From 022ca2fc7fe08d51f33a1d23a9be49e6d132914e Mon Sep 17 00:00:00 2001 From: Alan Somers Date: Sat, 2 Jan 2021 16:34:20 -0700 Subject: [PATCH] Add aio_writev and aio_readv POSIX AIO is great, but it lacks vectored I/O functions. This commit fixes that shortcoming by adding aio_writev and aio_readv. They aren't part of the standard, but they're an obvious extension. They work just like their synchronous equivalents pwritev and preadv. It isn't yet possible to use vectored aiocbs with lio_listio, but that could be added in the future. Reviewed by: jhb, kib, bcr Relnotes: yes Differential Revision: https://reviews.freebsd.org/D27743 --- lib/libc/sys/Makefile.inc | 2 + lib/libc/sys/Symbol.map | 2 + lib/libc/sys/aio_error.2 | 6 +- lib/libc/sys/aio_read.2 | 63 ++- lib/libc/sys/aio_return.2 | 4 +- lib/libc/sys/aio_write.2 | 63 ++- share/man/man4/aio.4 | 4 +- sys/bsm/audit_kevents.h | 2 + sys/compat/freebsd32/freebsd32_misc.c | 2 +- sys/compat/freebsd32/freebsd32_util.h | 2 + sys/compat/freebsd32/syscalls.master | 6 +- sys/kern/capabilities.conf | 2 + sys/kern/sys_socket.c | 35 +- sys/kern/syscalls.master | 12 +- sys/kern/vfs_aio.c | 501 +++++++++++------ sys/sys/aio.h | 22 +- tests/sys/aio/aio_test.c | 739 +++++++++++++++++++++++--- 17 files changed, 1173 insertions(+), 294 deletions(-) diff --git a/lib/libc/sys/Makefile.inc b/lib/libc/sys/Makefile.inc index d43a59719563..82d16fb81b6b 100644 --- a/lib/libc/sys/Makefile.inc +++ b/lib/libc/sys/Makefile.inc @@ -354,6 +354,8 @@ MAN+= sctp_generic_recvmsg.2 \ write.2 \ _umtx_op.2 +MLINKS+=aio_read.2 aio_readv.2 +MLINKS+=aio_write.2 aio_writev.2 MLINKS+=accept.2 accept4.2 MLINKS+=access.2 eaccess.2 \ access.2 faccessat.2 diff --git a/lib/libc/sys/Symbol.map b/lib/libc/sys/Symbol.map index 9f0d3749ac01..847dd9cca987 100644 --- a/lib/libc/sys/Symbol.map +++ b/lib/libc/sys/Symbol.map @@ -403,6 +403,8 @@ FBSD_1.5 { FBSD_1.6 { __sysctlbyname; + aio_readv; + aio_writev; close_range; copy_file_range; fhlink; diff --git a/lib/libc/sys/aio_error.2 b/lib/libc/sys/aio_error.2 index 030914616121..1ec6505a64aa 100644 --- a/lib/libc/sys/aio_error.2 +++ b/lib/libc/sys/aio_error.2 @@ -24,7 +24,7 @@ .\" .\" $FreeBSD$ .\" -.Dd June 2, 1999 +.Dd January 2, 2021 .Dt AIO_ERROR 2 .Os .Sh NAME @@ -52,7 +52,9 @@ is returned. If the request has completed unsuccessfully the error status is returned as described in .Xr read 2 , +.Xr readv 2 , .Xr write 2 , +.Xr writev 2 , or .Xr fsync 2 . On failure, @@ -76,9 +78,11 @@ does not reference an outstanding asynchronous I/O request. .Sh SEE ALSO .Xr aio_cancel 2 , .Xr aio_read 2 , +.Xr aio_readv 2 , .Xr aio_return 2 , .Xr aio_suspend 2 , .Xr aio_write 2 , +.Xr aio_writev 2 , .Xr fsync 2 , .Xr read 2 , .Xr write 2 , diff --git a/lib/libc/sys/aio_read.2 b/lib/libc/sys/aio_read.2 index bbf96cc89890..0327ef1f747b 100644 --- a/lib/libc/sys/aio_read.2 +++ b/lib/libc/sys/aio_read.2 @@ -24,11 +24,12 @@ .\" .\" $FreeBSD$ .\" -.Dd August 19, 2016 +.Dd January 2, 2021 .Dt AIO_READ 2 .Os .Sh NAME -.Nm aio_read +.Nm aio_read , +.Nm aio_readv .Nd asynchronous read from a file (REALTIME) .Sh LIBRARY .Lb libc @@ -36,21 +37,42 @@ .In aio.h .Ft int .Fn aio_read "struct aiocb *iocb" +.In sys/uio.h +.Ft int +.Fn aio_readv "struct aiocb *iocb" .Sh DESCRIPTION The .Fn aio_read -system call allows the calling process to read -.Fa iocb->aio_nbytes +and +.Fn aio_readv +system calls allow the calling process to read from the descriptor .Fa iocb->aio_fildes beginning at the offset -.Fa iocb->aio_offset -into the buffer pointed to by -.Fa iocb->aio_buf . -The call returns immediately after the read request has +.Fa iocb->aio_offset . +.Fn aio_read +will read +.Fa iocb->aio_nbytes +from the buffer pointed to by +.Fa iocb->aio_buf , +whereas +.Fn aio_readv +reads the data into the +.Fa iocb->aio_iovcnt +buffers specified by the members of the +.Fa iocb->aio_iov +array. +Both syscalls return immediately after the read request has been enqueued to the descriptor; the read may or may not have completed at the time the call returns. .Pp +For +.Fn aio_readv +the +.Fa iovec +structure is defined in +.Xr readv 2 . +.Pp If _POSIX_PRIORITIZED_IO is defined, and the descriptor supports it, then the enqueued operation is submitted at a priority equal to that of the calling process minus @@ -61,7 +83,9 @@ The argument is ignored by the .Fn aio_read -system call. +and +.Fn aio_readv +system calls. .Pp The .Fa iocb @@ -108,16 +132,22 @@ is past the offset maximum for .Fa iocb->aio_fildes , no I/O will occur. .Sh RETURN VALUES -.Rv -std aio_read +.Rv -std aio_read aio_readv .Sh DIAGNOSTICS None. .Sh ERRORS The .Fn aio_read -system call will fail if: +and +.Fn aio_readv +system calls will fail if: .Bl -tag -width Er .It Bq Er EAGAIN The request was not queued because of system resource limitations. +.It Bq Er EFAULT +Part of +.Fa aio_iov +points outside the process's allocated address space. .It Bq Er EINVAL The asynchronous notification method in .Fa iocb->aio_sigevent.sigev_notify @@ -130,10 +160,14 @@ are unsafe and unsafe asynchronous I/O operations are disabled. .Pp The following conditions may be synchronously detected when the .Fn aio_read +or +.Fn aio_readv system call is made, or asynchronously, at any time thereafter. If they are detected at call time, .Fn aio_read +or +.Fn aio_readv returns -1 and sets .Va errno appropriately; otherwise the @@ -207,11 +241,18 @@ The system call is expected to conform to the .St -p1003.1 standard. +The +.Fn aio_readv +system call is a FreeBSD extension, and should not be used in portable code. .Sh HISTORY The .Fn aio_read system call first appeared in .Fx 3.0 . +The +.Fn aio_readv +system call first appeared in +.Fx 13.0 . .Sh AUTHORS This manual page was written by diff --git a/lib/libc/sys/aio_return.2 b/lib/libc/sys/aio_return.2 index df558734ed41..d94fcc7eba62 100644 --- a/lib/libc/sys/aio_return.2 +++ b/lib/libc/sys/aio_return.2 @@ -24,7 +24,7 @@ .\" .\" $FreeBSD$ .\" -.Dd March 21, 2016 +.Dd January 2, 2021 .Dt AIO_RETURN 2 .Os .Sh NAME @@ -55,7 +55,9 @@ returns something other than If the asynchronous I/O request has completed, the status is returned as described in .Xr read 2 , +.Xr readv 2 , .Xr write 2 , +.Xr writev 2 , or .Xr fsync 2 . Otherwise, diff --git a/lib/libc/sys/aio_write.2 b/lib/libc/sys/aio_write.2 index a3268e50ea90..601515b0e7b0 100644 --- a/lib/libc/sys/aio_write.2 +++ b/lib/libc/sys/aio_write.2 @@ -24,11 +24,12 @@ .\" .\" $FreeBSD$ .\" -.Dd August 19, 2016 +.Dd January 2, 2021 .Dt AIO_WRITE 2 .Os .Sh NAME -.Nm aio_write +.Nm aio_write , +.Nm aio_writev .Nd asynchronous write to a file (REALTIME) .Sh LIBRARY .Lb libc @@ -36,28 +37,48 @@ .In aio.h .Ft int .Fn aio_write "struct aiocb *iocb" +.In sys/uio.h +.Ft int +.Fn aio_writev "struct aiocb *iocb" .Sh DESCRIPTION The .Fn aio_write -system call allows the calling process to write -.Fa iocb->aio_nbytes -from the buffer pointed to by -.Fa iocb->aio_buf +and +.Fn aio_writev +system calls allow the calling process to write to the descriptor .Fa iocb->aio_fildes . -The call returns immediately after the write request has been enqueued +.Fn aio_write +will write +.Fa iocb->aio_nbytes +from the buffer pointed to by +.Fa iocb->aio_buf , +whereas +.Fn aio_writev +gathers the data from the +.Fa iocb->aio_iovcnt +buffers specified by the members of the +.Fa iocb->aio_iov +array. +Both syscalls return immediately after the write request has been enqueued to the descriptor; the write may or may not have completed at the time the call returns. If the request could not be enqueued, generally due to invalid arguments, the call returns without having enqueued the request. .Pp +For +.Fn aio_writev +the +.Fa iovec +structure is defined in +.Xr writev 2 . +.Pp If .Dv O_APPEND is set for .Fa iocb->aio_fildes , -.Fn aio_write -operations append to the file in the same order as the calls were +write operations append to the file in the same order as the calls were made. If .Dv O_APPEND @@ -103,6 +124,8 @@ The asynchronous I/O control buffer .Fa iocb should be zeroed before the .Fn aio_write +or +.Fn aio_writev system call to avoid passing bogus context information to the kernel. .Pp Modifications of the Asynchronous I/O Control Block structure or the @@ -114,14 +137,20 @@ is past the offset maximum for .Fa iocb->aio_fildes , no I/O will occur. .Sh RETURN VALUES -.Rv -std aio_write +.Rv -std aio_write aio_writev .Sh ERRORS The .Fn aio_write -system call will fail if: +and +.Fn aio_writev +system calls will fail if: .Bl -tag -width Er .It Bq Er EAGAIN The request was not queued because of system resource limitations. +.It Bq Er EFAULT +Part of +.Fa aio_iov +points outside the process's allocated address space. .It Bq Er EINVAL The asynchronous notification method in .Fa iocb->aio_sigevent.sigev_notify @@ -134,10 +163,14 @@ are unsafe and unsafe asynchronous I/O operations are disabled. .Pp The following conditions may be synchronously detected when the .Fn aio_write +or +.Fn aio_writev system call is made, or asynchronously, at any time thereafter. If they are detected at call time, .Fn aio_write +or +.Fn aio_writev returns -1 and sets .Va errno appropriately; otherwise the @@ -203,11 +236,19 @@ system call is expected to conform to the .St -p1003.1 standard. +.Pp +The +.Fn aio_writev +system call is a FreeBSD extension, and should not be used in portable code. .Sh HISTORY The .Fn aio_write system call first appeared in .Fx 3.0 . +The +.Fn aio_writev +system call first appeared in +.Fx 13.0 . .Sh AUTHORS This manual page was written by .An Wes Peters Aq Mt wes@softweyr.com . diff --git a/share/man/man4/aio.4 b/share/man/man4/aio.4 index 0ea728499d13..513a5728defc 100644 --- a/share/man/man4/aio.4 +++ b/share/man/man4/aio.4 @@ -27,7 +27,7 @@ .\" .\" $FreeBSD$ .\" -.Dd June 22, 2017 +.Dd January 2, 2021 .Dt AIO 4 .Os .Sh NAME @@ -215,10 +215,12 @@ as described in .Xr aio_cancel 2 , .Xr aio_error 2 , .Xr aio_read 2 , +.Xr aio_readv 2 , .Xr aio_return 2 , .Xr aio_suspend 2 , .Xr aio_waitcomplete 2 , .Xr aio_write 2 , +.Xr aio_writev 2 , .Xr lio_listio 2 , .Xr sigevent 3 , .Xr sysctl 8 diff --git a/sys/bsm/audit_kevents.h b/sys/bsm/audit_kevents.h index 5b37329078a1..eeb928ecafdc 100644 --- a/sys/bsm/audit_kevents.h +++ b/sys/bsm/audit_kevents.h @@ -660,6 +660,8 @@ #define AUE_REALPATHAT 43264 /* FreeBSD-specific. */ #define AUE_CLOSERANGE 43265 /* FreeBSD-specific. */ #define AUE_SPECIALFD 43266 /* FreeBSD-specific. */ +#define AUE_AIO_WRITEV 43267 /* FreeBSD-specific. */ +#define AUE_AIO_READV 43268 /* FreeBSD-specific. */ /* * Darwin BSM uses a number of AUE_O_* definitions, which are aliased to the diff --git a/sys/compat/freebsd32/freebsd32_misc.c b/sys/compat/freebsd32/freebsd32_misc.c index 62fab95c68d1..14afd433d9f1 100644 --- a/sys/compat/freebsd32/freebsd32_misc.c +++ b/sys/compat/freebsd32/freebsd32_misc.c @@ -1070,7 +1070,7 @@ freebsd32_ptrace(struct thread *td, struct freebsd32_ptrace_args *uap) return (error); } -static int +int freebsd32_copyinuio(struct iovec32 *iovp, u_int iovcnt, struct uio **uiop) { struct iovec32 iov32; diff --git a/sys/compat/freebsd32/freebsd32_util.h b/sys/compat/freebsd32/freebsd32_util.h index a66038d4d36a..b126fbde0857 100644 --- a/sys/compat/freebsd32/freebsd32_util.h +++ b/sys/compat/freebsd32/freebsd32_util.h @@ -116,6 +116,8 @@ int freebsd32_copyout_strings(struct image_params *imgp, uintptr_t *stack_base); int freebsd32_copyiniov(struct iovec32 *iovp, u_int iovcnt, struct iovec **iov, int error); +int freebsd32_copyinuio(struct iovec32 *iovp, u_int iovcnt, + struct uio **uiop); void freebsd32_rusage_out(const struct rusage *s, struct rusage32 *s32); struct image_args; diff --git a/sys/compat/freebsd32/syscalls.master b/sys/compat/freebsd32/syscalls.master index f4339795781a..ca0db9a76b1e 100644 --- a/sys/compat/freebsd32/syscalls.master +++ b/sys/compat/freebsd32/syscalls.master @@ -493,8 +493,10 @@ 257 AUE_LIO_LISTIO STD { int freebsd32_lio_listio(int mode, \ struct aiocb32 * const *acb_list, \ int nent, struct sigevent32 *sig); } -258 AUE_NULL UNIMPL nosys -259 AUE_NULL UNIMPL nosys +258 AUE_AIO_WRITEV STD { int freebsd32_aio_writev( \ + struct aiocb32 *aiocbp); } +259 AUE_AIO_READV STD { int freebsd32_aio_readv( \ + struct aiocb32 *aiocbp); } 260 AUE_NULL UNIMPL nosys 261 AUE_NULL UNIMPL nosys 262 AUE_NULL UNIMPL nosys diff --git a/sys/kern/capabilities.conf b/sys/kern/capabilities.conf index 3d552255d823..602ec7088fc6 100644 --- a/sys/kern/capabilities.conf +++ b/sys/kern/capabilities.conf @@ -100,6 +100,8 @@ aio_return aio_suspend aio_waitcomplete aio_write +aio_writev +aio_readv ## ## audit(2) is a global operation, submitting to the global trail, but it is diff --git a/sys/kern/sys_socket.c b/sys/kern/sys_socket.c index 0fe200c119d2..18803b6a5ac0 100644 --- a/sys/kern/sys_socket.c +++ b/sys/kern/sys_socket.c @@ -600,9 +600,7 @@ soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) struct ucred *td_savedcred; struct thread *td; struct file *fp; - struct uio uio; - struct iovec iov; - size_t cnt, done; + size_t cnt, done, job_total_nbytes; long ru_before; int error, flags; @@ -614,16 +612,11 @@ soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) td_savedcred = td->td_ucred; td->td_ucred = job->cred; + job_total_nbytes = job->uiop->uio_resid + job->aio_done; done = job->aio_done; - cnt = job->uaiocb.aio_nbytes - done; - iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done); - iov.iov_len = cnt; - uio.uio_iov = &iov; - uio.uio_iovcnt = 1; - uio.uio_offset = 0; - uio.uio_resid = cnt; - uio.uio_segflg = UIO_USERSPACE; - uio.uio_td = td; + cnt = job->uiop->uio_resid; + job->uiop->uio_offset = 0; + job->uiop->uio_td = td; flags = MSG_NBIO; /* @@ -633,26 +626,26 @@ soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) */ if (sb == &so->so_rcv) { - uio.uio_rw = UIO_READ; ru_before = td->td_ru.ru_msgrcv; #ifdef MAC error = mac_socket_check_receive(fp->f_cred, so); if (error == 0) #endif - error = soreceive(so, NULL, &uio, NULL, NULL, &flags); + error = soreceive(so, NULL, job->uiop, NULL, NULL, + &flags); if (td->td_ru.ru_msgrcv != ru_before) job->msgrcv = 1; } else { if (!TAILQ_EMPTY(&sb->sb_aiojobq)) flags |= MSG_MORETOCOME; - uio.uio_rw = UIO_WRITE; ru_before = td->td_ru.ru_msgsnd; #ifdef MAC error = mac_socket_check_send(fp->f_cred, so); if (error == 0) #endif - error = sosend(so, NULL, &uio, NULL, NULL, flags, td); + error = sosend(so, NULL, job->uiop, NULL, NULL, flags, + td); if (td->td_ru.ru_msgsnd != ru_before) job->msgsnd = 1; if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) { @@ -662,7 +655,7 @@ soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) } } - done += cnt - uio.uio_resid; + done += cnt - job->uiop->uio_resid; job->aio_done = done; td->td_ucred = td_savedcred; @@ -676,7 +669,7 @@ soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) * been made, requeue this request at the head of the * queue to try again when the socket is ready. */ - MPASS(done != job->uaiocb.aio_nbytes); + MPASS(done != job_total_nbytes); SOCKBUF_LOCK(sb); if (done == 0 || !(so->so_state & SS_NBIO)) { empty_results++; @@ -782,10 +775,10 @@ soo_aio_cancel(struct kaiocb *job) so = job->fd_file->f_data; opcode = job->uaiocb.aio_lio_opcode; - if (opcode == LIO_READ) + if (opcode == LIO_READ || opcode == LIO_READV) sb = &so->so_rcv; else { - MPASS(opcode == LIO_WRITE); + MPASS(opcode == LIO_WRITE || opcode == LIO_WRITEV); sb = &so->so_snd; } @@ -817,9 +810,11 @@ soo_aio_queue(struct file *fp, struct kaiocb *job) switch (job->uaiocb.aio_lio_opcode) { case LIO_READ: + case LIO_READV: sb = &so->so_rcv; break; case LIO_WRITE: + case LIO_WRITEV: sb = &so->so_snd; break; default: diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master index b7ea5e939635..aaa0a1277461 100644 --- a/sys/kern/syscalls.master +++ b/sys/kern/syscalls.master @@ -1477,7 +1477,17 @@ _In_opt_ struct sigevent *sig ); } -258-271 AUE_NULL UNIMPL nosys +258 AUE_AIO_WRITEV STD { + int aio_writev( + _Inout_ struct aiocb *aiocbp + ); + } +259 AUE_AIO_READV STD { + int aio_readv( + _Inout_ struct aiocb *aiocbp + ); + } +260-271 AUE_NULL UNIMPL nosys 272 AUE_O_GETDENTS COMPAT11 { int getdents( int fd, diff --git a/sys/kern/vfs_aio.c b/sys/kern/vfs_aio.c index 37e19557d807..d83c9d725e68 100644 --- a/sys/kern/vfs_aio.c +++ b/sys/kern/vfs_aio.c @@ -292,7 +292,7 @@ struct kaioinfo { * Different ABIs provide their own operations. */ struct aiocb_ops { - int (*aio_copyin)(struct aiocb *ujob, struct aiocb *kjob); + int (*aio_copyin)(struct aiocb *ujob, struct kaiocb *kjob, int ty); long (*fetch_status)(struct aiocb *ujob); long (*fetch_error)(struct aiocb *ujob); int (*store_status)(struct aiocb *ujob, long status); @@ -307,6 +307,7 @@ static struct mtx aio_job_mtx; static TAILQ_HEAD(,kaiocb) aio_jobs; /* (c) Async job list */ static struct unrhdr *aiod_unr; +static void aio_biocleanup(struct bio *bp); void aio_init_aioinfo(struct proc *p); static int aio_onceonly(void); static int aio_free_entry(struct kaiocb *job); @@ -559,6 +560,8 @@ aio_free_entry(struct kaiocb *job) if (job->fd_file) fdrop(job->fd_file, curthread); crfree(job->cred); + if (job->uiop != &job->uio) + free(job->uiop, M_IOV); uma_zfree(aiocb_zone, job); AIO_LOCK(ki); @@ -754,36 +757,29 @@ aio_process_rw(struct kaiocb *job) struct thread *td; struct aiocb *cb; struct file *fp; - struct uio auio; - struct iovec aiov; ssize_t cnt; long msgsnd_st, msgsnd_end; long msgrcv_st, msgrcv_end; long oublock_st, oublock_end; long inblock_st, inblock_end; - int error; + int error, opcode; KASSERT(job->uaiocb.aio_lio_opcode == LIO_READ || - job->uaiocb.aio_lio_opcode == LIO_WRITE, + job->uaiocb.aio_lio_opcode == LIO_READV || + job->uaiocb.aio_lio_opcode == LIO_WRITE || + job->uaiocb.aio_lio_opcode == LIO_WRITEV, ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode)); aio_switch_vmspace(job); td = curthread; td_savedcred = td->td_ucred; td->td_ucred = job->cred; + job->uiop->uio_td = td; cb = &job->uaiocb; fp = job->fd_file; - aiov.iov_base = (void *)(uintptr_t)cb->aio_buf; - aiov.iov_len = cb->aio_nbytes; - - auio.uio_iov = &aiov; - auio.uio_iovcnt = 1; - auio.uio_offset = cb->aio_offset; - auio.uio_resid = cb->aio_nbytes; - cnt = cb->aio_nbytes; - auio.uio_segflg = UIO_USERSPACE; - auio.uio_td = td; + opcode = job->uaiocb.aio_lio_opcode; + cnt = job->uiop->uio_resid; msgrcv_st = td->td_ru.ru_msgrcv; msgsnd_st = td->td_ru.ru_msgsnd; @@ -794,17 +790,16 @@ aio_process_rw(struct kaiocb *job) * aio_aqueue() acquires a reference to the file that is * released in aio_free_entry(). */ - if (cb->aio_lio_opcode == LIO_READ) { - auio.uio_rw = UIO_READ; - if (auio.uio_resid == 0) + if (opcode == LIO_READ || opcode == LIO_READV) { + if (job->uiop->uio_resid == 0) error = 0; else - error = fo_read(fp, &auio, fp->f_cred, FOF_OFFSET, td); + error = fo_read(fp, job->uiop, fp->f_cred, FOF_OFFSET, + td); } else { if (fp->f_type == DTYPE_VNODE) bwillwrite(); - auio.uio_rw = UIO_WRITE; - error = fo_write(fp, &auio, fp->f_cred, FOF_OFFSET, td); + error = fo_write(fp, job->uiop, fp->f_cred, FOF_OFFSET, td); } msgrcv_end = td->td_ru.ru_msgrcv; msgsnd_end = td->td_ru.ru_msgsnd; @@ -816,17 +811,18 @@ aio_process_rw(struct kaiocb *job) job->inblock = inblock_end - inblock_st; job->outblock = oublock_end - oublock_st; - if ((error) && (auio.uio_resid != cnt)) { + if (error != 0 && job->uiop->uio_resid != cnt) { if (error == ERESTART || error == EINTR || error == EWOULDBLOCK) error = 0; - if ((error == EPIPE) && (cb->aio_lio_opcode == LIO_WRITE)) { + if (error == EPIPE && + (opcode == LIO_WRITE || opcode == LIO_WRITEV)) { PROC_LOCK(job->userproc); kern_psignal(job->userproc, SIGPIPE); PROC_UNLOCK(job->userproc); } } - cnt -= auio.uio_resid; + cnt -= job->uiop->uio_resid; td->td_ucred = td_savedcred; if (error) aio_complete(job, -1, error); @@ -1210,21 +1206,23 @@ aio_qbio(struct proc *p, struct kaiocb *job) { struct aiocb *cb; struct file *fp; - struct bio *bp; struct buf *pbuf; struct vnode *vp; struct cdevsw *csw; struct cdev *dev; struct kaioinfo *ki; - struct vm_page **pages; - int error, npages, poff, ref; + struct bio **bios = NULL; + off_t offset; + int bio_cmd, error, i, iovcnt, opcode, poff, ref; vm_prot_t prot; + bool use_unmapped; cb = &job->uaiocb; fp = job->fd_file; + opcode = cb->aio_lio_opcode; - if (!(cb->aio_lio_opcode == LIO_WRITE || - cb->aio_lio_opcode == LIO_READ)) + if (!(opcode == LIO_WRITE || opcode == LIO_WRITEV || + opcode == LIO_READ || opcode == LIO_READV)) return (-1); if (fp == NULL || fp->f_type != DTYPE_VNODE) return (-1); @@ -1234,8 +1232,21 @@ aio_qbio(struct proc *p, struct kaiocb *job) return (-1); if (vp->v_bufobj.bo_bsize == 0) return (-1); - if (cb->aio_nbytes % vp->v_bufobj.bo_bsize) + + bio_cmd = opcode == LIO_WRITE || opcode == LIO_WRITEV ? BIO_WRITE : + BIO_READ; + iovcnt = job->uiop->uio_iovcnt; + if (iovcnt > max_buf_aio) return (-1); + for (i = 0; i < iovcnt; i++) { + if (job->uiop->uio_iov[i].iov_len % vp->v_bufobj.bo_bsize != 0) + return (-1); + if (job->uiop->uio_iov[i].iov_len > maxphys) { + error = -1; + return (-1); + } + } + offset = cb->aio_offset; ref = 0; csw = devvn_refthread(vp, &dev, &ref); @@ -1246,89 +1257,106 @@ aio_qbio(struct proc *p, struct kaiocb *job) error = -1; goto unref; } - if (cb->aio_nbytes > dev->si_iosize_max) { + if (job->uiop->uio_resid > dev->si_iosize_max) { error = -1; goto unref; } ki = p->p_aioinfo; - poff = (vm_offset_t)cb->aio_buf & PAGE_MASK; - if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) { - if (cb->aio_nbytes > maxphys) { - error = -1; - goto unref; - } + job->error = 0; - pbuf = NULL; - pages = malloc(sizeof(vm_page_t) * (atop(round_page( - cb->aio_nbytes)) + 1), M_TEMP, M_WAITOK | M_ZERO); - } else { - if (cb->aio_nbytes > maxphys) { - error = -1; - goto unref; - } - if (ki->kaio_buffer_count >= max_buf_aio) { + use_unmapped = (dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed; + if (!use_unmapped) { + AIO_LOCK(ki); + if (ki->kaio_buffer_count + iovcnt > max_buf_aio) { + AIO_UNLOCK(ki); error = EAGAIN; goto unref; } - - pbuf = uma_zalloc(pbuf_zone, M_WAITOK); - BUF_KERNPROC(pbuf); - AIO_LOCK(ki); - ki->kaio_buffer_count++; + ki->kaio_buffer_count += iovcnt; AIO_UNLOCK(ki); - pages = pbuf->b_pages; } - bp = g_alloc_bio(); - bp->bio_length = cb->aio_nbytes; - bp->bio_bcount = cb->aio_nbytes; - bp->bio_done = aio_biowakeup; - bp->bio_offset = cb->aio_offset; - bp->bio_cmd = cb->aio_lio_opcode == LIO_WRITE ? BIO_WRITE : BIO_READ; - bp->bio_dev = dev; - bp->bio_caller1 = job; - bp->bio_caller2 = pbuf; + bios = malloc(sizeof(struct bio *) * iovcnt, M_TEMP, M_WAITOK); + atomic_store_int(&job->nbio, iovcnt); + for (i = 0; i < iovcnt; i++) { + struct vm_page** pages; + struct bio *bp; + void *buf; + size_t nbytes; + int npages; - prot = VM_PROT_READ; - if (cb->aio_lio_opcode == LIO_READ) - prot |= VM_PROT_WRITE; /* Less backwards than it looks */ - npages = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, - (vm_offset_t)cb->aio_buf, bp->bio_length, prot, pages, - atop(maxphys) + 1); - if (npages < 0) { - error = EFAULT; - goto doerror; - } - if (pbuf != NULL) { - pmap_qenter((vm_offset_t)pbuf->b_data, pages, npages); - bp->bio_data = pbuf->b_data + poff; - atomic_add_int(&num_buf_aio, 1); - pbuf->b_npages = npages; - } else { - bp->bio_ma = pages; - bp->bio_ma_n = npages; - bp->bio_ma_offset = poff; - bp->bio_data = unmapped_buf; - bp->bio_flags |= BIO_UNMAPPED; - atomic_add_int(&num_unmapped_aio, 1); + buf = job->uiop->uio_iov[i].iov_base; + nbytes = job->uiop->uio_iov[i].iov_len; + + bios[i] = g_alloc_bio(); + bp = bios[i]; + + poff = (vm_offset_t)buf & PAGE_MASK; + if (use_unmapped) { + pbuf = NULL; + pages = malloc(sizeof(vm_page_t) * (atop(round_page( + nbytes)) + 1), M_TEMP, M_WAITOK | M_ZERO); + } else { + pbuf = uma_zalloc(pbuf_zone, M_WAITOK); + BUF_KERNPROC(pbuf); + pages = pbuf->b_pages; + } + + bp->bio_length = nbytes; + bp->bio_bcount = nbytes; + bp->bio_done = aio_biowakeup; + bp->bio_offset = offset; + bp->bio_cmd = bio_cmd; + bp->bio_dev = dev; + bp->bio_caller1 = job; + bp->bio_caller2 = pbuf; + + prot = VM_PROT_READ; + if (opcode == LIO_READ || opcode == LIO_READV) + prot |= VM_PROT_WRITE; /* Less backwards than it looks */ + npages = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, + (vm_offset_t)buf, bp->bio_length, prot, pages, + atop(maxphys) + 1); + if (npages < 0) { + if (pbuf != NULL) + uma_zfree(pbuf_zone, pbuf); + else + free(pages, M_TEMP); + error = EFAULT; + g_destroy_bio(bp); + i--; + goto destroy_bios; + } + if (pbuf != NULL) { + pmap_qenter((vm_offset_t)pbuf->b_data, pages, npages); + bp->bio_data = pbuf->b_data + poff; + pbuf->b_npages = npages; + atomic_add_int(&num_buf_aio, 1); + } else { + bp->bio_ma = pages; + bp->bio_ma_n = npages; + bp->bio_ma_offset = poff; + bp->bio_data = unmapped_buf; + bp->bio_flags |= BIO_UNMAPPED; + atomic_add_int(&num_unmapped_aio, 1); + } + + offset += nbytes; } /* Perform transfer. */ - csw->d_strategy(bp); + for (i = 0; i < iovcnt; i++) + csw->d_strategy(bios[i]); + free(bios, M_TEMP); + dev_relthread(dev, ref); return (0); -doerror: - if (pbuf != NULL) { - AIO_LOCK(ki); - ki->kaio_buffer_count--; - AIO_UNLOCK(ki); - uma_zfree(pbuf_zone, pbuf); - } else { - free(pages, M_TEMP); - } - g_destroy_bio(bp); +destroy_bios: + for (; i >= 0; i--) + aio_biocleanup(bios[i]); + free(bios, M_TEMP); unref: dev_relthread(dev, ref); return (error); @@ -1362,25 +1390,39 @@ convert_old_sigevent(struct osigevent *osig, struct sigevent *nsig) } static int -aiocb_copyin_old_sigevent(struct aiocb *ujob, struct aiocb *kjob) +aiocb_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob, + int type __unused) { struct oaiocb *ojob; + struct aiocb *kcb = &kjob->uaiocb; int error; - bzero(kjob, sizeof(struct aiocb)); - error = copyin(ujob, kjob, sizeof(struct oaiocb)); + bzero(kcb, sizeof(struct aiocb)); + error = copyin(ujob, kcb, sizeof(struct oaiocb)); if (error) return (error); - ojob = (struct oaiocb *)kjob; - return (convert_old_sigevent(&ojob->aio_sigevent, &kjob->aio_sigevent)); + /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */ + ojob = (struct oaiocb *)kcb; + return (convert_old_sigevent(&ojob->aio_sigevent, &kcb->aio_sigevent)); } #endif static int -aiocb_copyin(struct aiocb *ujob, struct aiocb *kjob) +aiocb_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type) { + struct aiocb *kcb = &kjob->uaiocb; + int error; - return (copyin(ujob, kjob, sizeof(struct aiocb))); + error = copyin(ujob, kcb, sizeof(struct aiocb)); + if (error) + return (error); + if (type == LIO_READV || type == LIO_WRITEV) { + /* malloc a uio and copy in the iovec */ + error = copyinuio(__DEVOLATILE(struct iovec*, kcb->aio_iov), + kcb->aio_iovcnt, &kjob->uiop); + } + + return (error); } static long @@ -1456,7 +1498,7 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, int type, struct aiocb_ops *ops) { struct proc *p = td->td_proc; - struct file *fp; + struct file *fp = NULL; struct kaiocb *job; struct kaioinfo *ki; struct kevent kev; @@ -1477,39 +1519,35 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, if (num_queue_count >= max_queue_count || ki->kaio_count >= max_aio_queue_per_proc) { - ops->store_error(ujob, EAGAIN); - return (EAGAIN); + error = EAGAIN; + goto err1; } job = uma_zalloc(aiocb_zone, M_WAITOK | M_ZERO); knlist_init_mtx(&job->klist, AIO_MTX(ki)); - error = ops->aio_copyin(ujob, &job->uaiocb); - if (error) { - ops->store_error(ujob, error); - uma_zfree(aiocb_zone, job); - return (error); - } + error = ops->aio_copyin(ujob, job, type); + if (error) + goto err2; if (job->uaiocb.aio_nbytes > IOSIZE_MAX) { - uma_zfree(aiocb_zone, job); - return (EINVAL); + error = EINVAL; + goto err2; } if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_SIGNAL && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_THREAD_ID && job->uaiocb.aio_sigevent.sigev_notify != SIGEV_NONE) { - ops->store_error(ujob, EINVAL); - uma_zfree(aiocb_zone, job); - return (EINVAL); + error = EINVAL; + goto err2; } if ((job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL || job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID) && !_SIG_VALID(job->uaiocb.aio_sigevent.sigev_signo)) { - uma_zfree(aiocb_zone, job); - return (EINVAL); + error = EINVAL; + goto err2; } ksiginfo_init(&job->ksi); @@ -1533,16 +1571,17 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, fd = job->uaiocb.aio_fildes; switch (opcode) { case LIO_WRITE: + case LIO_WRITEV: error = fget_write(td, fd, &cap_pwrite_rights, &fp); break; case LIO_READ: + case LIO_READV: error = fget_read(td, fd, &cap_pread_rights, &fp); break; case LIO_SYNC: error = fget(td, fd, &cap_fsync_rights, &fp); break; case LIO_MLOCK: - fp = NULL; break; case LIO_NOP: error = fget(td, fd, &cap_no_rights, &fp); @@ -1550,22 +1589,20 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, default: error = EINVAL; } - if (error) { - uma_zfree(aiocb_zone, job); - ops->store_error(ujob, error); - return (error); - } + if (error) + goto err3; if (opcode == LIO_SYNC && fp->f_vnode == NULL) { error = EINVAL; - goto aqueue_fail; + goto err3; } - if ((opcode == LIO_READ || opcode == LIO_WRITE) && + if ((opcode == LIO_READ || opcode == LIO_READV || + opcode == LIO_WRITE || opcode == LIO_WRITEV) && job->uaiocb.aio_offset < 0 && (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) { error = EINVAL; - goto aqueue_fail; + goto err3; } job->fd_file = fp; @@ -1577,12 +1614,13 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, error = ops->store_kernelinfo(ujob, jid); if (error) { error = EINVAL; - goto aqueue_fail; + goto err3; } job->uaiocb._aiocb_private.kernelinfo = (void *)(intptr_t)jid; if (opcode == LIO_NOP) { fdrop(fp, td); + MPASS(job->uiop == &job->uio || job->uiop == NULL); uma_zfree(aiocb_zone, job); return (0); } @@ -1592,7 +1630,7 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, evflags = job->uaiocb.aio_sigevent.sigev_notify_kevent_flags; if ((evflags & ~(EV_CLEAR | EV_DISPATCH | EV_ONESHOT)) != 0) { error = EINVAL; - goto aqueue_fail; + goto err3; } kqfd = job->uaiocb.aio_sigevent.sigev_notify_kqueue; memset(&kev, 0, sizeof(kev)); @@ -1603,7 +1641,7 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, kev.udata = job->uaiocb.aio_sigevent.sigev_value.sival_ptr; error = kqfd_register(kqfd, &kev, td, M_WAITOK); if (error) - goto aqueue_fail; + goto err3; no_kqueue: @@ -1614,6 +1652,39 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, job->jobflags = KAIOCB_QUEUEING; job->lio = lj; + switch (opcode) { + case LIO_READV: + case LIO_WRITEV: + /* Use the uio copied in by aio_copyin */ + MPASS(job->uiop != &job->uio && job->uiop != NULL); + break; + case LIO_READ: + case LIO_WRITE: + /* Setup the inline uio */ + job->iov[0].iov_base = (void *)(uintptr_t)job->uaiocb.aio_buf; + job->iov[0].iov_len = job->uaiocb.aio_nbytes; + job->uio.uio_iov = job->iov; + job->uio.uio_iovcnt = 1; + job->uio.uio_resid = job->uaiocb.aio_nbytes; + job->uio.uio_segflg = UIO_USERSPACE; + /* FALLTHROUGH */ + default: + job->uiop = &job->uio; + break; + } + switch (opcode) { + case LIO_READ: + case LIO_READV: + job->uiop->uio_rw = UIO_READ; + break; + case LIO_WRITE: + case LIO_WRITEV: + job->uiop->uio_rw = UIO_WRITE; + break; + } + job->uiop->uio_offset = job->uaiocb.aio_offset; + job->uiop->uio_td = td; + if (opcode == LIO_MLOCK) { aio_schedule(job, aio_process_mlock); error = 0; @@ -1622,7 +1693,7 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, else error = fo_aio_queue(fp, job); if (error) - goto aqueue_fail; + goto err3; AIO_LOCK(ki); job->jobflags &= ~KAIOCB_QUEUEING; @@ -1643,11 +1714,15 @@ aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, AIO_UNLOCK(ki); return (0); -aqueue_fail: - knlist_delete(&job->klist, curthread, 0); +err3: if (fp) fdrop(fp, td); + knlist_delete(&job->klist, curthread, 0); +err2: + if (job->uiop != &job->uio) + free(job->uiop, M_IOV); uma_zfree(aiocb_zone, job); +err1: ops->store_error(ujob, error); return (error); } @@ -1723,7 +1798,9 @@ aio_queue_file(struct file *fp, struct kaiocb *job) switch (job->uaiocb.aio_lio_opcode) { case LIO_READ: + case LIO_READV: case LIO_WRITE: + case LIO_WRITEV: aio_schedule(job, aio_process_rw); error = 0; break; @@ -2097,6 +2174,13 @@ sys_aio_read(struct thread *td, struct aio_read_args *uap) return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READ, &aiocb_ops)); } +int +sys_aio_readv(struct thread *td, struct aio_readv_args *uap) +{ + + return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READV, &aiocb_ops)); +} + /* syscall - asynchronous write to a file (REALTIME) */ #ifdef COMPAT_FREEBSD6 int @@ -2115,6 +2199,13 @@ sys_aio_write(struct thread *td, struct aio_write_args *uap) return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITE, &aiocb_ops)); } +int +sys_aio_writev(struct thread *td, struct aio_writev_args *uap) +{ + + return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITEV, &aiocb_ops)); +} + int sys_aio_mlock(struct thread *td, struct aio_mlock_args *uap) { @@ -2337,13 +2428,11 @@ sys_lio_listio(struct thread *td, struct lio_listio_args *uap) } static void -aio_biowakeup(struct bio *bp) +aio_biocleanup(struct bio *bp) { struct kaiocb *job = (struct kaiocb *)bp->bio_caller1; struct kaioinfo *ki; - struct buf *pbuf = (struct buf*)bp->bio_caller2; - size_t nbytes; - int error, nblks; + struct buf *pbuf = (struct buf *)bp->bio_caller2; /* Release mapping into kernel space. */ if (pbuf != NULL) { @@ -2362,25 +2451,49 @@ aio_biowakeup(struct bio *bp) free(bp->bio_ma, M_TEMP); atomic_subtract_int(&num_unmapped_aio, 1); } - - nbytes = job->uaiocb.aio_nbytes - bp->bio_resid; - error = 0; - if (bp->bio_flags & BIO_ERROR) - error = bp->bio_error; - nblks = btodb(nbytes); - if (job->uaiocb.aio_lio_opcode == LIO_WRITE) - job->outblock += nblks; - else - job->inblock += nblks; - - if (error) - aio_complete(job, -1, error); - else - aio_complete(job, nbytes, 0); - g_destroy_bio(bp); } +static void +aio_biowakeup(struct bio *bp) +{ + struct kaiocb *job = (struct kaiocb *)bp->bio_caller1; + size_t nbytes; + long bcount = bp->bio_bcount; + long resid = bp->bio_resid; + int error, opcode, nblks; + int bio_error = bp->bio_error; + uint16_t flags = bp->bio_flags; + + opcode = job->uaiocb.aio_lio_opcode; + + aio_biocleanup(bp); + + nbytes =bcount - resid; + atomic_add_acq_long(&job->nbytes, nbytes); + nblks = btodb(nbytes); + error = 0; + /* + * If multiple bios experienced an error, the job will reflect the + * error of whichever failed bio completed last. + */ + if (flags & BIO_ERROR) + atomic_set_int(&job->error, bio_error); + if (opcode == LIO_WRITE || opcode == LIO_WRITEV) + atomic_add_int(&job->outblock, nblks); + else + atomic_add_int(&job->inblock, nblks); + atomic_subtract_int(&job->nbio, 1); + + + if (atomic_load_int(&job->nbio) == 0) { + if (atomic_load_int(&job->error)) + aio_complete(job, -1, job->error); + else + aio_complete(job, atomic_load_long(&job->nbytes), 0); + } +} + /* syscall - wait for the next completion of an aio request */ static int kern_aio_waitcomplete(struct thread *td, struct aiocb **ujobp, @@ -2614,8 +2727,8 @@ typedef struct oaiocb32 { typedef struct aiocb32 { int32_t aio_fildes; /* File descriptor */ uint64_t aio_offset __packed; /* File offset for I/O */ - uint32_t aio_buf; /* I/O buffer in process space */ - uint32_t aio_nbytes; /* Number of bytes for I/O */ + uint32_t aio_buf; /* I/O buffer in process space */ + uint32_t aio_nbytes; /* Number of bytes for I/O */ int __spare__[2]; uint32_t __spare2__; int aio_lio_opcode; /* LIO opcode */ @@ -2652,49 +2765,67 @@ convert_old_sigevent32(struct osigevent32 *osig, struct sigevent *nsig) } static int -aiocb32_copyin_old_sigevent(struct aiocb *ujob, struct aiocb *kjob) +aiocb32_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob, + int type __unused) { struct oaiocb32 job32; + struct aiocb *kcb = &kjob->uaiocb; int error; - bzero(kjob, sizeof(struct aiocb)); + bzero(kcb, sizeof(struct aiocb)); error = copyin(ujob, &job32, sizeof(job32)); if (error) return (error); - CP(job32, *kjob, aio_fildes); - CP(job32, *kjob, aio_offset); - PTRIN_CP(job32, *kjob, aio_buf); - CP(job32, *kjob, aio_nbytes); - CP(job32, *kjob, aio_lio_opcode); - CP(job32, *kjob, aio_reqprio); - CP(job32, *kjob, _aiocb_private.status); - CP(job32, *kjob, _aiocb_private.error); - PTRIN_CP(job32, *kjob, _aiocb_private.kernelinfo); + /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */ + + CP(job32, *kcb, aio_fildes); + CP(job32, *kcb, aio_offset); + PTRIN_CP(job32, *kcb, aio_buf); + CP(job32, *kcb, aio_nbytes); + CP(job32, *kcb, aio_lio_opcode); + CP(job32, *kcb, aio_reqprio); + CP(job32, *kcb, _aiocb_private.status); + CP(job32, *kcb, _aiocb_private.error); + PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo); return (convert_old_sigevent32(&job32.aio_sigevent, - &kjob->aio_sigevent)); + &kcb->aio_sigevent)); } #endif static int -aiocb32_copyin(struct aiocb *ujob, struct aiocb *kjob) +aiocb32_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type) { struct aiocb32 job32; + struct aiocb *kcb = &kjob->uaiocb; + struct iovec32 *iov32; int error; error = copyin(ujob, &job32, sizeof(job32)); if (error) return (error); - CP(job32, *kjob, aio_fildes); - CP(job32, *kjob, aio_offset); - PTRIN_CP(job32, *kjob, aio_buf); - CP(job32, *kjob, aio_nbytes); - CP(job32, *kjob, aio_lio_opcode); - CP(job32, *kjob, aio_reqprio); - CP(job32, *kjob, _aiocb_private.status); - CP(job32, *kjob, _aiocb_private.error); - PTRIN_CP(job32, *kjob, _aiocb_private.kernelinfo); - return (convert_sigevent32(&job32.aio_sigevent, &kjob->aio_sigevent)); + CP(job32, *kcb, aio_fildes); + CP(job32, *kcb, aio_offset); + CP(job32, *kcb, aio_lio_opcode); + if (type == LIO_READV || type == LIO_WRITEV) { + iov32 = PTRIN(job32.aio_iov); + CP(job32, *kcb, aio_iovcnt); + /* malloc a uio and copy in the iovec */ + error = freebsd32_copyinuio(iov32, + kcb->aio_iovcnt, &kjob->uiop); + if (error) + return (error); + } else { + PTRIN_CP(job32, *kcb, aio_buf); + CP(job32, *kcb, aio_nbytes); + } + CP(job32, *kcb, aio_reqprio); + CP(job32, *kcb, _aiocb_private.status); + CP(job32, *kcb, _aiocb_private.error); + PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo); + error = convert_sigevent32(&job32.aio_sigevent, &kcb->aio_sigevent); + + return (error); } static long @@ -2840,6 +2971,14 @@ freebsd32_aio_read(struct thread *td, struct freebsd32_aio_read_args *uap) &aiocb32_ops)); } +int +freebsd32_aio_readv(struct thread *td, struct freebsd32_aio_readv_args *uap) +{ + + return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READV, + &aiocb32_ops)); +} + #ifdef COMPAT_FREEBSD6 int freebsd6_freebsd32_aio_write(struct thread *td, @@ -2859,6 +2998,14 @@ freebsd32_aio_write(struct thread *td, struct freebsd32_aio_write_args *uap) &aiocb32_ops)); } +int +freebsd32_aio_writev(struct thread *td, struct freebsd32_aio_writev_args *uap) +{ + + return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITEV, + &aiocb32_ops)); +} + int freebsd32_aio_mlock(struct thread *td, struct freebsd32_aio_mlock_args *uap) { diff --git a/sys/sys/aio.h b/sys/sys/aio.h index 71afdbf31b35..c0e2b4eaaaf6 100644 --- a/sys/sys/aio.h +++ b/sys/sys/aio.h @@ -27,6 +27,7 @@ #include #include #include +#include #endif /* @@ -45,6 +46,8 @@ #ifdef _KERNEL #define LIO_SYNC 0x3 #define LIO_MLOCK 0x4 +#define LIO_WRITEV 0x5 +#define LIO_READV 0x6 #endif /* @@ -92,7 +95,7 @@ struct __aiocb_private { typedef struct aiocb { int aio_fildes; /* File descriptor */ off_t aio_offset; /* File offset for I/O */ - volatile void *aio_buf; /* I/O buffer in process space */ + volatile void *aio_buf; /* I/O buffer in process space */ size_t aio_nbytes; /* Number of bytes for I/O */ int __spare__[2]; void *__spare2__; @@ -102,6 +105,9 @@ typedef struct aiocb { struct sigevent aio_sigevent; /* Signal to deliver */ } aiocb_t; +#define aio_iov aio_buf /* I/O scatter/gather list */ +#define aio_iovcnt aio_nbytes /* Length of aio_iov */ + #ifdef _KERNEL typedef void aio_cancel_fn_t(struct kaiocb *); @@ -132,11 +138,19 @@ struct kaiocb { struct aiocb *ujob; /* (*) pointer in userspace of aiocb */ struct knlist klist; /* (a) list of knotes */ struct aiocb uaiocb; /* (*) copy of user I/O control block */ + struct uio uio; /* (*) storage for non-vectored uio */ + struct iovec iov[1]; /* (*) storage for non-vectored uio */ + struct uio *uiop; /* (*) Possibly malloced uio */ ksiginfo_t ksi; /* (a) realtime signal info */ uint64_t seqno; /* (*) job number */ aio_cancel_fn_t *cancel_fn; /* (a) backend cancel function */ aio_handle_fn_t *handle_fn; /* (c) backend handle function */ union { /* Backend-specific data fields */ + struct { /* BIO backend */ + int nbio; /* Number of remaining bios */ + int error; /* Worst error of all bios */ + long nbytes; /* Bytes completed so far */ + }; struct { /* fsync() requests */ int pending; /* (a) number of pending I/O */ }; @@ -202,11 +216,17 @@ __BEGIN_DECLS * Asynchronously read from a file */ int aio_read(struct aiocb *); +#if __BSD_VISIBLE +int aio_readv(struct aiocb *); +#endif /* * Asynchronously write to file */ int aio_write(struct aiocb *); +#if __BSD_VISIBLE +int aio_writev(struct aiocb *); +#endif /* * List I/O Asynchronously/synchronously read/write to/from file diff --git a/tests/sys/aio/aio_test.c b/tests/sys/aio/aio_test.c index a9216335d768..891892e5e757 100644 --- a/tests/sys/aio/aio_test.c +++ b/tests/sys/aio/aio_test.c @@ -281,6 +281,47 @@ aio_write_test(struct aio_context *ac, completion comp, struct sigevent *sev) atf_tc_fail("aio short write (%jd)", (intmax_t)len); } +/* + * Perform a vectored I/O test of our initialized data buffer to the provided + * file descriptor. + * + * To vectorize the linear buffer, chop it up into two pieces of dissimilar + * size, and swap their offsets. + */ +static void +aio_writev_test(struct aio_context *ac, completion comp, struct sigevent *sev) +{ + struct aiocb aio; + struct iovec iov[2]; + size_t len0, len1; + ssize_t len; + + bzero(&aio, sizeof(aio)); + + aio.aio_fildes = ac->ac_write_fd; + aio.aio_offset = 0; + len0 = ac->ac_buflen * 3 / 4; + len1 = ac->ac_buflen / 4; + iov[0].iov_base = ac->ac_buffer + len1; + iov[0].iov_len = len0; + iov[1].iov_base = ac->ac_buffer; + iov[1].iov_len = len1; + aio.aio_iov = iov; + aio.aio_iovcnt = 2; + if (sev) + aio.aio_sigevent = *sev; + + if (aio_writev(&aio) < 0) + atf_tc_fail("aio_writev failed: %s", strerror(errno)); + + len = comp(&aio); + if (len < 0) + atf_tc_fail("aio failed: %s", strerror(errno)); + + if (len != ac->ac_buflen) + atf_tc_fail("aio short write (%jd)", (intmax_t)len); +} + /* * Perform a simple read test of our initialized data buffer from the * provided file descriptor. @@ -314,6 +355,43 @@ aio_read_test(struct aio_context *ac, completion comp, struct sigevent *sev) atf_tc_fail("buffer mismatched"); } +static void +aio_readv_test(struct aio_context *ac, completion comp, struct sigevent *sev) +{ + struct aiocb aio; + struct iovec iov[2]; + size_t len0, len1; + ssize_t len; + + bzero(ac->ac_buffer, ac->ac_buflen); + bzero(&aio, sizeof(aio)); + aio.aio_fildes = ac->ac_read_fd; + aio.aio_offset = 0; + len0 = ac->ac_buflen * 3 / 4; + len1 = ac->ac_buflen / 4; + iov[0].iov_base = ac->ac_buffer + len1; + iov[0].iov_len = len0; + iov[1].iov_base = ac->ac_buffer; + iov[1].iov_len = len1; + aio.aio_iov = iov; + aio.aio_iovcnt = 2; + if (sev) + aio.aio_sigevent = *sev; + + if (aio_readv(&aio) < 0) + atf_tc_fail("aio_read failed: %s", strerror(errno)); + + len = comp(&aio); + if (len < 0) + atf_tc_fail("aio failed: %s", strerror(errno)); + + ATF_REQUIRE_EQ_MSG(len, ac->ac_buflen, + "aio short read (%jd)", (intmax_t)len); + + if (aio_test_buffer(ac->ac_buffer, ac->ac_buflen, ac->ac_seed) == 0) + atf_tc_fail("buffer mismatched"); +} + /* * Series of type-specific tests for AIO. For now, we just make sure we can * issue a write and then a read to each type. We assume that once a write @@ -328,7 +406,7 @@ aio_read_test(struct aio_context *ac, completion comp, struct sigevent *sev) #define FILE_PATHNAME "testfile" static void -aio_file_test(completion comp, struct sigevent *sev) +aio_file_test(completion comp, struct sigevent *sev, bool vectored) { struct aio_context ac; int fd; @@ -340,39 +418,44 @@ aio_file_test(completion comp, struct sigevent *sev) ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); aio_context_init(&ac, fd, fd, FILE_LEN); - aio_write_test(&ac, comp, sev); - aio_read_test(&ac, comp, sev); + if (vectored) { + aio_writev_test(&ac, comp, sev); + aio_readv_test(&ac, comp, sev); + } else { + aio_write_test(&ac, comp, sev); + aio_read_test(&ac, comp, sev); + } close(fd); } ATF_TC_WITHOUT_HEAD(file_poll); ATF_TC_BODY(file_poll, tc) { - aio_file_test(poll, NULL); + aio_file_test(poll, NULL, false); } ATF_TC_WITHOUT_HEAD(file_signal); ATF_TC_BODY(file_signal, tc) { - aio_file_test(poll_signaled, setup_signal()); + aio_file_test(poll_signaled, setup_signal(), false); } ATF_TC_WITHOUT_HEAD(file_suspend); ATF_TC_BODY(file_suspend, tc) { - aio_file_test(suspend, NULL); + aio_file_test(suspend, NULL, false); } ATF_TC_WITHOUT_HEAD(file_thread); ATF_TC_BODY(file_thread, tc) { - aio_file_test(poll_signaled, setup_thread()); + aio_file_test(poll_signaled, setup_thread(), false); } ATF_TC_WITHOUT_HEAD(file_waitcomplete); ATF_TC_BODY(file_waitcomplete, tc) { - aio_file_test(waitcomplete, NULL); + aio_file_test(waitcomplete, NULL, false); } #define FIFO_LEN 256 @@ -446,7 +529,7 @@ ATF_TC_BODY(fifo_waitcomplete, tc) #define UNIX_SOCKETPAIR_LEN 256 static void -aio_unix_socketpair_test(completion comp, struct sigevent *sev) +aio_unix_socketpair_test(completion comp, struct sigevent *sev, bool vectored) { struct aio_context ac; struct rusage ru_before, ru_after; @@ -460,14 +543,16 @@ aio_unix_socketpair_test(completion comp, struct sigevent *sev) aio_context_init(&ac, sockets[0], sockets[1], UNIX_SOCKETPAIR_LEN); ATF_REQUIRE_MSG(getrusage(RUSAGE_SELF, &ru_before) != -1, "getrusage failed: %s", strerror(errno)); - aio_write_test(&ac, comp, sev); + if (vectored) { + aio_writev_test(&ac, comp, sev); + aio_readv_test(&ac, comp, sev); + } else { + aio_write_test(&ac, comp, sev); + aio_read_test(&ac, comp, sev); + } ATF_REQUIRE_MSG(getrusage(RUSAGE_SELF, &ru_after) != -1, "getrusage failed: %s", strerror(errno)); ATF_REQUIRE(ru_after.ru_msgsnd == ru_before.ru_msgsnd + 1); - ru_before = ru_after; - aio_read_test(&ac, comp, sev); - ATF_REQUIRE_MSG(getrusage(RUSAGE_SELF, &ru_after) != -1, - "getrusage failed: %s", strerror(errno)); ATF_REQUIRE(ru_after.ru_msgrcv == ru_before.ru_msgrcv + 1); close(sockets[0]); @@ -477,31 +562,31 @@ aio_unix_socketpair_test(completion comp, struct sigevent *sev) ATF_TC_WITHOUT_HEAD(socket_poll); ATF_TC_BODY(socket_poll, tc) { - aio_unix_socketpair_test(poll, NULL); + aio_unix_socketpair_test(poll, NULL, false); } ATF_TC_WITHOUT_HEAD(socket_signal); ATF_TC_BODY(socket_signal, tc) { - aio_unix_socketpair_test(poll_signaled, setup_signal()); + aio_unix_socketpair_test(poll_signaled, setup_signal(), false); } ATF_TC_WITHOUT_HEAD(socket_suspend); ATF_TC_BODY(socket_suspend, tc) { - aio_unix_socketpair_test(suspend, NULL); + aio_unix_socketpair_test(suspend, NULL, false); } ATF_TC_WITHOUT_HEAD(socket_thread); ATF_TC_BODY(socket_thread, tc) { - aio_unix_socketpair_test(poll_signaled, setup_thread()); + aio_unix_socketpair_test(poll_signaled, setup_thread(), false); } ATF_TC_WITHOUT_HEAD(socket_waitcomplete); ATF_TC_BODY(socket_waitcomplete, tc) { - aio_unix_socketpair_test(waitcomplete, NULL); + aio_unix_socketpair_test(waitcomplete, NULL, false); } struct aio_pty_arg { @@ -629,40 +714,11 @@ ATF_TC_BODY(pipe_waitcomplete, tc) #define MD_LEN GLOBAL_MAX #define MDUNIT_LINK "mdunit_link" -static void -aio_md_cleanup(void) -{ - struct md_ioctl mdio; - int mdctl_fd, error, n, unit; - char buf[80]; - - mdctl_fd = open("/dev/" MDCTL_NAME, O_RDWR, 0); - ATF_REQUIRE(mdctl_fd >= 0); - n = readlink(MDUNIT_LINK, buf, sizeof(buf)); - if (n > 0) { - if (sscanf(buf, "%d", &unit) == 1 && unit >= 0) { - bzero(&mdio, sizeof(mdio)); - mdio.md_version = MDIOVERSION; - mdio.md_unit = unit; - if (ioctl(mdctl_fd, MDIOCDETACH, &mdio) == -1) { - error = errno; - close(mdctl_fd); - errno = error; - atf_tc_fail("ioctl MDIOCDETACH failed: %s", - strerror(errno)); - } - } - } - - close(mdctl_fd); -} - -static void -aio_md_test(completion comp, struct sigevent *sev) +static int +aio_md_setup(void) { int error, fd, mdctl_fd, unit; char pathname[PATH_MAX]; - struct aio_context ac; struct md_ioctl mdio; char buf[80]; @@ -695,9 +751,52 @@ aio_md_test(completion comp, struct sigevent *sev) ATF_REQUIRE_MSG(fd != -1, "opening %s failed: %s", pathname, strerror(errno)); + return (fd); +} + +static void +aio_md_cleanup(void) +{ + struct md_ioctl mdio; + int mdctl_fd, error, n, unit; + char buf[80]; + + mdctl_fd = open("/dev/" MDCTL_NAME, O_RDWR, 0); + ATF_REQUIRE(mdctl_fd >= 0); + n = readlink(MDUNIT_LINK, buf, sizeof(buf)); + if (n > 0) { + if (sscanf(buf, "%d", &unit) == 1 && unit >= 0) { + bzero(&mdio, sizeof(mdio)); + mdio.md_version = MDIOVERSION; + mdio.md_unit = unit; + if (ioctl(mdctl_fd, MDIOCDETACH, &mdio) == -1) { + error = errno; + close(mdctl_fd); + errno = error; + atf_tc_fail("ioctl MDIOCDETACH failed: %s", + strerror(errno)); + } + } + } + + close(mdctl_fd); +} + +static void +aio_md_test(completion comp, struct sigevent *sev, bool vectored) +{ + struct aio_context ac; + int fd; + + fd = aio_md_setup(); aio_context_init(&ac, fd, fd, MD_LEN); - aio_write_test(&ac, comp, sev); - aio_read_test(&ac, comp, sev); + if (vectored) { + aio_writev_test(&ac, comp, sev); + aio_readv_test(&ac, comp, sev); + } else { + aio_write_test(&ac, comp, sev); + aio_read_test(&ac, comp, sev); + } close(fd); } @@ -710,7 +809,7 @@ ATF_TC_HEAD(md_poll, tc) } ATF_TC_BODY(md_poll, tc) { - aio_md_test(poll, NULL); + aio_md_test(poll, NULL, false); } ATF_TC_CLEANUP(md_poll, tc) { @@ -725,7 +824,7 @@ ATF_TC_HEAD(md_signal, tc) } ATF_TC_BODY(md_signal, tc) { - aio_md_test(poll_signaled, setup_signal()); + aio_md_test(poll_signaled, setup_signal(), false); } ATF_TC_CLEANUP(md_signal, tc) { @@ -740,7 +839,7 @@ ATF_TC_HEAD(md_suspend, tc) } ATF_TC_BODY(md_suspend, tc) { - aio_md_test(suspend, NULL); + aio_md_test(suspend, NULL, false); } ATF_TC_CLEANUP(md_suspend, tc) { @@ -755,7 +854,7 @@ ATF_TC_HEAD(md_thread, tc) } ATF_TC_BODY(md_thread, tc) { - aio_md_test(poll_signaled, setup_thread()); + aio_md_test(poll_signaled, setup_thread(), false); } ATF_TC_CLEANUP(md_thread, tc) { @@ -770,13 +869,89 @@ ATF_TC_HEAD(md_waitcomplete, tc) } ATF_TC_BODY(md_waitcomplete, tc) { - aio_md_test(waitcomplete, NULL); + aio_md_test(waitcomplete, NULL, false); } ATF_TC_CLEANUP(md_waitcomplete, tc) { aio_md_cleanup(); } +#define ZVOL_VDEV_PATHNAME "test_vdev" +#define POOL_SIZE (1 << 28) /* 256 MB */ +#define ZVOL_SIZE "64m" +#define POOL_NAME "aio_testpool" +#define ZVOL_NAME "aio_testvol" + +static int +aio_zvol_setup(void) +{ + FILE *pidfile; + int fd; + pid_t pid; + char pool_name[80]; + char cmd[160]; + char zvol_name[160]; + char devname[160]; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_KERNEL_MODULE("zfs"); + + fd = open(ZVOL_VDEV_PATHNAME, O_RDWR | O_CREAT, 0600); + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + ATF_REQUIRE_EQ_MSG(0, + ftruncate(fd, POOL_SIZE), "ftruncate failed: %s", strerror(errno)); + close(fd); + + pid = getpid(); + pidfile = fopen("pidfile", "w"); + ATF_REQUIRE_MSG(NULL != pidfile, "fopen: %s", strerror(errno)); + fprintf(pidfile, "%d", pid); + fclose(pidfile); + + snprintf(pool_name, sizeof(pool_name), POOL_NAME ".%d", pid); + snprintf(zvol_name, sizeof(zvol_name), "%s/" ZVOL_NAME, pool_name); + snprintf(cmd, sizeof(cmd), "zpool create %s $PWD/" ZVOL_VDEV_PATHNAME, + pool_name); + ATF_REQUIRE_EQ_MSG(0, system(cmd), + "zpool create failed: %s", strerror(errno)); + snprintf(cmd, sizeof(cmd), + "zfs create -o volblocksize=8192 -o volmode=dev -V " + ZVOL_SIZE " %s", zvol_name); + ATF_REQUIRE_EQ_MSG(0, system(cmd), + "zfs create failed: %s", strerror(errno)); + /* + * XXX Due to bug 251828, we need an extra "zfs set" here + * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=251828 + */ + snprintf(cmd, sizeof(cmd), "zfs set volmode=dev %s", zvol_name); + ATF_REQUIRE_EQ_MSG(0, system(cmd), + "zfs set failed: %s", strerror(errno)); + + snprintf(devname, sizeof(devname), "/dev/zvol/%s", zvol_name); + do { + fd = open(devname, O_RDWR); + } while (fd == -1 && errno == EINTR) ; + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + return (fd); +} + +static void +aio_zvol_cleanup(void) +{ + FILE *pidfile; + pid_t testpid; + char cmd[160]; + + pidfile = fopen("pidfile", "r"); + ATF_REQUIRE_MSG(NULL != pidfile, "fopen: %s", strerror(errno)); + ATF_REQUIRE_EQ(1, fscanf(pidfile, "%d", &testpid)); + fclose(pidfile); + + snprintf(cmd, sizeof(cmd), "zpool destroy " POOL_NAME ".%d", testpid); + system(cmd); +} + + ATF_TC_WITHOUT_HEAD(aio_large_read_test); ATF_TC_BODY(aio_large_read_test, tc) { @@ -907,16 +1082,13 @@ ATF_TC_BODY(aio_socket_two_reads, tc) close(s[0]); } -/* - * This test ensures that aio_write() on a blocking socket of a "large" - * buffer does not return a short completion. - */ -ATF_TC_WITHOUT_HEAD(aio_socket_blocking_short_write); -ATF_TC_BODY(aio_socket_blocking_short_write, tc) +static void +aio_socket_blocking_short_write_test(bool vectored) { struct aiocb iocb, *iocbp; + struct iovec iov[2]; char *buffer[2]; - ssize_t done; + ssize_t done, r; int buffer_size, sb_size; socklen_t len; int s[2]; @@ -954,9 +1126,21 @@ ATF_TC_BODY(aio_socket_blocking_short_write, tc) memset(&iocb, 0, sizeof(iocb)); iocb.aio_fildes = s[1]; - iocb.aio_buf = buffer[1]; - iocb.aio_nbytes = buffer_size; - ATF_REQUIRE(aio_write(&iocb) == 0); + if (vectored) { + iov[0].iov_base = buffer[1]; + iov[0].iov_len = buffer_size / 2 + 1; + iov[1].iov_base = buffer[1] + buffer_size / 2 + 1; + iov[1].iov_len = buffer_size / 2 - 1; + iocb.aio_iov = iov; + iocb.aio_iovcnt = 2; + r = aio_writev(&iocb); + ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r); + } else { + iocb.aio_buf = buffer[1]; + iocb.aio_nbytes = buffer_size; + r = aio_write(&iocb); + ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r); + } done = recv(s[0], buffer[0], buffer_size, MSG_WAITALL); ATF_REQUIRE(done == buffer_size); @@ -971,6 +1155,26 @@ ATF_TC_BODY(aio_socket_blocking_short_write, tc) close(s[0]); } +/* + * This test ensures that aio_write() on a blocking socket of a "large" + * buffer does not return a short completion. + */ +ATF_TC_WITHOUT_HEAD(aio_socket_blocking_short_write); +ATF_TC_BODY(aio_socket_blocking_short_write, tc) +{ + aio_socket_blocking_short_write_test(false); +} + +/* + * Like aio_socket_blocking_short_write, but also tests that partially + * completed vectored sends can be retried correctly. + */ +ATF_TC_WITHOUT_HEAD(aio_socket_blocking_short_write_vectored); +ATF_TC_BODY(aio_socket_blocking_short_write_vectored, tc) +{ + aio_socket_blocking_short_write_test(true); +} + /* * This test verifies that cancelling a partially completed socket write * returns a short write rather than ECANCELED. @@ -1155,6 +1359,395 @@ ATF_TC_BODY(aio_fsync_test, tc) close(fd); } +/* + * We shouldn't be able to DoS the system by setting iov_len to an insane + * value + */ +ATF_TC_WITHOUT_HEAD(aio_writev_dos_iov_len); +ATF_TC_BODY(aio_writev_dos_iov_len, tc) +{ + struct aiocb aio; + const struct aiocb *const iocbs[] = {&aio}; + const char *wbuf = "Hello, world!"; + struct iovec iov[1]; + ssize_t len, r; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + fd = open("testfile", O_RDWR | O_CREAT, 0600); + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + + len = strlen(wbuf); + iov[0].iov_base = __DECONST(void*, wbuf); + iov[0].iov_len = 1 << 30; + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = 1; + + r = aio_writev(&aio); + ATF_CHECK_EQ_MSG(0, r, "aio_writev returned %zd", r); + ATF_REQUIRE_EQ(0, aio_suspend(iocbs, 1, NULL)); + r = aio_return(&aio); + ATF_CHECK_EQ_MSG(-1, r, "aio_return returned %zd", r); + ATF_CHECK_MSG(errno == EFAULT || errno == EINVAL, + "aio_writev: %s", strerror(errno)); + + close(fd); +} + +/* + * We shouldn't be able to DoS the system by setting aio_iovcnt to an insane + * value + */ +ATF_TC_WITHOUT_HEAD(aio_writev_dos_iovcnt); +ATF_TC_BODY(aio_writev_dos_iovcnt, tc) +{ + struct aiocb aio; + const char *wbuf = "Hello, world!"; + struct iovec iov[1]; + ssize_t len; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + fd = open("testfile", O_RDWR | O_CREAT, 0600); + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + + len = strlen(wbuf); + iov[0].iov_base = __DECONST(void*, wbuf); + iov[0].iov_len = len; + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = 1 << 30; + + ATF_REQUIRE_EQ(-1, aio_writev(&aio)); + ATF_CHECK_EQ(EINVAL, errno); + + close(fd); +} + +ATF_TC_WITH_CLEANUP(aio_writev_efault); +ATF_TC_HEAD(aio_writev_efault, tc) +{ + atf_tc_set_md_var(tc, "descr", + "Vectored AIO should gracefully handle invalid addresses"); + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(aio_writev_efault, tc) +{ + struct aiocb aio; + ssize_t buflen; + char *buffer; + struct iovec iov[2]; + long seed; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + fd = aio_md_setup(); + + seed = random(); + buflen = 4096; + buffer = malloc(buflen); + aio_fill_buffer(buffer, buflen, seed); + iov[0].iov_base = buffer; + iov[0].iov_len = buflen; + iov[1].iov_base = (void*)-1; /* Invalid! */ + iov[1].iov_len = buflen; + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = nitems(iov); + + ATF_REQUIRE_EQ(-1, aio_writev(&aio)); + ATF_CHECK_EQ(EFAULT, errno); + + close(fd); +} +ATF_TC_CLEANUP(aio_writev_efault, tc) +{ + aio_md_cleanup(); +} + +ATF_TC_WITHOUT_HEAD(aio_writev_empty_file_poll); +ATF_TC_BODY(aio_writev_empty_file_poll, tc) +{ + struct aiocb aio; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + fd = open("testfile", O_RDWR | O_CREAT, 0600); + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iovcnt = 0; + + ATF_REQUIRE_EQ(0, aio_writev(&aio)); + ATF_REQUIRE_EQ(0, suspend(&aio)); + + close(fd); +} + +ATF_TC_WITHOUT_HEAD(aio_writev_empty_file_signal); +ATF_TC_BODY(aio_writev_empty_file_signal, tc) +{ + struct aiocb aio; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + fd = open("testfile", O_RDWR | O_CREAT, 0600); + ATF_REQUIRE_MSG(fd != -1, "open failed: %s", strerror(errno)); + + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iovcnt = 0; + aio.aio_sigevent = *setup_signal(); + + ATF_REQUIRE_EQ(0, aio_writev(&aio)); + ATF_REQUIRE_EQ(0, poll_signaled(&aio)); + + close(fd); +} + +// aio_writev and aio_readv should still work even if the iovcnt is greater +// than the number of buffered AIO operations permitted per process. +ATF_TC_WITH_CLEANUP(vectored_big_iovcnt); +ATF_TC_HEAD(vectored_big_iovcnt, tc) +{ + atf_tc_set_md_var(tc, "descr", + "Vectored AIO should still work even if the iovcnt is greater than " + "the number of buffered AIO operations permitted by the process"); + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(vectored_big_iovcnt, tc) +{ + struct aiocb aio; + struct iovec *iov; + ssize_t len, buflen; + char *buffer; + const char *oid = "vfs.aio.max_buf_aio"; + long seed; + int max_buf_aio; + int fd, i; + ssize_t sysctl_len = sizeof(max_buf_aio); + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + if (sysctlbyname(oid, &max_buf_aio, &sysctl_len, NULL, 0) == -1) + atf_libc_error(errno, "Failed to read %s", oid); + + seed = random(); + buflen = 512 * (max_buf_aio + 1); + buffer = malloc(buflen); + aio_fill_buffer(buffer, buflen, seed); + iov = calloc(max_buf_aio + 1, sizeof(struct iovec)); + + fd = aio_md_setup(); + + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + for (i = 0; i < max_buf_aio + 1; i++) { + iov[i].iov_base = &buffer[i * 512]; + iov[i].iov_len = 512; + } + aio.aio_iov = iov; + aio.aio_iovcnt = max_buf_aio + 1; + + if (aio_writev(&aio) < 0) + atf_tc_fail("aio_writev failed: %s", strerror(errno)); + + len = poll(&aio); + if (len < 0) + atf_tc_fail("aio failed: %s", strerror(errno)); + + if (len != buflen) + atf_tc_fail("aio short write (%jd)", (intmax_t)len); + + bzero(&aio, sizeof(aio)); + aio.aio_fildes = fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = max_buf_aio + 1; + + if (aio_readv(&aio) < 0) + atf_tc_fail("aio_readv failed: %s", strerror(errno)); + + len = poll(&aio); + if (len < 0) + atf_tc_fail("aio failed: %s", strerror(errno)); + + if (len != buflen) + atf_tc_fail("aio short read (%jd)", (intmax_t)len); + + if (aio_test_buffer(buffer, buflen, seed) == 0) + atf_tc_fail("buffer mismatched"); + + close(fd); +} +ATF_TC_CLEANUP(vectored_big_iovcnt, tc) +{ + aio_md_cleanup(); +} + +ATF_TC_WITHOUT_HEAD(vectored_file_poll); +ATF_TC_BODY(vectored_file_poll, tc) +{ + aio_file_test(poll, NULL, true); +} + +ATF_TC_WITH_CLEANUP(vectored_md_poll); +ATF_TC_HEAD(vectored_md_poll, tc) +{ + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(vectored_md_poll, tc) +{ + aio_md_test(poll, NULL, true); +} +ATF_TC_CLEANUP(vectored_md_poll, tc) +{ + aio_md_cleanup(); +} + +ATF_TC_WITHOUT_HEAD(vectored_socket_poll); +ATF_TC_BODY(vectored_socket_poll, tc) +{ + aio_unix_socketpair_test(poll, NULL, true); +} + +// aio_writev and aio_readv should still work even if the iov contains elements +// that aren't a multiple of the device's sector size, and even if the total +// amount if I/O _is_ a multiple of the device's sector size. +ATF_TC_WITH_CLEANUP(vectored_unaligned); +ATF_TC_HEAD(vectored_unaligned, tc) +{ + atf_tc_set_md_var(tc, "descr", + "Vectored AIO should still work even if the iov contains elements " + "that aren't a multiple of the sector size."); + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(vectored_unaligned, tc) +{ + struct aio_context ac; + struct aiocb aio; + struct iovec iov[3]; + ssize_t len, total_len; + int fd; + + ATF_REQUIRE_KERNEL_MODULE("aio"); + ATF_REQUIRE_UNSAFE_AIO(); + + /* + * Use a zvol with volmode=dev, so it will allow .d_write with + * unaligned uio. geom devices use physio, which doesn't allow that. + */ + fd = aio_zvol_setup(); + aio_context_init(&ac, fd, fd, FILE_LEN); + + /* Break the buffer into 3 parts: + * * A 4kB part, aligned to 4kB + * * Two other parts that add up to 4kB: + * - 256B + * - 4kB - 256B + */ + iov[0].iov_base = ac.ac_buffer; + iov[0].iov_len = 4096; + iov[1].iov_base = (void*)((uintptr_t)iov[0].iov_base + iov[0].iov_len); + iov[1].iov_len = 256; + iov[2].iov_base = (void*)((uintptr_t)iov[1].iov_base + iov[1].iov_len); + iov[2].iov_len = 4096 - iov[1].iov_len; + total_len = iov[0].iov_len + iov[1].iov_len + iov[2].iov_len; + bzero(&aio, sizeof(aio)); + aio.aio_fildes = ac.ac_write_fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = 3; + + if (aio_writev(&aio) < 0) + atf_tc_fail("aio_writev failed: %s", strerror(errno)); + + len = poll(&aio); + if (len < 0) + atf_tc_fail("aio failed: %s", strerror(errno)); + + if (len != total_len) + atf_tc_fail("aio short write (%jd)", (intmax_t)len); + + bzero(&aio, sizeof(aio)); + aio.aio_fildes = ac.ac_read_fd; + aio.aio_offset = 0; + aio.aio_iov = iov; + aio.aio_iovcnt = 3; + + if (aio_readv(&aio) < 0) + atf_tc_fail("aio_readv failed: %s", strerror(errno)); + len = poll(&aio); + + ATF_REQUIRE_MSG(aio_test_buffer(ac.ac_buffer, total_len, + ac.ac_seed) != 0, "aio_test_buffer: internal error"); + + close(fd); +} +ATF_TC_CLEANUP(vectored_unaligned, tc) +{ + aio_zvol_cleanup(); +} + +static void +aio_zvol_test(completion comp, struct sigevent *sev, bool vectored) +{ + struct aio_context ac; + int fd; + + fd = aio_zvol_setup(); + aio_context_init(&ac, fd, fd, MD_LEN); + if (vectored) { + aio_writev_test(&ac, comp, sev); + aio_readv_test(&ac, comp, sev); + } else { + aio_write_test(&ac, comp, sev); + aio_read_test(&ac, comp, sev); + } + + close(fd); +} + +/* + * Note that unlike md, the zvol is not a geom device, does not allow unmapped + * buffers, and does not use physio. + */ +ATF_TC_WITH_CLEANUP(vectored_zvol_poll); +ATF_TC_HEAD(vectored_zvol_poll, tc) +{ + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(vectored_zvol_poll, tc) +{ + aio_zvol_test(poll, NULL, true); +} +ATF_TC_CLEANUP(vectored_zvol_poll, tc) +{ + aio_zvol_cleanup(); +} + ATF_TP_ADD_TCS(tp) { @@ -1193,7 +1786,19 @@ ATF_TP_ADD_TCS(tp) ATF_TP_ADD_TC(tp, aio_large_read_test); ATF_TP_ADD_TC(tp, aio_socket_two_reads); ATF_TP_ADD_TC(tp, aio_socket_blocking_short_write); + ATF_TP_ADD_TC(tp, aio_socket_blocking_short_write_vectored); ATF_TP_ADD_TC(tp, aio_socket_short_write_cancel); + ATF_TP_ADD_TC(tp, aio_writev_dos_iov_len); + ATF_TP_ADD_TC(tp, aio_writev_dos_iovcnt); + ATF_TP_ADD_TC(tp, aio_writev_efault); + ATF_TP_ADD_TC(tp, aio_writev_empty_file_poll); + ATF_TP_ADD_TC(tp, aio_writev_empty_file_signal); + ATF_TP_ADD_TC(tp, vectored_big_iovcnt); + ATF_TP_ADD_TC(tp, vectored_file_poll); + ATF_TP_ADD_TC(tp, vectored_md_poll); + ATF_TP_ADD_TC(tp, vectored_zvol_poll); + ATF_TP_ADD_TC(tp, vectored_unaligned); + ATF_TP_ADD_TC(tp, vectored_socket_poll); return (atf_no_error()); }