public inbox for devel@edk2.groups.io
 help / color / mirror / Atom feed
From: "Li, Songpeng" <songpeng.li@intel.com>
To: Laszlo Ersek <lersek@redhat.com>,
	"edk2-devel@lists.01.org" <edk2-devel@lists.01.org>
Cc: "Carsey, Jaben" <jaben.carsey@intel.com>,
	"Ni, Ray" <ray.ni@intel.com>, "Wu, Jiaxin" <jiaxin.wu@intel.com>
Subject: Re: [PATCH v2] ShellPkg/TftpDynamicCommand: Change file writing method in tftp
Date: Mon, 14 Jan 2019 08:07:14 +0000	[thread overview]
Message-ID: <F4A2861CFE7EF04096224C048078577D0517BF99@shsmsx102.ccr.corp.intel.com> (raw)
In-Reply-To: <168d2e88-e1e4-730d-ac39-474670c10bc6@redhat.com>

Hi Laszlo,

Thank you for the comments.

For your questions below:

> Before the patch, the shell would write the entire file in "one go".
> 
> In this particular case, I'm not interested in the memory allocation
> size, but the transfer block size that the shell would ultimately use
> with EFI_FILE_WRITE. I find it interesting because some lower-level
> storage drivers may have limitations themselves (for example, a single
> SCSI transfer might not be larger than 1GB). What layer would handle
> (split) a downloaded file larger than that into smaller chunks? Maybe
> the FS (FAT) driver?

Although underlying driver can handle split writing,  we still need split writing in application layer because:
1. To improve user experiences -- if write the entire file in "one go" way, the shell looks like hanged after download finished. 
     After this patch, shell prompt will show up immediately after the downloading progress bar finished.
2. Before this patch, if the file size is larger than physical memory size, we cannot download the file. 

> And, after the patch, we're now writing TFTP packet-sized chunks to the
> file. Is this not a performance problem for traditional spindle disks
> ("spinning rust")? If neither the rest of the driver stack, nor the disk
> controller itself, support some form of write caching, then writing out
> every TFTP packet in isolation (especially for a multi-GB file) could be
> detrimental to performance (and maybe to the health of the spindle disk).
> 
> Currently I'm not affected by this use case either way, so feel free to
> ignore, but it might make sense to implement some larger chunking here,
> like 64KB or 128KB -- later, as a separate Feature BZ.

We agree with this comment. It is a performance problem and will be fixed. 

Best,
Songpeng

> -----Original Message-----
> From: Laszlo Ersek [mailto:lersek@redhat.com]
> Sent: Friday, January 11, 2019 6:25 PM
> To: Li, Songpeng <songpeng.li@intel.com>; edk2-devel@lists.01.org
> Cc: Carsey, Jaben <jaben.carsey@intel.com>; Ni, Ray <ray.ni@intel.com>;
> Wu, Jiaxin <jiaxin.wu@intel.com>
> Subject: Re: [edk2] [PATCH v2] ShellPkg/TftpDynamicCommand: Change file
> writing method in tftp
> 
> On 01/10/19 03:54, Songpeng Li wrote:
> > REF: https://bugzilla.tianocore.org/show_bug.cgi?id=1433
> >
> > v2: Remove an unused variable.
> >
> > Current logic of shell tftp download was writing file after tftp
> > download finished, when the file is large, it looks like the shell
> > tftp command hanged after download was finished. To improve
> > end-user experience, the solution is using split file writing
> > instead.
> >
> > This patch update the code to open and close file inside
> > DownloadFile(), and save each packet to file within callback
> > function CheckPacket().
> 
> I have some questions, just out of curiosity:
> 
> >
> > Since AllocatePage() is no-longer needed, This patch can also
> > remove the memory limitation. The download file can be larger
> > than system free memory now.
> >
> > Cc: Jaben Carsey <jaben.carsey@intel.com>
> > Cc: Ruiyu Ni <ruiyu.ni@intel.com>
> > Cc: Wu Jiaxin <jiaxin.wu@intel.com>
> > Contributed-under: TianoCore Contribution Agreement 1.1
> > Signed-off-by: Songpeng Li <songpeng.li@intel.com>
> > ---
> >  ShellPkg/DynamicCommand/TftpDynamicCommand/Tftp.c | 154
> ++++++++------------
> >  1 file changed, 64 insertions(+), 90 deletions(-)
> >
> > diff --git a/ShellPkg/DynamicCommand/TftpDynamicCommand/Tftp.c
> b/ShellPkg/DynamicCommand/TftpDynamicCommand/Tftp.c
> > index ed081b5bad7c..ba753a279b00 100644
> > --- a/ShellPkg/DynamicCommand/TftpDynamicCommand/Tftp.c
> > +++ b/ShellPkg/DynamicCommand/TftpDynamicCommand/Tftp.c
> > @@ -41,6 +41,12 @@ STATIC CONST CHAR16 mTftpProgressFrame[] = L"[
> >  // (TFTP_PROGRESS_MESSAGE_SIZE-1) '\b'
> >  STATIC CONST CHAR16 mTftpProgressDelete[] =
> L"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\
> b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b";
> >
> > +// Local File Handle
> > +SHELL_FILE_HANDLE     mFileHandle;
> > +
> > +// Path of the local file, Unicode encoded
> > +CONST CHAR16         *mLocalFilePath;
> > +
> >  /**
> >    Check and convert the UINT16 option values of the 'tftp' command
> >
> > @@ -166,9 +172,6 @@ GetFileSize (
> >    @param[in]   FileSize       Size of the file in number of bytes
> >    @param[in]   BlockSize      Value of the TFTP blksize option
> >    @param[in]   WindowSize     Value of the TFTP window size option
> > -  @param[out]  Data           Address where to store the address of the
> buffer
> > -                              where the data of the file were downloaded in
> > -                              case of success.
> >
> >    @retval  EFI_SUCCESS           The file was downloaded.
> >    @retval  EFI_OUT_OF_RESOURCES  A memory allocation failed.
> > @@ -184,8 +187,7 @@ DownloadFile (
> >    IN   CONST CHAR8          *AsciiFilePath,
> >    IN   UINTN                FileSize,
> >    IN   UINT16               BlockSize,
> > -  IN   UINT16               WindowSize,
> > -  OUT  VOID                 **Data
> > +  IN   UINT16               WindowSize
> >    );
> >
> >  /**
> > @@ -287,7 +289,6 @@ RunTftp (
> >    CHAR8                   *AsciiRemoteFilePath;
> >    UINTN                   FilePathSize;
> >    CONST CHAR16            *Walker;
> > -  CONST CHAR16            *LocalFilePath;
> >    EFI_MTFTP4_CONFIG_DATA  Mtftp4ConfigData;
> >    EFI_HANDLE              *Handles;
> >    UINTN                   HandleCount;
> > @@ -297,9 +298,6 @@ RunTftp (
> >    EFI_HANDLE              Mtftp4ChildHandle;
> >    EFI_MTFTP4_PROTOCOL     *Mtftp4;
> >    UINTN                   FileSize;
> > -  UINTN                   DataSize;
> > -  VOID                    *Data;
> > -  SHELL_FILE_HANDLE       FileHandle;
> >    UINT16                  BlockSize;
> >    UINT16                  WindowSize;
> >
> > @@ -309,7 +307,6 @@ RunTftp (
> >    AsciiRemoteFilePath = NULL;
> >    Handles             = NULL;
> >    FileSize            = 0;
> > -  DataSize            = 0;
> >    BlockSize           = MTFTP_DEFAULT_BLKSIZE;
> >    WindowSize          = MTFTP_DEFAULT_WINDOWSIZE;
> >
> > @@ -385,7 +382,7 @@ RunTftp (
> >    UnicodeStrToAsciiStrS (RemoteFilePath, AsciiRemoteFilePath,
> FilePathSize);
> >
> >    if (ParamCount == 4) {
> > -    LocalFilePath = ShellCommandLineGetRawValue (CheckPackage, 3);
> > +    mLocalFilePath = ShellCommandLineGetRawValue (CheckPackage, 3);
> >    } else {
> >      Walker = RemoteFilePath + StrLen (RemoteFilePath);
> >      while ((--Walker) >= RemoteFilePath) {
> > @@ -394,7 +391,7 @@ RunTftp (
> >          break;
> >        }
> >      }
> > -    LocalFilePath = Walker + 1;
> > +    mLocalFilePath = Walker + 1;
> >    }
> >
> >    //
> > @@ -492,7 +489,6 @@ RunTftp (
> >         (NicNumber < HandleCount) && (ShellStatus != SHELL_SUCCESS);
> >         NicNumber++) {
> >      ControllerHandle = Handles[NicNumber];
> > -    Data = NULL;
> >
> >      Status = GetNicName (ControllerHandle, NicNumber, NicName);
> >      if (EFI_ERROR (Status)) {
> > @@ -543,7 +539,7 @@ RunTftp (
> >        goto NextHandle;
> >      }
> >
> > -    Status = DownloadFile (Mtftp4, RemoteFilePath, AsciiRemoteFilePath,
> FileSize, BlockSize, WindowSize, &Data);
> > +    Status = DownloadFile (Mtftp4, RemoteFilePath, AsciiRemoteFilePath,
> FileSize, BlockSize, WindowSize);
> >      if (EFI_ERROR (Status)) {
> >        ShellPrintHiiEx (
> >          -1, -1, NULL, STRING_TOKEN (STR_TFTP_ERR_DOWNLOAD),
> > @@ -552,45 +548,8 @@ RunTftp (
> >        goto NextHandle;
> >      }
> >
> > -    DataSize = FileSize;
> > -
> > -    if (!EFI_ERROR (ShellFileExists (LocalFilePath))) {
> > -      ShellDeleteFileByName (LocalFilePath);
> > -    }
> > -
> > -    Status = ShellOpenFileByName (
> > -               LocalFilePath,
> > -               &FileHandle,
> > -               EFI_FILE_MODE_CREATE |
> > -               EFI_FILE_MODE_WRITE  |
> > -               EFI_FILE_MODE_READ,
> > -               0
> > -               );
> > -    if (EFI_ERROR (Status)) {
> > -      ShellPrintHiiEx (
> > -        -1, -1, NULL, STRING_TOKEN (STR_GEN_FILE_OPEN_FAIL),
> > -        mTftpHiiHandle, L"tftp", LocalFilePath
> > -      );
> > -      goto NextHandle;
> > -    }
> > -
> > -    Status = ShellWriteFile (FileHandle, &FileSize, Data);
> 
> Before the patch, the shell would write the entire file in "one go".
> 
> In this particular case, I'm not interested in the memory allocation
> size, but the transfer block size that the shell would ultimately use
> with EFI_FILE_WRITE. I find it interesting because some lower-level
> storage drivers may have limitations themselves (for example, a single
> SCSI transfer might not be larger than 1GB). What layer would handle
> (split) a downloaded file larger than that into smaller chunks? Maybe
> the FS (FAT) driver?
> 
> 
> > -    if (!EFI_ERROR (Status)) {
> > -      ShellStatus = SHELL_SUCCESS;
> > -    } else {
> > -      ShellPrintHiiEx (
> > -        -1, -1, NULL, STRING_TOKEN (STR_TFTP_ERR_WRITE),
> > -        mTftpHiiHandle, LocalFilePath, Status
> > -      );
> > -    }
> > -    ShellCloseFile (&FileHandle);
> > -
> >      NextHandle:
> >
> > -    if (Data != NULL) {
> > -      gBS->FreePages ((EFI_PHYSICAL_ADDRESS)(UINTN)Data,
> EFI_SIZE_TO_PAGES (DataSize));
> > -    }
> > -
> >      CloseProtocolAndDestroyServiceChild (
> >        ControllerHandle,
> >        &gEfiMtftp4ServiceBindingProtocolGuid,
> > @@ -912,9 +871,6 @@ Error :
> >    @param[in]   FileSize       Size of the file in number of bytes
> >    @param[in]   BlockSize      Value of the TFTP blksize option
> >    @param[in]   WindowSize     Value of the TFTP window size option
> > -  @param[out]  Data           Address where to store the address of the
> buffer
> > -                              where the data of the file were downloaded in
> > -                              case of success.
> >
> >    @retval  EFI_SUCCESS           The file was downloaded.
> >    @retval  EFI_OUT_OF_RESOURCES  A memory allocation failed.
> > @@ -930,13 +886,10 @@ DownloadFile (
> >    IN   CONST CHAR8          *AsciiFilePath,
> >    IN   UINTN                FileSize,
> >    IN   UINT16               BlockSize,
> > -  IN   UINT16               WindowSize,
> > -  OUT  VOID                 **Data
> > +  IN   UINT16               WindowSize
> >    )
> >  {
> >    EFI_STATUS            Status;
> > -  EFI_PHYSICAL_ADDRESS  PagesAddress;
> > -  VOID                  *Buffer;
> >    DOWNLOAD_CONTEXT      *TftpContext;
> >    EFI_MTFTP4_TOKEN      Mtftp4Token;
> >    UINT8                 BlksizeBuf[10];
> > @@ -944,22 +897,6 @@ DownloadFile (
> >
> >    ZeroMem (&Mtftp4Token, sizeof (EFI_MTFTP4_TOKEN));
> >
> > -  // Downloaded file can be large. BS.AllocatePages() is more faster
> > -  // than AllocatePool() and avoid fragmentation.
> > -  // The downloaded file could be an EFI application. Marking the
> > -  // allocated page as EfiBootServicesCode would allow to execute a
> > -  // potential downloaded EFI application.
> > -  Status = gBS->AllocatePages (
> > -                   AllocateAnyPages,
> > -                   EfiBootServicesCode,
> > -                   EFI_SIZE_TO_PAGES (FileSize),
> > -                   &PagesAddress
> > -                   );
> > -  if (EFI_ERROR (Status)) {
> > -    return Status;
> > -  }
> > -
> > -  Buffer = (VOID*)(UINTN)PagesAddress;
> >    TftpContext = AllocatePool (sizeof (DOWNLOAD_CONTEXT));
> >    if (TftpContext == NULL) {
> >      Status = EFI_OUT_OF_RESOURCES;
> > @@ -970,8 +907,6 @@ DownloadFile (
> >    TftpContext->LastReportedNbOfBytes = 0;
> >
> >    Mtftp4Token.Filename    = (UINT8*)AsciiFilePath;
> > -  Mtftp4Token.BufferSize  = FileSize;
> > -  Mtftp4Token.Buffer      = Buffer;
> >    Mtftp4Token.CheckPacket = CheckPacket;
> >    Mtftp4Token.Context     = (VOID*)TftpContext;
> >    Mtftp4Token.OptionCount = 0;
> > @@ -1000,14 +935,41 @@ DownloadFile (
> >      mTftpHiiHandle, FilePath
> >      );
> >
> > +  //
> > +  // OPEN FILE
> > +  //
> > +  if (!EFI_ERROR (ShellFileExists (mLocalFilePath))) {
> > +    ShellDeleteFileByName (mLocalFilePath);
> > +  }
> > +
> > +  Status = ShellOpenFileByName (
> > +              mLocalFilePath,
> > +              &mFileHandle,
> > +              EFI_FILE_MODE_CREATE |
> > +              EFI_FILE_MODE_WRITE  |
> > +              EFI_FILE_MODE_READ,
> > +              0
> > +              );
> > +  if (EFI_ERROR (Status)) {
> > +    ShellPrintHiiEx (
> > +      -1, -1, NULL, STRING_TOKEN (STR_GEN_FILE_OPEN_FAIL),
> > +      mTftpHiiHandle, L"tftp", mLocalFilePath
> > +    );
> > +    goto Error;
> > +  }
> > +
> >    Status = Mtftp4->ReadFile (Mtftp4, &Mtftp4Token);
> >    ShellPrintHiiEx (
> >      -1, -1, NULL, STRING_TOKEN (STR_GEN_CRLF),
> >      mTftpHiiHandle
> >      );
> >
> > -Error :
> > +  //
> > +  // CLOSE FILE
> > +  //
> > +  ShellCloseFile (&mFileHandle);
> >
> > +Error :
> >    if (TftpContext != NULL) {
> >      FreePool (TftpContext);
> >    }
> > @@ -1016,14 +978,7 @@ Error :
> >      FreePool (Mtftp4Token.OptionList);
> >    }
> >
> > -  if (EFI_ERROR (Status)) {
> > -    gBS->FreePages (PagesAddress, EFI_SIZE_TO_PAGES (FileSize));
> > -    return Status;
> > -  }
> > -
> > -  *Data = Buffer;
> > -
> > -  return EFI_SUCCESS;
> > +  return Status;
> >  }
> >
> >  /**
> > @@ -1054,6 +1009,7 @@ CheckPacket (
> >    UINTN             Index;
> >    UINTN             LastStep;
> >    UINTN             Step;
> > +  UINTN             DownloadLen;
> >    EFI_STATUS        Status;
> >
> >    if ((NTOHS (Packet->OpCode)) != EFI_MTFTP4_OPCODE_DATA) {
> > @@ -1061,17 +1017,35 @@ CheckPacket (
> >    }
> >
> >    Context = (DOWNLOAD_CONTEXT*)Token->Context;
> > -  if (Context->DownloadedNbOfBytes == 0) {
> > -    ShellPrintEx (-1, -1, L"%s       0 Kb", mTftpProgressFrame);
> > -  }
> >
> >    //
> >    // The data in the packet are prepended with two UINT16 :
> >    // . OpCode = EFI_MTFTP4_OPCODE_DATA
> >    // . Block  = the number of this block of data
> >    //
> > -  Context->DownloadedNbOfBytes += PacketLen - sizeof (Packet-
> >OpCode)
> > -                                            - sizeof (Packet->Data.Block);
> > +  DownloadLen = (UINTN)PacketLen - sizeof (Packet->OpCode) - sizeof
> (Packet->Data.Block);
> > +
> > +  ShellSetFilePosition(mFileHandle, Context->DownloadedNbOfBytes);
> > +  Status = ShellWriteFile (mFileHandle, &DownloadLen, Packet-
> >Data.Data);
> 
> And, after the patch, we're now writing TFTP packet-sized chunks to the
> file. Is this not a performance problem for traditional spindle disks
> ("spinning rust")? If neither the rest of the driver stack, nor the disk
> controller itself, support some form of write caching, then writing out
> every TFTP packet in isolation (especially for a multi-GB file) could be
> detrimental to performance (and maybe to the health of the spindle disk).
> 
> Currently I'm not affected by this use case either way, so feel free to
> ignore, but it might make sense to implement some larger chunking here,
> like 64KB or 128KB -- later, as a separate Feature BZ.
> 
> Thanks,
> Laszlo
> 
> 
> > +  if (EFI_ERROR (Status)) {
> > +    if (Context->DownloadedNbOfBytes > 0) {
> > +      ShellPrintHiiEx (
> > +        -1, -1, NULL, STRING_TOKEN (STR_GEN_CRLF),
> > +        mTftpHiiHandle
> > +      );
> > +    }
> > +    ShellPrintHiiEx (
> > +      -1, -1, NULL, STRING_TOKEN (STR_TFTP_ERR_WRITE),
> > +      mTftpHiiHandle, mLocalFilePath, Status
> > +    );
> > +    return Status;
> > +  }
> > +
> > +  if (Context->DownloadedNbOfBytes == 0) {
> > +    ShellPrintEx (-1, -1, L"%s       0 Kb", mTftpProgressFrame);
> > +  }
> > +
> > +  Context->DownloadedNbOfBytes += DownloadLen;
> >    NbOfKb = Context->DownloadedNbOfBytes / 1024;
> >
> >    Progress[0] = L'\0';
> >


  reply	other threads:[~2019-01-14  8:07 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-01-10  2:54 [PATCH v2] ShellPkg/TftpDynamicCommand: Change file writing method in tftp Songpeng Li
2019-01-10 15:19 ` Carsey, Jaben
2019-01-11  0:44 ` Wu, Jiaxin
2019-01-11 10:24 ` Laszlo Ersek
2019-01-14  8:07   ` Li, Songpeng [this message]
  -- strict thread matches above, loose matches on Subject: below --
2019-01-10  2:00 Songpeng Li

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-list from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=F4A2861CFE7EF04096224C048078577D0517BF99@shsmsx102.ccr.corp.intel.com \
    --to=devel@edk2.groups.io \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox