Re: [PATCH RFC v2] Integration coroutines into fuse export

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 10390 bytes --]

On Sun, Mar 16, 2025 at 01:30:20AM +0800, saz97 wrote:
> Signed-off-by: Changzhi Xie <sa.z@qq.com>
> 
> This commit refactors the FUSE export to process read and write operations
> using coroutines, improving concurrency and avoiding blocking the main loop.
> 
> The main changes include:
> 1.  Introduce FuseIORequest structure to encapsulate I/O parameters and state
> 2.  Move read/write processing into coroutine fuse_read_coroutine and fuse_write_coroutine
> 3.  Use blk_co_pread/pwrite for async block layer access
> ---
>  block/export/fuse.c | 189 +++++++++++++++++++++++++++++++-------------
>  1 file changed, 132 insertions(+), 57 deletions(-)
> 
> diff --git a/block/export/fuse.c b/block/export/fuse.c
> index 465cc9891d..3314f64706 100644
> --- a/block/export/fuse.c
> +++ b/block/export/fuse.c
> @@ -39,6 +39,7 @@
>  
>  #ifdef __linux__
>  #include <linux/fs.h>
> +#include <linux/fuse.h>
>  #endif
>  
>  /* Prevent overly long bounce buffer allocations */
> @@ -49,7 +50,6 @@ typedef struct FuseExport {
>      BlockExport common;
>  
>      struct fuse_session *fuse_session;
> -    struct fuse_buf fuse_buf;
>      unsigned int in_flight; /* atomic */
>      bool mounted, fd_handler_set_up;
>  
> @@ -64,6 +64,14 @@ typedef struct FuseExport {
>      gid_t st_gid;
>  } FuseExport;
>  
> +typedef struct FuseIORequest {
> +    fuse_req_t req;
> +    size_t size;
> +    off_t offset;
> +    FuseExport *exp;
> +    char *write_buf;
> +} FuseIORequest;
> +
>  static GHashTable *exports;
>  static const struct fuse_lowlevel_ops fuse_ops;
>  
> @@ -288,6 +296,7 @@ fail:
>  static void read_from_fuse_export(void *opaque)
>  {
>      FuseExport *exp = opaque;
> +    struct fuse_buf buf = {};
>      int ret;
>  
>      blk_exp_ref(&exp->common);
> @@ -295,20 +304,30 @@ static void read_from_fuse_export(void *opaque)
>      qatomic_inc(&exp->in_flight);
>  
>      do {
> -        ret = fuse_session_receive_buf(exp->fuse_session, &exp->fuse_buf);
> +        ret = fuse_session_receive_buf(exp->fuse_session, &buf);
>      } while (ret == -EINTR);
>      if (ret < 0) {
>          goto out;
>      }
>  
> -    fuse_session_process_buf(exp->fuse_session, &exp->fuse_buf);
> +    fuse_session_process_buf(exp->fuse_session, &buf);
>  
>  out:
> +    struct fuse_in_header *in = (struct fuse_in_header *)buf.mem;
> +
> +    if (in->opcode == FUSE_WRITE || in->opcode == FUSE_READ) {
> +        g_free(buf.mem);
> +        return;

Returning here is not safe because &buf was passed to
fuse_session_process_buf() and is located on read_from_fuse_export()'s
stack. The coroutine must not access buf after this function returns.

I suggest moving most of this function into a coroutine so that struct
fuse_buf can be on the coroutine's stack. That way it outlives
fuse_session_process_buf().

Doing this also avoids duplicating the code below and eliminates the
need for FuseIORequest.

> +    }
> +
>      if (qatomic_fetch_dec(&exp->in_flight) == 1) {
>          aio_wait_kick(); /* wake AIO_WAIT_WHILE() */
>      }
>  
>      blk_exp_unref(&exp->common);
> +
> +    g_free(buf.mem);

Please make this free(buf.mem) since libfuse uses malloc(3).

> +
>  }
>  
>  static void fuse_export_shutdown(BlockExport *blk_exp)
> @@ -347,7 +366,6 @@ static void fuse_export_delete(BlockExport *blk_exp)
>          fuse_session_destroy(exp->fuse_session);
>      }
>  
> -    free(exp->fuse_buf.mem);
>      g_free(exp->mountpoint);
>  }
>  
> @@ -570,102 +588,159 @@ static void fuse_open(fuse_req_t req, fuse_ino_t inode,
>      fuse_reply_open(req, fi);
>  }
>  
> -/**
> - * Handle client reads from the exported image.
> - */
> -static void fuse_read(fuse_req_t req, fuse_ino_t inode,
> -                      size_t size, off_t offset, struct fuse_file_info *fi)
> +static void coroutine_fn fuse_read_coroutine(void *opaque)
>  {
> -    FuseExport *exp = fuse_req_userdata(req);
> +    FuseIORequest *io_req = opaque;
> +    FuseExport *exp = io_req->exp;
>      int64_t length;
> -    void *buf;
> +    void *buffer;
>      int ret;
>  
> -    /* Limited by max_read, should not happen */
> -    if (size > FUSE_MAX_BOUNCE_BYTES) {
> -        fuse_reply_err(req, EINVAL);
> -        return;
> +    if (io_req->size > FUSE_MAX_BOUNCE_BYTES) {
> +        fuse_reply_err(io_req->req, EINVAL);
> +        goto cleanup;
>      }
>  
> -    /**
> -     * Clients will expect short reads at EOF, so we have to limit
> -     * offset+size to the image length.
> -     */
>      length = blk_getlength(exp->common.blk);
>      if (length < 0) {
> -        fuse_reply_err(req, -length);
> -        return;
> +        fuse_reply_err(io_req->req, -length);
> +        goto cleanup;
>      }
>  
> -    if (offset + size > length) {
> -        size = length - offset;
> +    if (io_req->offset + io_req->size > length) {
> +        io_req->size = length - io_req->offset;
>      }
>  
> -    buf = qemu_try_blockalign(blk_bs(exp->common.blk), size);
> -    if (!buf) {
> -        fuse_reply_err(req, ENOMEM);
> -        return;
> +    if (io_req->size == 0) {
> +        fuse_reply_buf(io_req->req, NULL, 0);
> +        goto cleanup;
>      }
>  
> -    ret = blk_pread(exp->common.blk, offset, size, buf, 0);
> +    buffer = qemu_try_blockalign(blk_bs(exp->common.blk), io_req->size);
> +    if (!buffer) {
> +        fuse_reply_err(io_req->req, ENOMEM);
> +        goto cleanup;
> +    }
> +
> +    ret = blk_co_pread(exp->common.blk, io_req->offset,
> +                       io_req->size, buffer, 0);
>      if (ret >= 0) {
> -        fuse_reply_buf(req, buf, size);
> +        fuse_reply_buf(io_req->req, buffer, io_req->size);
>      } else {
> -        fuse_reply_err(req, -ret);
> +        fuse_reply_err(io_req->req, -ret);
> +    }
> +
> +    qemu_vfree(buffer);
> +
> +cleanup:
> +    if (qatomic_fetch_dec(&exp->in_flight) == 1) {
> +        aio_wait_kick(); /* wake AIO_WAIT_WHILE() */
>      }
>  
> -    qemu_vfree(buf);
> +    blk_exp_unref(&exp->common);
> +
> +    g_free(io_req);
>  }
>  
> -/**
> - * Handle client writes to the exported image.
> - */
> -static void fuse_write(fuse_req_t req, fuse_ino_t inode, const char *buf,
> -                       size_t size, off_t offset, struct fuse_file_info *fi)
> +static void coroutine_fn fuse_write_coroutine(void *opaque)
>  {
> -    FuseExport *exp = fuse_req_userdata(req);
> +    FuseIORequest *io_req = opaque;
> +    FuseExport *exp = io_req->exp;
>      int64_t length;
>      int ret;
>  
> -    /* Limited by max_write, should not happen */
> -    if (size > BDRV_REQUEST_MAX_BYTES) {
> -        fuse_reply_err(req, EINVAL);
> -        return;
> +    if (io_req->size > BDRV_REQUEST_MAX_BYTES) {
> +        fuse_reply_err(io_req->req, EINVAL);
> +        goto cleanup;
>      }
>  
>      if (!exp->writable) {
> -        fuse_reply_err(req, EACCES);
> -        return;
> +        fuse_reply_err(io_req->req, EACCES);
> +        goto cleanup;
>      }
>  
> -    /**
> -     * Clients will expect short writes at EOF, so we have to limit
> -     * offset+size to the image length.
> -     */
>      length = blk_getlength(exp->common.blk);
>      if (length < 0) {
> -        fuse_reply_err(req, -length);
> -        return;
> +        fuse_reply_err(io_req->req, -length);
> +        goto cleanup;
>      }
>  
> -    if (offset + size > length) {
> +    if (io_req->offset + io_req->size > length) {
>          if (exp->growable) {
> -            ret = fuse_do_truncate(exp, offset + size, true, PREALLOC_MODE_OFF);
> +            ret = fuse_do_truncate(exp, io_req->offset + io_req->size,
> +                                   true, PREALLOC_MODE_OFF);
>              if (ret < 0) {
> -                fuse_reply_err(req, -ret);
> -                return;
> +                fuse_reply_err(io_req->req, -ret);
> +                goto cleanup;
>              }
>          } else {
> -            size = length - offset;
> +            io_req->size = MAX(0, length - io_req->offset);
> +            if (io_req->size == 0) {
> +                fuse_reply_write(io_req->req, 0);
> +                goto cleanup;
> +            }
>          }
>      }
>  
> -    ret = blk_pwrite(exp->common.blk, offset, size, buf, 0);
> +    ret = blk_co_pwrite(exp->common.blk, io_req->offset, io_req->size,
> +                        io_req->write_buf, 0);
>      if (ret >= 0) {
> -        fuse_reply_write(req, size);
> +        fuse_reply_write(io_req->req, io_req->size);
>      } else {
> -        fuse_reply_err(req, -ret);
> +        fuse_reply_err(io_req->req, -ret);
> +    }
> +
> +cleanup:
> +    if (qatomic_fetch_dec(&exp->in_flight) == 1) {
> +        aio_wait_kick(); /* wake AIO_WAIT_WHILE() */
>      }
> +
> +    blk_exp_unref(&exp->common);
> +
> +    g_free(io_req->write_buf);
> +    g_free(io_req);
> +}
> +
> +/**
> + * Handle client reads from the exported image.
> + */
> +static void fuse_read(fuse_req_t req, fuse_ino_t inode,
> +                      size_t size, off_t offset, struct fuse_file_info *fi)
> +{
> +    FuseExport *exp = fuse_req_userdata(req);
> +    FuseIORequest *io_req = g_new(FuseIORequest, 1);
> +
> +    *io_req = (FuseIORequest) {
> +        .req = req,
> +        .size = size,
> +        .offset = offset,
> +        .exp = exp,
> +    };
> +
> +    Coroutine *co = qemu_coroutine_create(fuse_read_coroutine, io_req);
> +    qemu_coroutine_enter(co);
> +}
> +
> +
> +/**
> + * Handle client writes to the exported image.
> + */
> +static void fuse_write(fuse_req_t req, fuse_ino_t inode, const char *buf,
> +                       size_t size, off_t offset, struct fuse_file_info *fi)
> +{
> +    FuseExport *exp = fuse_req_userdata(req);
> +    FuseIORequest *io_req = g_new(FuseIORequest, 1);
> +
> +    *io_req = (FuseIORequest) {
> +        .req = req,
> +        .size = size,
> +        .offset = offset,
> +        .exp = exp,
> +        .write_buf = g_memdup2_qemu(buf, size),
> +    };
> +
> +    Coroutine *co = qemu_coroutine_create(fuse_write_coroutine, io_req);
> +    qemu_coroutine_enter(co);
>  }
>  
>  /**
> -- 
> 2.34.1
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]