/* Unix SMB/Netbios implementation. Version 3.0 async_io read handling using POSIX async io. Copyright (C) Jeremy Allison 2005. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see . */ #include "includes.h" #if defined(WITH_AIO) /* The signal we'll use to signify aio done. */ #ifndef RT_SIGNAL_AIO #define RT_SIGNAL_AIO (SIGRTMIN+3) #endif /**************************************************************************** The buffer we keep around whilst an aio request is in process. *****************************************************************************/ struct aio_extra { struct aio_extra *next, *prev; SMB_STRUCT_AIOCB acb; files_struct *fsp; bool read_req; uint16 mid; char *inbuf; char *outbuf; }; static struct aio_extra *aio_list_head; /**************************************************************************** Create the extended aio struct we must keep around for the lifetime of the aio_read call. *****************************************************************************/ static struct aio_extra *create_aio_ex_read(files_struct *fsp, size_t buflen, uint16 mid) { struct aio_extra *aio_ex = SMB_MALLOC_P(struct aio_extra); if (!aio_ex) { return NULL; } ZERO_STRUCTP(aio_ex); /* The output buffer stored in the aio_ex is the start of the smb return buffer. The buffer used in the acb is the start of the reply data portion of that buffer. */ aio_ex->outbuf = SMB_MALLOC_ARRAY(char, buflen); if (!aio_ex->outbuf) { SAFE_FREE(aio_ex); return NULL; } DLIST_ADD(aio_list_head, aio_ex); aio_ex->fsp = fsp; aio_ex->read_req = True; aio_ex->mid = mid; return aio_ex; } /**************************************************************************** Create the extended aio struct we must keep around for the lifetime of the aio_write call. *****************************************************************************/ static struct aio_extra *create_aio_ex_write(files_struct *fsp, size_t inbuflen, size_t outbuflen, uint16 mid) { struct aio_extra *aio_ex = SMB_MALLOC_P(struct aio_extra); if (!aio_ex) { return NULL; } ZERO_STRUCTP(aio_ex); /* We need space for an output reply of outbuflen bytes. */ aio_ex->outbuf = SMB_MALLOC_ARRAY(char, outbuflen); if (!aio_ex->outbuf) { SAFE_FREE(aio_ex); return NULL; } if (!(aio_ex->inbuf = SMB_MALLOC_ARRAY(char, inbuflen))) { SAFE_FREE(aio_ex->outbuf); SAFE_FREE(aio_ex); return NULL; } DLIST_ADD(aio_list_head, aio_ex); aio_ex->fsp = fsp; aio_ex->read_req = False; aio_ex->mid = mid; return aio_ex; } /**************************************************************************** Delete the extended aio struct. *****************************************************************************/ static void delete_aio_ex(struct aio_extra *aio_ex) { DLIST_REMOVE(aio_list_head, aio_ex); SAFE_FREE(aio_ex->inbuf); SAFE_FREE(aio_ex->outbuf); SAFE_FREE(aio_ex); } /**************************************************************************** Given the aiocb struct find the extended aio struct containing it. *****************************************************************************/ static struct aio_extra *find_aio_ex(uint16 mid) { struct aio_extra *p; for( p = aio_list_head; p; p = p->next) { if (mid == p->mid) { return p; } } return NULL; } /**************************************************************************** We can have these many aio buffers in flight. *****************************************************************************/ #define AIO_PENDING_SIZE 10 static sig_atomic_t signals_received; static int outstanding_aio_calls; static uint16 aio_pending_array[AIO_PENDING_SIZE]; /**************************************************************************** Signal handler when an aio request completes. *****************************************************************************/ static void signal_handler(int sig, siginfo_t *info, void *unused) { if (signals_received < AIO_PENDING_SIZE) { aio_pending_array[signals_received] = info->si_value.sival_int; signals_received++; } /* Else signal is lost. */ sys_select_signal(RT_SIGNAL_AIO); } /**************************************************************************** Is there a signal waiting ? *****************************************************************************/ bool aio_finished(void) { return (signals_received != 0); } /**************************************************************************** Initialize the signal handler for aio read/write. *****************************************************************************/ void initialize_async_io_handler(void) { struct sigaction act; ZERO_STRUCT(act); act.sa_sigaction = signal_handler; act.sa_flags = SA_SIGINFO; sigemptyset( &act.sa_mask ); if (sigaction(RT_SIGNAL_AIO, &act, NULL) != 0) { DEBUG(0,("Failed to setup RT_SIGNAL_AIO handler\n")); } /* the signal can start off blocked due to a bug in bash */ BlockSignals(False, RT_SIGNAL_AIO); } /**************************************************************************** Set up an aio request from a SMBreadX call. *****************************************************************************/ bool schedule_aio_read_and_X(connection_struct *conn, struct smb_request *req, files_struct *fsp, SMB_OFF_T startpos, size_t smb_maxcnt) { struct aio_extra *aio_ex; SMB_STRUCT_AIOCB *a; size_t bufsize; size_t min_aio_read_size = lp_aio_read_size(SNUM(conn)); if ((!min_aio_read_size || (smb_maxcnt < min_aio_read_size)) && !SMB_VFS_AIO_FORCE(fsp)) { /* Too small a read for aio request. */ DEBUG(10,("schedule_aio_read_and_X: read size (%u) too small " "for minimum aio_read of %u\n", (unsigned int)smb_maxcnt, (unsigned int)min_aio_read_size )); return False; } /* Only do this on non-chained and non-chaining reads not using the * write cache. */ if (chain_size !=0 || (CVAL(req->inbuf,smb_vwv0) != 0xFF) || (lp_write_cache_size(SNUM(conn)) != 0) ) { return False; } if (outstanding_aio_calls >= AIO_PENDING_SIZE) { DEBUG(10,("schedule_aio_read_and_X: Already have %d aio " "activities outstanding.\n", outstanding_aio_calls )); return False; } /* The following is safe from integer wrap as we've already checked smb_maxcnt is 128k or less. Wct is 12 for read replies */ bufsize = smb_size + 12 * 2 + smb_maxcnt; if ((aio_ex = create_aio_ex_read(fsp, bufsize, req->mid)) == NULL) { DEBUG(10,("schedule_aio_read_and_X: malloc fail.\n")); return False; } construct_reply_common((char *)req->inbuf, aio_ex->outbuf); srv_set_message(aio_ex->outbuf, 12, 0, True); SCVAL(aio_ex->outbuf,smb_vwv0,0xFF); /* Never a chained reply. */ a = &aio_ex->acb; /* Now set up the aio record for the read call. */ a->aio_fildes = fsp->fh->fd; a->aio_buf = smb_buf(aio_ex->outbuf); a->aio_nbytes = smb_maxcnt; a->aio_offset = startpos; a->aio_sigevent.sigev_notify = SIGEV_SIGNAL; a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO; a->aio_sigevent.sigev_value.sival_int = aio_ex->mid; if (SMB_VFS_AIO_READ(fsp,a) == -1) { DEBUG(0,("schedule_aio_read_and_X: aio_read failed. " "Error %s\n", strerror(errno) )); delete_aio_ex(aio_ex); return False; } DEBUG(10,("schedule_aio_read_and_X: scheduled aio_read for file %s, " "offset %.0f, len = %u (mid = %u)\n", fsp->fsp_name, (double)startpos, (unsigned int)smb_maxcnt, (unsigned int)aio_ex->mid )); srv_defer_sign_response(aio_ex->mid); outstanding_aio_calls++; return True; } /**************************************************************************** Set up an aio request from a SMBwriteX call. *****************************************************************************/ bool schedule_aio_write_and_X(connection_struct *conn, struct smb_request *req, files_struct *fsp, char *data, SMB_OFF_T startpos, size_t numtowrite) { struct aio_extra *aio_ex; SMB_STRUCT_AIOCB *a; size_t inbufsize, outbufsize; bool write_through = BITSETW(req->inbuf+smb_vwv7,0); size_t min_aio_write_size = lp_aio_write_size(SNUM(conn)); if ((!min_aio_write_size || (numtowrite < min_aio_write_size)) && !SMB_VFS_AIO_FORCE(fsp)) { /* Too small a write for aio request. */ DEBUG(10,("schedule_aio_write_and_X: write size (%u) too " "small for minimum aio_write of %u\n", (unsigned int)numtowrite, (unsigned int)min_aio_write_size )); return False; } /* Only do this on non-chained and non-chaining reads not using the * write cache. */ if (chain_size !=0 || (CVAL(req->inbuf,smb_vwv0) != 0xFF) || (lp_write_cache_size(SNUM(conn)) != 0) ) { return False; } if (outstanding_aio_calls >= AIO_PENDING_SIZE) { DEBUG(3,("schedule_aio_write_and_X: Already have %d aio " "activities outstanding.\n", outstanding_aio_calls )); DEBUG(10,("schedule_aio_write_and_X: failed to schedule " "aio_write for file %s, offset %.0f, len = %u " "(mid = %u)\n", fsp->fsp_name, (double)startpos, (unsigned int)numtowrite, (unsigned int)req->mid )); return False; } inbufsize = smb_len(req->inbuf) + 4; reply_outbuf(req, 6, 0); outbufsize = smb_len(req->outbuf) + 4; if (!(aio_ex = create_aio_ex_write(fsp, inbufsize, outbufsize, req->mid))) { DEBUG(0,("schedule_aio_write_and_X: malloc fail.\n")); return False; } /* Copy the SMB header already setup in outbuf. */ memcpy(aio_ex->inbuf, req->inbuf, inbufsize); /* Copy the SMB header already setup in outbuf. */ memcpy(aio_ex->outbuf, req->outbuf, outbufsize); TALLOC_FREE(req->outbuf); SCVAL(aio_ex->outbuf,smb_vwv0,0xFF); /* Never a chained reply. */ a = &aio_ex->acb; /* Now set up the aio record for the write call. */ a->aio_fildes = fsp->fh->fd; a->aio_buf = aio_ex->inbuf + (PTR_DIFF(data, req->inbuf)); a->aio_nbytes = numtowrite; a->aio_offset = startpos; a->aio_sigevent.sigev_notify = SIGEV_SIGNAL; a->aio_sigevent.sigev_signo = RT_SIGNAL_AIO; a->aio_sigevent.sigev_value.sival_int = aio_ex->mid; if (SMB_VFS_AIO_WRITE(fsp,a) == -1) { DEBUG(3,("schedule_aio_wrote_and_X: aio_write failed. " "Error %s\n", strerror(errno) )); delete_aio_ex(aio_ex); return False; } release_level_2_oplocks_on_change(fsp); if (!write_through && !lp_syncalways(SNUM(fsp->conn)) && fsp->aio_write_behind) { /* Lie to the client and immediately claim we finished the * write. */ SSVAL(aio_ex->outbuf,smb_vwv2,numtowrite); SSVAL(aio_ex->outbuf,smb_vwv4,(numtowrite>>16)&1); show_msg(aio_ex->outbuf); if (!srv_send_smb(smbd_server_fd(),aio_ex->outbuf, IS_CONN_ENCRYPTED(fsp->conn))) { exit_server_cleanly("handle_aio_write: srv_send_smb " "failed."); } DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write " "behind for file %s\n", fsp->fsp_name )); } else { srv_defer_sign_response(aio_ex->mid); } outstanding_aio_calls++; DEBUG(10,("schedule_aio_write_and_X: scheduled aio_write for file " "%s, offset %.0f, len = %u (mid = %u) " "outstanding_aio_calls = %d\n", fsp->fsp_name, (double)startpos, (unsigned int)numtowrite, (unsigned int)aio_ex->mid, outstanding_aio_calls )); return True; } /**************************************************************************** Complete the read and return the data or error back to the client. Returns errno or zero if all ok. *****************************************************************************/ static int handle_aio_read_complete(struct aio_extra *aio_ex) { int ret = 0; int outsize; char *outbuf = aio_ex->outbuf; char *data = smb_buf(outbuf); ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb); if (nread < 0) { /* We're relying here on the fact that if the fd is closed then the aio will complete and aio_return will return an error. Hopefully this is true.... JRA. */ /* If errno is ECANCELED then don't return anything to the * client. */ if (errno == ECANCELED) { srv_cancel_sign_response(aio_ex->mid); return 0; } DEBUG( 3,( "handle_aio_read_complete: file %s nread == -1. " "Error = %s\n", aio_ex->fsp->fsp_name, strerror(errno) )); ret = errno; ERROR_NT(map_nt_error_from_unix(ret)); outsize = srv_set_message(outbuf,0,0,true); } else { outsize = srv_set_message(outbuf,12,nread,False); SSVAL(outbuf,smb_vwv2,0xFFFF); /* Remaining - must be * -1. */ SSVAL(outbuf,smb_vwv5,nread); SSVAL(outbuf,smb_vwv6,smb_offset(data,outbuf)); SSVAL(outbuf,smb_vwv7,((nread >> 16) & 1)); SSVAL(smb_buf(outbuf),-2,nread); DEBUG( 3, ( "handle_aio_read_complete file %s max=%d " "nread=%d\n", aio_ex->fsp->fsp_name, (int)aio_ex->acb.aio_nbytes, (int)nread ) ); } smb_setlen(outbuf,outsize - 4); show_msg(outbuf); if (!srv_send_smb(smbd_server_fd(),outbuf, IS_CONN_ENCRYPTED(aio_ex->fsp->conn))) { exit_server_cleanly("handle_aio_read_complete: srv_send_smb " "failed."); } DEBUG(10,("handle_aio_read_complete: scheduled aio_read completed " "for file %s, offset %.0f, len = %u\n", aio_ex->fsp->fsp_name, (double)aio_ex->acb.aio_offset, (unsigned int)nread )); return ret; } /**************************************************************************** Complete the write and return the data or error back to the client. Returns errno or zero if all ok. *****************************************************************************/ static int handle_aio_write_complete(struct aio_extra *aio_ex) { int ret = 0; files_struct *fsp = aio_ex->fsp; char *outbuf = aio_ex->outbuf; ssize_t numtowrite = aio_ex->acb.aio_nbytes; ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb); if (fsp->aio_write_behind) { if (nwritten != numtowrite) { if (nwritten == -1) { DEBUG(5,("handle_aio_write_complete: " "aio_write_behind failed ! File %s " "is corrupt ! Error %s\n", fsp->fsp_name, strerror(errno) )); ret = errno; } else { DEBUG(0,("handle_aio_write_complete: " "aio_write_behind failed ! File %s " "is corrupt ! Wanted %u bytes but " "only wrote %d\n", fsp->fsp_name, (unsigned int)numtowrite, (int)nwritten )); ret = EIO; } } else { DEBUG(10,("handle_aio_write_complete: " "aio_write_behind completed for file %s\n", fsp->fsp_name )); } return 0; } /* We don't need outsize or set_message here as we've already set the fixed size length when we set up the aio call. */ if(nwritten == -1) { DEBUG( 3,( "handle_aio_write: file %s wanted %u bytes. " "nwritten == %d. Error = %s\n", fsp->fsp_name, (unsigned int)numtowrite, (int)nwritten, strerror(errno) )); /* If errno is ECANCELED then don't return anything to the * client. */ if (errno == ECANCELED) { srv_cancel_sign_response(aio_ex->mid); return 0; } ret = errno; ERROR_BOTH(map_nt_error_from_unix(ret), ERRHRD, ERRdiskfull); srv_set_message(outbuf,0,0,true); } else { bool write_through = BITSETW(aio_ex->inbuf+smb_vwv7,0); NTSTATUS status; SSVAL(outbuf,smb_vwv2,nwritten); SSVAL(outbuf,smb_vwv4,(nwritten>>16)&1); if (nwritten < (ssize_t)numtowrite) { SCVAL(outbuf,smb_rcls,ERRHRD); SSVAL(outbuf,smb_err,ERRdiskfull); } DEBUG(3,("handle_aio_write: fnum=%d num=%d wrote=%d\n", fsp->fnum, (int)numtowrite, (int)nwritten)); status = sync_file(fsp->conn,fsp, write_through); if (!NT_STATUS_IS_OK(status)) { ret = errno; ERROR_BOTH(map_nt_error_from_unix(ret), ERRHRD, ERRdiskfull); srv_set_message(outbuf,0,0,true); DEBUG(5,("handle_aio_write: sync_file for %s returned %s\n", fsp->fsp_name, nt_errstr(status) )); } } show_msg(outbuf); if (!srv_send_smb(smbd_server_fd(),outbuf,IS_CONN_ENCRYPTED(fsp->conn))) { exit_server_cleanly("handle_aio_write: srv_send_smb failed."); } DEBUG(10,("handle_aio_write_complete: scheduled aio_write completed " "for file %s, offset %.0f, requested %u, written = %u\n", fsp->fsp_name, (double)aio_ex->acb.aio_offset, (unsigned int)numtowrite, (unsigned int)nwritten )); return ret; } /**************************************************************************** Handle any aio completion. Returns True if finished (and sets *perr if err was non-zero), False if not. *****************************************************************************/ static bool handle_aio_completed(struct aio_extra *aio_ex, int *perr) { int err; /* Ensure the operation has really completed. */ if (SMB_VFS_AIO_ERROR(aio_ex->fsp, &aio_ex->acb) == EINPROGRESS) { DEBUG(10,( "handle_aio_completed: operation mid %u still in " "process for file %s\n", aio_ex->mid, aio_ex->fsp->fsp_name )); return False; } if (aio_ex->read_req) { err = handle_aio_read_complete(aio_ex); } else { err = handle_aio_write_complete(aio_ex); } if (err) { *perr = err; /* Only save non-zero errors. */ } return True; } /**************************************************************************** Handle any aio completion inline. Returns non-zero errno if fail or zero if all ok. *****************************************************************************/ int process_aio_queue(void) { int i; int ret = 0; BlockSignals(True, RT_SIGNAL_AIO); DEBUG(10,("process_aio_queue: signals_received = %d\n", (int)signals_received)); DEBUG(10,("process_aio_queue: outstanding_aio_calls = %d\n", outstanding_aio_calls)); if (!signals_received) { BlockSignals(False, RT_SIGNAL_AIO); return 0; } /* Drain all the complete aio_reads. */ for (i = 0; i < signals_received; i++) { uint16 mid = aio_pending_array[i]; files_struct *fsp = NULL; struct aio_extra *aio_ex = find_aio_ex(mid); if (!aio_ex) { DEBUG(3,("process_aio_queue: Can't find record to " "match mid %u.\n", (unsigned int)mid)); srv_cancel_sign_response(mid); continue; } fsp = aio_ex->fsp; if (fsp == NULL) { /* file was closed whilst I/O was outstanding. Just * ignore. */ DEBUG( 3,( "process_aio_queue: file closed whilst " "aio outstanding.\n")); srv_cancel_sign_response(mid); continue; } if (!handle_aio_completed(aio_ex, &ret)) { continue; } delete_aio_ex(aio_ex); } outstanding_aio_calls -= signals_received; signals_received = 0; BlockSignals(False, RT_SIGNAL_AIO); return ret; } /**************************************************************************** We're doing write behind and the client closed the file. Wait up to 30 seconds (my arbitrary choice) for the aio to complete. Return 0 if all writes completed, errno to return if not. *****************************************************************************/ #define SMB_TIME_FOR_AIO_COMPLETE_WAIT 29 int wait_for_aio_completion(files_struct *fsp) { struct aio_extra *aio_ex; const SMB_STRUCT_AIOCB **aiocb_list; int aio_completion_count = 0; time_t start_time = time(NULL); int seconds_left; for (seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT; seconds_left >= 0;) { int err = 0; int i; struct timespec ts; aio_completion_count = 0; for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) { if (aio_ex->fsp == fsp) { aio_completion_count++; } } if (!aio_completion_count) { return 0; } DEBUG(3,("wait_for_aio_completion: waiting for %d aio events " "to complete.\n", aio_completion_count )); aiocb_list = SMB_MALLOC_ARRAY(const SMB_STRUCT_AIOCB *, aio_completion_count); if (!aiocb_list) { return ENOMEM; } for( i = 0, aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) { if (aio_ex->fsp == fsp) { aiocb_list[i++] = &aio_ex->acb; } } /* Now wait up to seconds_left for completion. */ ts.tv_sec = seconds_left; ts.tv_nsec = 0; DEBUG(10,("wait_for_aio_completion: %d events, doing a wait " "of %d seconds.\n", aio_completion_count, seconds_left )); err = SMB_VFS_AIO_SUSPEND(fsp, aiocb_list, aio_completion_count, &ts); DEBUG(10,("wait_for_aio_completion: returned err = %d, " "errno = %s\n", err, strerror(errno) )); if (err == -1 && errno == EAGAIN) { DEBUG(0,("wait_for_aio_completion: aio_suspend timed " "out waiting for %d events after a wait of " "%d seconds\n", aio_completion_count, seconds_left)); /* Timeout. */ cancel_aio_by_fsp(fsp); SAFE_FREE(aiocb_list); return EIO; } /* One or more events might have completed - process them if * so. */ for( i = 0; i < aio_completion_count; i++) { uint16 mid = aiocb_list[i]->aio_sigevent.sigev_value.sival_int; aio_ex = find_aio_ex(mid); if (!aio_ex) { DEBUG(0, ("wait_for_aio_completion: mid %u " "doesn't match an aio record\n", (unsigned int)mid )); continue; } if (!handle_aio_completed(aio_ex, &err)) { continue; } delete_aio_ex(aio_ex); } SAFE_FREE(aiocb_list); seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT - (time(NULL) - start_time); } /* We timed out - we don't know why. Return ret if already an error, * else EIO. */ DEBUG(10,("wait_for_aio_completion: aio_suspend timed out waiting " "for %d events\n", aio_completion_count)); return EIO; } /**************************************************************************** Cancel any outstanding aio requests. The client doesn't care about the reply. *****************************************************************************/ void cancel_aio_by_fsp(files_struct *fsp) { struct aio_extra *aio_ex; for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) { if (aio_ex->fsp == fsp) { /* Don't delete the aio_extra record as we may have completed and don't yet know it. Just do the aio_cancel call and return. */ SMB_VFS_AIO_CANCEL(fsp, &aio_ex->acb); aio_ex->fsp = NULL; /* fsp will be closed when we * return. */ } } } #else bool aio_finished(void) { return False; } void initialize_async_io_handler(void) { } int process_aio_queue(void) { return False; } bool schedule_aio_read_and_X(connection_struct *conn, struct smb_request *req, files_struct *fsp, SMB_OFF_T startpos, size_t smb_maxcnt) { return False; } bool schedule_aio_write_and_X(connection_struct *conn, struct smb_request *req, files_struct *fsp, char *data, SMB_OFF_T startpos, size_t numtowrite) { return False; } void cancel_aio_by_fsp(files_struct *fsp) { } int wait_for_aio_completion(files_struct *fsp) { return ENOSYS; } #endif