pread (3),
+.BR io_prep_pwrite (3),
+.BR io_queue_init (3),
+.BR io_queue_release (3),
+.BR io_queue_run (3),
+.BR io_queue_wait (3),
+.BR io_set_callback (3),
+.BR errno (3).
--- /dev/null
+prefix=/usr
+includedir=$(prefix)/include
+libdir=$(prefix)/lib
+
+CFLAGS ?= -g -fomit-frame-pointer -O2
+CFLAGS += -Wall -I. -fPIC
+SO_CFLAGS=-shared $(CFLAGS)
+L_CFLAGS=$(CFLAGS)
+LINK_FLAGS=
+LINK_FLAGS+=$(LDFLAGS)
+ENABLE_SHARED ?= 1
+
+soname=libaio.so.1
+minor=0
+micro=2
+libname=$(soname).$(minor).$(micro)
+all_targets += libaio.a
+
+ifeq ($(ENABLE_SHARED),1)
+all_targets += $(libname)
+endif
+
+all: $(all_targets)
+
+# libaio provided functions
+libaio_srcs := io_queue_init.c io_queue_release.c
+libaio_srcs += io_queue_wait.c io_queue_run.c
+
+# real syscalls
+libaio_srcs += io_getevents.c io_submit.c io_cancel.c
+libaio_srcs += io_setup.c io_destroy.c io_pgetevents.c
+
+# internal functions
+libaio_srcs += raw_syscall.c
+
+# old symbols
+libaio_srcs += compat-0_1.c
+
+libaio_objs := $(patsubst %.c,%.ol,$(libaio_srcs))
+libaio_sobjs := $(patsubst %.c,%.os,$(libaio_srcs))
+
+$(libaio_objs) $(libaio_sobjs): libaio.h vsys_def.h
+
+%.os: %.c
+ $(CC) $(SO_CFLAGS) -c -o $@ $<
+
+%.ol: %.c
+ $(CC) $(L_CFLAGS) -c -o $@ $<
+
+AR ?= ar
+RANLIB ?= ranlib
+libaio.a: $(libaio_objs)
+ rm -f libaio.a
+ $(AR) r libaio.a $^
+ $(RANLIB) libaio.a
+
+$(libname): $(libaio_sobjs) libaio.map
+ $(CC) $(CFLAGS) -c struct_offsets.c
+ $(CC) $(SO_CFLAGS) -Wl,--version-script=libaio.map -Wl,-soname=$(soname) -o $@ $(libaio_sobjs) $(LINK_FLAGS)
+
+install: $(all_targets)
+ install -D -m 644 libaio.h $(includedir)/libaio.h
+ install -D -m 644 libaio.a $(libdir)/libaio.a
+ifeq ($(ENABLE_SHARED),1)
+ install -D -m 755 $(libname) $(libdir)/$(libname)
+ ln -sf $(libname) $(libdir)/$(soname)
+ ln -sf $(libname) $(libdir)/libaio.so
+endif
+
+$(libaio_objs): libaio.h
+
+clean:
+ rm -f $(all_targets) $(libaio_objs) $(libaio_sobjs) $(soname).new
+ rm -f *.so* *.a *.o
--- /dev/null
+/*
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#ifndef _AIO_RING_H
+#define _AIO_RING_H
+
+#define AIO_RING_MAGIC 0xa10a10a1
+
+struct aio_ring {
+ unsigned id; /* kernel internal index number */
+ unsigned nr; /* number of io_events */
+ unsigned head;
+ unsigned tail;
+
+ unsigned magic;
+ unsigned compat_features;
+ unsigned incompat_features;
+ unsigned header_length; /* size of aio_ring */
+};
+
+static inline int aio_ring_is_empty(io_context_t ctx, struct timespec *timeout)
+{
+ struct aio_ring *ring = (struct aio_ring *)ctx;
+
+ if (!ring || ring->magic != AIO_RING_MAGIC)
+ return 0;
+ if (!timeout || timeout->tv_sec || timeout->tv_nsec)
+ return 0;
+ if (ring->head != ring->tail)
+ return 0;
+ return 1;
+}
+
+#endif /* _AIO_RING_H */
--- /dev/null
+/* libaio Linux async I/O interface
+
+ compat-0_1.c : compatibility symbols for libaio 0.1.x-0.3.x
+
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <stdlib.h>
+#include <asm/errno.h>
+
+#include "libaio.h"
+#include "vsys_def.h"
+
+#include "syscall.h"
+
+
+/* ABI change. Provide partial compatibility on this one for now. */
+SYMVER(compat0_1_io_cancel, io_cancel, 0.1);
+int compat0_1_io_cancel(io_context_t ctx, struct iocb *iocb)
+{
+ struct io_event event;
+
+ /* FIXME: the old ABI would return the event on the completion queue */
+ return io_cancel(ctx, iocb, &event);
+}
+
+SYMVER(compat0_1_io_queue_wait, io_queue_wait, 0.1);
+int compat0_1_io_queue_wait(io_context_t ctx, struct timespec *when)
+{
+ struct timespec timeout;
+ if (when)
+ timeout = *when;
+ return io_getevents(ctx, 0, 0, NULL, when ? &timeout : NULL);
+}
+
+
+/* ABI change. Provide backwards compatibility for this one. */
+SYMVER(compat0_1_io_getevents, io_getevents, 0.1);
+int compat0_1_io_getevents(io_context_t ctx, long nr,
+ struct io_event *events,
+ const struct timespec *const_timeout)
+{
+ struct timespec timeout;
+ if (const_timeout)
+ timeout = *const_timeout;
+ return io_getevents(ctx, 1, nr, events,
+ const_timeout ? &timeout : NULL);
+}
+
--- /dev/null
+/* io_cancel.c
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include "syscall.h"
+
+io_syscall3(int, io_cancel_0_4, io_cancel, io_context_t, ctx, struct iocb *, iocb, struct io_event *, event)
+DEFSYMVER(io_cancel_0_4, io_cancel, 0.4)
--- /dev/null
+/* io_destroy
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <errno.h>
+#include <libaio.h>
+#include "syscall.h"
+
+io_syscall1(int, io_destroy, io_destroy, io_context_t, ctx)
--- /dev/null
+/* io_getevents.c
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <time.h>
+#include "syscall.h"
+#include "aio_ring.h"
+
+io_syscall5(int, __io_getevents_0_4, io_getevents, io_context_t, ctx, long, min_nr, long, nr, struct io_event *, events, struct timespec *, timeout)
+
+int io_getevents_0_4(io_context_t ctx, long min_nr, long nr, struct io_event * events, struct timespec * timeout)
+{
+ if (aio_ring_is_empty(ctx, timeout))
+ return 0;
+ return __io_getevents_0_4(ctx, min_nr, nr, events, timeout);
+}
+
+DEFSYMVER(io_getevents_0_4, io_getevents, 0.4)
--- /dev/null
+/*
+ libaio Linux async I/O interface
+ Copyright 2018 Christoph Hellwig.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <time.h>
+#include <signal.h>
+#include "syscall.h"
+#include "aio_ring.h"
+
+#ifdef __NR_io_pgetevents
+io_syscall6(int, __io_pgetevents, io_pgetevents, io_context_t, ctx, long,
+ min_nr, long, nr, struct io_event *, events,
+ struct timespec *, timeout, void *, sigmask);
+
+int io_pgetevents(io_context_t ctx, long min_nr, long nr,
+ struct io_event *events, struct timespec *timeout,
+ sigset_t *sigmask)
+{
+ struct {
+ unsigned long ss;
+ unsigned long ss_len;
+ } data;
+
+ if (aio_ring_is_empty(ctx, timeout))
+ return 0;
+
+ data.ss = (unsigned long)sigmask;
+ data.ss_len = _NSIG / 8;
+ return __io_pgetevents(ctx, min_nr, nr, events, timeout, &data);
+}
+#else
+int io_pgetevents(io_context_t ctx, long min_nr, long nr,
+ struct io_event *events, struct timespec *timeout,
+ sigset_t *sigmask)
+
+{
+ return -ENOSYS;
+}
+#endif /* __NR_io_pgetevents */
--- /dev/null
+/* io_queue_init.c
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <errno.h>
+
+#include "syscall.h"
+
+int io_queue_init(int maxevents, io_context_t *ctxp)
+{
+ if (maxevents > 0) {
+ *ctxp = NULL;
+ return io_setup(maxevents, ctxp);
+ }
+ return -EINVAL;
+}
--- /dev/null
+/* io_queue_release.c
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <errno.h>
+
+int io_queue_release(io_context_t ctx)
+{
+ return io_destroy(ctx);
+}
--- /dev/null
+/* io_submit
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <libaio.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <time.h>
+
+int io_queue_run(io_context_t ctx)
+{
+ static struct timespec timeout = { 0, 0 };
+ struct io_event event;
+ int ret;
+
+ /* FIXME: batch requests? */
+ while (1 == (ret = io_getevents(ctx, 0, 1, &event, &timeout))) {
+ io_callback_t cb = (io_callback_t)event.data;
+ struct iocb *iocb = event.obj;
+
+ cb(ctx, iocb, event.res, event.res2);
+ }
+
+ return ret;
+}
--- /dev/null
+/* io_submit
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#define NO_SYSCALL_ERRNO
+#include <sys/types.h>
+#include <libaio.h>
+#include <errno.h>
+#include "syscall.h"
+
+struct timespec;
+
+int io_queue_wait_0_4(io_context_t ctx, struct timespec *timeout)
+{
+ return io_getevents(ctx, 0, 0, NULL, timeout);
+}
+DEFSYMVER(io_queue_wait_0_4, io_queue_wait, 0.4)
--- /dev/null
+/* io_setup
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <errno.h>
+#include <libaio.h>
+#include "syscall.h"
+
+io_syscall2(int, io_setup, io_setup, int, maxevents, io_context_t *, ctxp)
--- /dev/null
+/* io_submit
+ libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#include <errno.h>
+#include <libaio.h>
+#include "syscall.h"
+
+io_syscall3(int, io_submit, io_submit, io_context_t, ctx, long, nr, struct iocb **, iocbs)
--- /dev/null
+/* /usr/include/libaio.h
+ *
+ * Copyright 2000,2001,2002 Red Hat, Inc.
+ *
+ * Written by Benjamin LaHaise <bcrl@redhat.com>
+ *
+ * libaio Linux async I/O interface
+ *
+ * This library 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 2 of the License, or (at your option) any later version.
+ *
+ * This library 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+#ifndef __LIBAIO_H
+#define __LIBAIO_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <sys/types.h>
+#include <string.h>
+#include <signal.h>
+
+struct timespec;
+struct sockaddr;
+struct iovec;
+
+typedef struct io_context *io_context_t;
+
+typedef enum io_iocb_cmd {
+ IO_CMD_PREAD = 0,
+ IO_CMD_PWRITE = 1,
+
+ IO_CMD_FSYNC = 2,
+ IO_CMD_FDSYNC = 3,
+
+ IO_CMD_POLL = 5,
+ IO_CMD_NOOP = 6,
+ IO_CMD_PREADV = 7,
+ IO_CMD_PWRITEV = 8,
+} io_iocb_cmd_t;
+
+/* little endian, 32 bits */
+#if defined(__i386__) || (defined(__arm__) && !defined(__ARMEB__)) || \
+ defined(__sh__) || defined(__bfin__) || defined(__MIPSEL__) || \
+ defined(__cris__) || defined(__loongarch32) || \
+ (defined(__riscv) && __riscv_xlen == 32) || \
+ (defined(__GNUC__) && defined(__BYTE_ORDER__) && \
+ __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_LONG__ == 4)
+#define PADDED(x, y) x; unsigned y
+#define PADDEDptr(x, y) x; unsigned y
+#define PADDEDul(x, y) unsigned long x; unsigned y
+
+/* little endian, 64 bits */
+#elif defined(__ia64__) || defined(__x86_64__) || defined(__alpha__) || \
+ (defined(__aarch64__) && defined(__AARCH64EL__)) || \
+ defined(__loongarch64) || \
+ (defined(__riscv) && __riscv_xlen == 64) || \
+ (defined(__GNUC__) && defined(__BYTE_ORDER__) && \
+ __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_LONG__ == 8)
+#define PADDED(x, y) x, y
+#define PADDEDptr(x, y) x
+#define PADDEDul(x, y) unsigned long x
+
+/* big endian, 64 bits */
+#elif defined(__powerpc64__) || defined(__s390x__) || \
+ (defined(__sparc__) && defined(__arch64__)) || \
+ (defined(__aarch64__) && defined(__AARCH64EB__)) || \
+ (defined(__GNUC__) && defined(__BYTE_ORDER__) && \
+ __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_LONG__ == 8)
+#define PADDED(x, y) unsigned y; x
+#define PADDEDptr(x,y) x
+#define PADDEDul(x, y) unsigned long x
+
+/* big endian, 32 bits */
+#elif defined(__PPC__) || defined(__s390__) || \
+ (defined(__arm__) && defined(__ARMEB__)) || \
+ defined(__sparc__) || defined(__MIPSEB__) || defined(__m68k__) || \
+ defined(__hppa__) || defined(__frv__) || defined(__avr32__) || \
+ (defined(__GNUC__) && defined(__BYTE_ORDER__) && \
+ __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_LONG__ == 4)
+#define PADDED(x, y) unsigned y; x
+#define PADDEDptr(x, y) unsigned y; x
+#define PADDEDul(x, y) unsigned y; unsigned long x
+
+#else
+#error endian?
+#endif
+
+struct io_iocb_poll {
+ PADDED(int events, __pad1);
+}; /* result code is the set of result flags or -'ve errno */
+
+struct io_iocb_sockaddr {
+ PADDEDptr(struct sockaddr *addr, __pad1);
+ PADDEDul(len, __pad2);
+}; /* result code is the length of the sockaddr, or -'ve errno */
+
+struct io_iocb_common {
+ PADDEDptr(void *buf, __pad1);
+ PADDEDul(nbytes, __pad2);
+ long long offset;
+ long long __pad3;
+ unsigned flags;
+ unsigned resfd;
+}; /* result code is the amount read or -'ve errno */
+
+struct io_iocb_vector {
+ PADDEDptr(const struct iovec *vec, __pad1);
+ PADDEDul(nr, __pad2);
+ long long offset;
+}; /* result code is the amount read or -'ve errno */
+
+struct iocb {
+ PADDEDptr(void *data, __pad1); /* Return in the io completion event */
+ /* key: For use in identifying io requests */
+ /* aio_rw_flags: RWF_* flags (such as RWF_NOWAIT) */
+ PADDED(unsigned key, aio_rw_flags);
+
+ short aio_lio_opcode;
+ short aio_reqprio;
+ int aio_fildes;
+
+ union {
+ struct io_iocb_common c;
+ struct io_iocb_vector v;
+ struct io_iocb_poll poll;
+ struct io_iocb_sockaddr saddr;
+ } u;
+};
+
+struct io_event {
+ PADDEDptr(void *data, __pad1);
+ PADDEDptr(struct iocb *obj, __pad2);
+ PADDEDul(res, __pad3);
+ PADDEDul(res2, __pad4);
+};
+
+#undef PADDED
+#undef PADDEDptr
+#undef PADDEDul
+
+typedef void (*io_callback_t)(io_context_t ctx, struct iocb *iocb, long res, long res2);
+
+/* library wrappers */
+extern int io_queue_init(int maxevents, io_context_t *ctxp);
+/*extern int io_queue_grow(io_context_t ctx, int new_maxevents);*/
+extern int io_queue_release(io_context_t ctx);
+/*extern int io_queue_wait(io_context_t ctx, struct timespec *timeout);*/
+extern int io_queue_run(io_context_t ctx);
+
+/* Actual syscalls */
+extern int io_setup(int maxevents, io_context_t *ctxp);
+extern int io_destroy(io_context_t ctx);
+extern int io_submit(io_context_t ctx, long nr, struct iocb *ios[]);
+extern int io_cancel(io_context_t ctx, struct iocb *iocb, struct io_event *evt);
+extern int io_getevents(io_context_t ctx, long min_nr, long nr, struct io_event *events, struct timespec *timeout);
+extern int io_pgetevents(io_context_t ctx, long min_nr, long nr,
+ struct io_event *events, struct timespec *timeout,
+ sigset_t *sigmask);
+
+
+static inline void io_set_callback(struct iocb *iocb, io_callback_t cb)
+{
+ iocb->data = (void *)cb;
+}
+
+static inline void io_prep_pread(struct iocb *iocb, int fd, void *buf, size_t count, long long offset)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PREAD;
+ iocb->aio_reqprio = 0;
+ iocb->u.c.buf = buf;
+ iocb->u.c.nbytes = count;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_pwrite(struct iocb *iocb, int fd, void *buf, size_t count, long long offset)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PWRITE;
+ iocb->aio_reqprio = 0;
+ iocb->u.c.buf = buf;
+ iocb->u.c.nbytes = count;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_preadv(struct iocb *iocb, int fd, const struct iovec *iov, int iovcnt, long long offset)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PREADV;
+ iocb->aio_reqprio = 0;
+ iocb->u.c.buf = (void *)iov;
+ iocb->u.c.nbytes = iovcnt;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_pwritev(struct iocb *iocb, int fd, const struct iovec *iov, int iovcnt, long long offset)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PWRITEV;
+ iocb->aio_reqprio = 0;
+ iocb->u.c.buf = (void *)iov;
+ iocb->u.c.nbytes = iovcnt;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_preadv2(struct iocb *iocb, int fd, const struct iovec *iov, int iovcnt, long long offset, int flags)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PREADV;
+ iocb->aio_reqprio = 0;
+ iocb->aio_rw_flags = flags;
+ iocb->u.c.buf = (void *)iov;
+ iocb->u.c.nbytes = iovcnt;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_pwritev2(struct iocb *iocb, int fd, const struct iovec *iov, int iovcnt, long long offset, int flags)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_PWRITEV;
+ iocb->aio_reqprio = 0;
+ iocb->aio_rw_flags = flags;
+ iocb->u.c.buf = (void *)iov;
+ iocb->u.c.nbytes = iovcnt;
+ iocb->u.c.offset = offset;
+}
+
+static inline void io_prep_poll(struct iocb *iocb, int fd, int events)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_POLL;
+ iocb->aio_reqprio = 0;
+ iocb->u.poll.events = events;
+}
+
+static inline int io_poll(io_context_t ctx, struct iocb *iocb, io_callback_t cb, int fd, int events)
+{
+ io_prep_poll(iocb, fd, events);
+ io_set_callback(iocb, cb);
+ return io_submit(ctx, 1, &iocb);
+}
+
+static inline void io_prep_fsync(struct iocb *iocb, int fd)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_FSYNC;
+ iocb->aio_reqprio = 0;
+}
+
+static inline int io_fsync(io_context_t ctx, struct iocb *iocb, io_callback_t cb, int fd)
+{
+ io_prep_fsync(iocb, fd);
+ io_set_callback(iocb, cb);
+ return io_submit(ctx, 1, &iocb);
+}
+
+static inline void io_prep_fdsync(struct iocb *iocb, int fd)
+{
+ memset(iocb, 0, sizeof(*iocb));
+ iocb->aio_fildes = fd;
+ iocb->aio_lio_opcode = IO_CMD_FDSYNC;
+ iocb->aio_reqprio = 0;
+}
+
+static inline int io_fdsync(io_context_t ctx, struct iocb *iocb, io_callback_t cb, int fd)
+{
+ io_prep_fdsync(iocb, fd);
+ io_set_callback(iocb, cb);
+ return io_submit(ctx, 1, &iocb);
+}
+
+static inline void io_set_eventfd(struct iocb *iocb, int eventfd)
+{
+ iocb->u.c.flags |= (1 << 0) /* IOCB_FLAG_RESFD */;
+ iocb->u.c.resfd = eventfd;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __LIBAIO_H */
--- /dev/null
+LIBAIO_0.1 {
+ global:
+ io_queue_init;
+ io_queue_run;
+ io_queue_wait;
+ io_queue_release;
+ io_cancel;
+ io_submit;
+ io_getevents;
+ local:
+ *;
+
+};
+
+LIBAIO_0.4 {
+ global:
+ io_setup;
+ io_destroy;
+ io_cancel;
+ io_getevents;
+ io_queue_wait;
+} LIBAIO_0.1;
+
+LIBAIO_0.5 {
+ global:
+ io_pgetevents;
+} LIBAIO_0.4;
--- /dev/null
+#include "syscall.h"
+
+#if defined(__ia64__)
+/* based on code from glibc by Jes Sorensen */
+__asm__(".text\n"
+ ".globl __ia64_aio_raw_syscall\n"
+ ".proc __ia64_aio_raw_syscall\n"
+ "__ia64_aio_raw_syscall:\n"
+ "alloc r2=ar.pfs,1,0,8,0\n"
+ "mov r15=r32\n"
+ "break 0x100000\n"
+ ";;"
+ "br.ret.sptk.few b0\n"
+ ".size __ia64_aio_raw_syscall, . - __ia64_aio_raw_syscall\n"
+ ".endp __ia64_aio_raw_syscall"
+);
+#endif
+
+;
--- /dev/null
+/*
+ * Ensure that data structure offsets in the iocb.u union match.
+ * Note that this code does not end up in the compiled object files.
+ * Its sole purpose is to abort the build if the structure padding
+ * is incorrect.
+ */
+#include <stddef.h>
+#include <assert.h>
+#include <libaio.h>
+
+void
+offset_check(void)
+{
+ static_assert(offsetof(struct iocb, u.v.nr) ==
+ offsetof(struct iocb, u.c.nbytes),
+ "Error: iocb.u.v.nr does not match the offset of iocb.u.c.nbytes.");
+ static_assert(offsetof(struct iocb, u.v.offset) ==
+ offsetof(struct iocb, u.c.offset),
+ "Error: iocb.u.v.offset does not match the offset of iocb.u.c.offset");
+ static_assert(offsetof(struct iocb, u.saddr.len) ==
+ offsetof(struct iocb, u.c.nbytes),
+ "Error: iocb.u.saddr.len does not match the offset of iocb.u.c.nbytes");
+}
--- /dev/null
+#define __NR_io_setup 398
+#define __NR_io_destroy 399
+#define __NR_io_getevents 400
+#define __NR_io_submit 401
+#define __NR_io_cancel 402
--- /dev/null
+/*
+ * linux/include/asm-arm/unistd.h
+ *
+ * Copyright (C) 2001-2005 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Please forward _all_ changes to this file to rmk@arm.linux.org.uk,
+ * no matter what the change is. Thanks!
+ */
+
+#define __NR_OABI_SYSCALL_BASE 0x900000
+
+#if defined(__thumb__) || defined(__ARM_EABI__)
+#define __NR_SYSCALL_BASE 0
+#else
+#define __NR_SYSCALL_BASE __NR_OABI_SYSCALL_BASE
+#endif
+
+#define __NR_io_setup (__NR_SYSCALL_BASE+243)
+#define __NR_io_destroy (__NR_SYSCALL_BASE+244)
+#define __NR_io_getevents (__NR_SYSCALL_BASE+245)
+#define __NR_io_submit (__NR_SYSCALL_BASE+246)
+#define __NR_io_cancel (__NR_SYSCALL_BASE+247)
--- /dev/null
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * This is based on the include/uapi/asm-generic/unistd.h header file
+ * in the kernel, which is a generic syscall schema for new architectures.
+ */
+
+#define __NR_io_setup 0
+#define __NR_io_destroy 1
+#define __NR_io_submit 2
+#define __NR_io_cancel 3
+#define __NR_io_getevents 4
--- /dev/null
+#define __NR_io_setup 245
+#define __NR_io_destroy 246
+#define __NR_io_getevents 247
+#define __NR_io_submit 248
+#define __NR_io_cancel 249
+#define __NR_io_pgetevents 385
--- /dev/null
+#define __NR_io_setup 1238
+#define __NR_io_destroy 1239
+#define __NR_io_getevents 1240
+#define __NR_io_submit 1241
+#define __NR_io_cancel 1242
--- /dev/null
+#define __NR_io_setup 227
+#define __NR_io_destroy 228
+#define __NR_io_getevents 229
+#define __NR_io_submit 230
+#define __NR_io_cancel 231
--- /dev/null
+#define __NR_io_setup 243
+#define __NR_io_destroy 244
+#define __NR_io_getevents 245
+#define __NR_io_submit 246
+#define __NR_io_cancel 247
--- /dev/null
+#define __NR_io_setup 268
+#define __NR_io_destroy 269
+#define __NR_io_submit 270
+#define __NR_io_cancel 271
+#define __NR_io_getevents 272
--- /dev/null
+#define __NR_io_setup 206
+#define __NR_io_destroy 207
+#define __NR_io_getevents 208
+#define __NR_io_submit 209
+#define __NR_io_cancel 210
+#define __NR_io_pgetevents 333
--- /dev/null
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <errno.h>
+
+#define _SYMSTR(str) #str
+#define SYMSTR(str) _SYMSTR(str)
+
+#define SYMVER(compat_sym, orig_sym, ver_sym) \
+ __asm__(".symver " SYMSTR(compat_sym) "," SYMSTR(orig_sym) "@LIBAIO_" SYMSTR(ver_sym));
+
+#define DEFSYMVER(compat_sym, orig_sym, ver_sym) \
+ __asm__(".symver " SYMSTR(compat_sym) "," SYMSTR(orig_sym) "@@LIBAIO_" SYMSTR(ver_sym));
+
+#if defined(__i386__)
+#include "syscall-i386.h"
+#elif defined(__x86_64__)
+#include "syscall-x86_64.h"
+#elif defined(__ia64__)
+#include "syscall-ia64.h"
+#elif defined(__PPC__)
+#include "syscall-ppc.h"
+#elif defined(__s390__)
+#include "syscall-s390.h"
+#elif defined(__alpha__)
+#include "syscall-alpha.h"
+#elif defined(__arm__)
+#include "syscall-arm.h"
+#elif defined(__sparc__)
+#include "syscall-sparc.h"
+#elif defined(__aarch64__) || defined(__loongarch__) || defined(__riscv)
+#include "syscall-generic.h"
+#else
+#warning "using system call numbers from sys/syscall.h"
+#endif
+
+#define _body_io_syscall(sname, args...) \
+{ \
+ int ret, saved_errno; \
+ saved_errno = errno; \
+ ret= syscall(__NR_##sname, ## args); \
+ if (ret < 0) { \
+ ret = -errno; \
+ errno = saved_errno; \
+ } \
+ return ret; \
+}
+
+#define io_syscall1(type,fname,sname,type1,arg1) \
+type fname(type1 arg1) \
+_body_io_syscall(sname, (long)arg1)
+
+#define io_syscall2(type,fname,sname,type1,arg1,type2,arg2) \
+type fname(type1 arg1,type2 arg2) \
+_body_io_syscall(sname, (long)arg1, (long)arg2)
+
+#define io_syscall3(type,fname,sname,type1,arg1,type2,arg2,type3,arg3) \
+type fname(type1 arg1,type2 arg2,type3 arg3) \
+_body_io_syscall(sname, (long)arg1, (long)arg2, (long)arg3)
+
+#define io_syscall4(type,fname,sname,type1,arg1,type2,arg2,type3,arg3,type4,arg4) \
+type fname (type1 arg1, type2 arg2, type3 arg3, type4 arg4) \
+_body_io_syscall(sname, (long)arg1, (long)arg2, (long)arg3, (long)arg4)
+
+#define io_syscall5(type,fname,sname,type1,arg1,type2,arg2,type3,arg3,type4,arg4, type5,arg5) \
+type fname (type1 arg1,type2 arg2,type3 arg3,type4 arg4,type5 arg5) \
+_body_io_syscall(sname, (long)arg1, (long)arg2, (long)arg3, (long)arg4, (long)arg5)
+
+#define io_syscall6(type,fname,sname,type1,arg1,type2,arg2,type3,arg3, \
+ type4,arg4,type5,arg5,type6,arg6) \
+type fname (type1 arg1,type2 arg2,type3 arg3,type4 arg4,type5 arg5, \
+ type6 arg6) \
+_body_io_syscall(sname, (long)arg1, (long)arg2, (long)arg3, (long)arg4, \
+ (long)arg5, (long)arg6)
--- /dev/null
+/* libaio Linux async I/O interface
+ Copyright 2002 Red Hat, Inc.
+
+ This library 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 2 of the License, or (at your option) any later version.
+
+ This library 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
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with this library; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+extern int vsys_io_setup(unsigned nr_reqs, io_context_t *ctxp);
+extern int vsys_io_destroy(io_context_t ctx);
+extern int vsys_io_submit(io_context_t ctx, long nr, struct iocb *iocbs[]);
+extern int vsys_io_cancel(io_context_t ctx, struct iocb *iocb);
+extern int vsys_io_wait(io_context_t ctx, struct iocb *iocb, const struct timespec *when);
+extern int vsys_io_getevents(io_context_t ctx, long nr, struct io_event *events, const struct timespec *timeout);
+
projects / libaio.git / commitdiff