aio: kill ki_retry
authorKent Overstreet <koverstreet@google.com>
Tue, 7 May 2013 23:19:11 +0000 (16:19 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 8 May 2013 02:46:02 +0000 (19:46 -0700)
Thanks to Zach Brown's work to rip out the retry infrastructure, we don't
need this anymore - ki_retry was only called right after the kiocb was
initialized.

This also refactors and trims some duplicated code, as well as cleaning up
the refcounting/error handling a bit.

[akpm@linux-foundation.org: use fmode_t in aio_run_iocb()]
[akpm@linux-foundation.org: fix file_start_write/file_end_write tests]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/aio.c
include/linux/aio.h

index 5d7dad365f5fe25f0f394132f2481d4c3456e914..c5b1a8c10411ab108960eb74ed20f1b1a4bed601 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -903,30 +903,21 @@ static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
        BUG_ON(ret > 0 && iocb->ki_left == 0);
 }
 
-static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
+typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *,
+                           unsigned long, loff_t);
+
+static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op)
 {
        struct file *file = iocb->ki_filp;
        struct address_space *mapping = file->f_mapping;
        struct inode *inode = mapping->host;
-       ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
-                        unsigned long, loff_t);
        ssize_t ret = 0;
-       unsigned short opcode;
-
-       if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
-               (iocb->ki_opcode == IOCB_CMD_PREAD)) {
-               rw_op = file->f_op->aio_read;
-               opcode = IOCB_CMD_PREADV;
-       } else {
-               rw_op = file->f_op->aio_write;
-               opcode = IOCB_CMD_PWRITEV;
-       }
 
        /* This matches the pread()/pwrite() logic */
        if (iocb->ki_pos < 0)
                return -EINVAL;
 
-       if (opcode == IOCB_CMD_PWRITEV)
+       if (rw == WRITE)
                file_start_write(file);
        do {
                ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
@@ -938,9 +929,9 @@ static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
        /* retry all partial writes.  retry partial reads as long as its a
         * regular file. */
        } while (ret > 0 && iocb->ki_left > 0 &&
-                (opcode == IOCB_CMD_PWRITEV ||
+                (rw == WRITE ||
                  (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
-       if (opcode == IOCB_CMD_PWRITEV)
+       if (rw == WRITE)
                file_end_write(file);
 
        /* This means we must have transferred all that we could */
@@ -950,7 +941,7 @@ static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
 
        /* If we managed to write some out we return that, rather than
         * the eventual error. */
-       if (opcode == IOCB_CMD_PWRITEV
+       if (rw == WRITE
            && ret < 0 && ret != -EIOCBQUEUED
            && iocb->ki_nbytes - iocb->ki_left)
                ret = iocb->ki_nbytes - iocb->ki_left;
@@ -958,73 +949,41 @@ static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
        return ret;
 }
 
-static ssize_t aio_fdsync(struct kiocb *iocb)
-{
-       struct file *file = iocb->ki_filp;
-       ssize_t ret = -EINVAL;
-
-       if (file->f_op->aio_fsync)
-               ret = file->f_op->aio_fsync(iocb, 1);
-       return ret;
-}
-
-static ssize_t aio_fsync(struct kiocb *iocb)
-{
-       struct file *file = iocb->ki_filp;
-       ssize_t ret = -EINVAL;
-
-       if (file->f_op->aio_fsync)
-               ret = file->f_op->aio_fsync(iocb, 0);
-       return ret;
-}
-
-static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat)
+static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat)
 {
        ssize_t ret;
 
+       kiocb->ki_nr_segs = kiocb->ki_nbytes;
+
 #ifdef CONFIG_COMPAT
        if (compat)
-               ret = compat_rw_copy_check_uvector(type,
+               ret = compat_rw_copy_check_uvector(rw,
                                (struct compat_iovec __user *)kiocb->ki_buf,
-                               kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
+                               kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec,
                                &kiocb->ki_iovec);
        else
 #endif
-               ret = rw_copy_check_uvector(type,
+               ret = rw_copy_check_uvector(rw,
                                (struct iovec __user *)kiocb->ki_buf,
-                               kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
+                               kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec,
                                &kiocb->ki_iovec);
        if (ret < 0)
-               goto out;
-
-       ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret);
-       if (ret < 0)
-               goto out;
+               return ret;
 
-       kiocb->ki_nr_segs = kiocb->ki_nbytes;
-       kiocb->ki_cur_seg = 0;
-       /* ki_nbytes/left now reflect bytes instead of segs */
+       /* ki_nbytes now reflect bytes instead of segs */
        kiocb->ki_nbytes = ret;
-       kiocb->ki_left = ret;
-
-       ret = 0;
-out:
-       return ret;
+       return 0;
 }
 
-static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb)
+static ssize_t aio_setup_single_vector(int rw, struct kiocb *kiocb)
 {
-       int bytes;
-
-       bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left);
-       if (bytes < 0)
-               return bytes;
+       if (unlikely(!access_ok(!rw, kiocb->ki_buf, kiocb->ki_nbytes)))
+               return -EFAULT;
 
        kiocb->ki_iovec = &kiocb->ki_inline_vec;
        kiocb->ki_iovec->iov_base = kiocb->ki_buf;
-       kiocb->ki_iovec->iov_len = bytes;
+       kiocb->ki_iovec->iov_len = kiocb->ki_nbytes;
        kiocb->ki_nr_segs = 1;
-       kiocb->ki_cur_seg = 0;
        return 0;
 }
 
@@ -1033,81 +992,82 @@ static ssize_t aio_setup_single_vector(int type, struct file * file, struct kioc
  *     Performs the initial checks and aio retry method
  *     setup for the kiocb at the time of io submission.
  */
-static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat)
+static ssize_t aio_run_iocb(struct kiocb *req, bool compat)
 {
-       struct file *file = kiocb->ki_filp;
-       ssize_t ret = 0;
+       struct file *file = req->ki_filp;
+       ssize_t ret;
+       int rw;
+       fmode_t mode;
+       aio_rw_op *rw_op;
 
-       switch (kiocb->ki_opcode) {
+       switch (req->ki_opcode) {
        case IOCB_CMD_PREAD:
-               ret = -EBADF;
-               if (unlikely(!(file->f_mode & FMODE_READ)))
-                       break;
-               ret = -EFAULT;
-               if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
-                       kiocb->ki_left)))
-                       break;
-               ret = aio_setup_single_vector(READ, file, kiocb);
-               if (ret)
-                       break;
-               ret = -EINVAL;
-               if (file->f_op->aio_read)
-                       kiocb->ki_retry = aio_rw_vect_retry;
-               break;
-       case IOCB_CMD_PWRITE:
-               ret = -EBADF;
-               if (unlikely(!(file->f_mode & FMODE_WRITE)))
-                       break;
-               ret = -EFAULT;
-               if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
-                       kiocb->ki_left)))
-                       break;
-               ret = aio_setup_single_vector(WRITE, file, kiocb);
-               if (ret)
-                       break;
-               ret = -EINVAL;
-               if (file->f_op->aio_write)
-                       kiocb->ki_retry = aio_rw_vect_retry;
-               break;
        case IOCB_CMD_PREADV:
-               ret = -EBADF;
-               if (unlikely(!(file->f_mode & FMODE_READ)))
-                       break;
-               ret = aio_setup_vectored_rw(READ, kiocb, compat);
-               if (ret)
-                       break;
-               ret = -EINVAL;
-               if (file->f_op->aio_read)
-                       kiocb->ki_retry = aio_rw_vect_retry;
-               break;
+               mode    = FMODE_READ;
+               rw      = READ;
+               rw_op   = file->f_op->aio_read;
+               goto rw_common;
+
+       case IOCB_CMD_PWRITE:
        case IOCB_CMD_PWRITEV:
-               ret = -EBADF;
-               if (unlikely(!(file->f_mode & FMODE_WRITE)))
-                       break;
-               ret = aio_setup_vectored_rw(WRITE, kiocb, compat);
+               mode    = FMODE_WRITE;
+               rw      = WRITE;
+               rw_op   = file->f_op->aio_write;
+               goto rw_common;
+rw_common:
+               if (unlikely(!(file->f_mode & mode)))
+                       return -EBADF;
+
+               if (!rw_op)
+                       return -EINVAL;
+
+               ret = (req->ki_opcode == IOCB_CMD_PREADV ||
+                      req->ki_opcode == IOCB_CMD_PWRITEV)
+                       ? aio_setup_vectored_rw(rw, req, compat)
+                       : aio_setup_single_vector(rw, req);
                if (ret)
-                       break;
-               ret = -EINVAL;
-               if (file->f_op->aio_write)
-                       kiocb->ki_retry = aio_rw_vect_retry;
+                       return ret;
+
+               ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes);
+               if (ret < 0)
+                       return ret;
+
+               req->ki_nbytes = ret;
+               req->ki_left = ret;
+
+               ret = aio_rw_vect_retry(req, rw, rw_op);
                break;
+
        case IOCB_CMD_FDSYNC:
-               ret = -EINVAL;
-               if (file->f_op->aio_fsync)
-                       kiocb->ki_retry = aio_fdsync;
+               if (!file->f_op->aio_fsync)
+                       return -EINVAL;
+
+               ret = file->f_op->aio_fsync(req, 1);
                break;
+
        case IOCB_CMD_FSYNC:
-               ret = -EINVAL;
-               if (file->f_op->aio_fsync)
-                       kiocb->ki_retry = aio_fsync;
+               if (!file->f_op->aio_fsync)
+                       return -EINVAL;
+
+               ret = file->f_op->aio_fsync(req, 0);
                break;
+
        default:
                pr_debug("EINVAL: no operation provided\n");
-               ret = -EINVAL;
+               return -EINVAL;
        }
 
-       if (!kiocb->ki_retry)
-               return ret;
+       if (ret != -EIOCBQUEUED) {
+               /*
+                * There's no easy way to restart the syscall since other AIO's
+                * may be already running. Just fail this IO with EINTR.
+                */
+               if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
+                            ret == -ERESTARTNOHAND ||
+                            ret == -ERESTART_RESTARTBLOCK))
+                       ret = -EINTR;
+               aio_complete(req, ret, 0);
+       }
 
        return 0;
 }
@@ -1134,7 +1094,7 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
                return -EINVAL;
        }
 
-       req = aio_get_req(ctx);  /* returns with 2 references to req */
+       req = aio_get_req(ctx);
        if (unlikely(!req))
                return -EAGAIN;
 
@@ -1173,26 +1133,12 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
        req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
        req->ki_opcode = iocb->aio_lio_opcode;
 
-       ret = aio_setup_iocb(req, compat);
+       ret = aio_run_iocb(req, compat);
        if (ret)
                goto out_put_req;
 
-       ret = req->ki_retry(req);
-       if (ret != -EIOCBQUEUED) {
-               /*
-                * There's no easy way to restart the syscall since other AIO's
-                * may be already running. Just fail this IO with EINTR.
-                */
-               if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
-                            ret == -ERESTARTNOHAND ||
-                            ret == -ERESTART_RESTARTBLOCK))
-                       ret = -EINTR;
-               aio_complete(req, ret, 0);
-       }
-
        aio_put_req(req);       /* drop extra ref to req */
        return 0;
-
 out_put_req:
        atomic_dec(&ctx->reqs_active);
        aio_put_req(req);       /* drop extra ref to req */
index 7308836dd0452efe1768c3752336f6ade04d358f..1bdf965339f9bef4222a5bc739efc0b23807e35e 100644 (file)
@@ -29,38 +29,12 @@ struct kiocb;
 
 typedef int (kiocb_cancel_fn)(struct kiocb *, struct io_event *);
 
-/* is there a better place to document function pointer methods? */
-/**
- * ki_retry    -       iocb forward progress callback
- * @kiocb:     The kiocb struct to advance by performing an operation.
- *
- * This callback is called when the AIO core wants a given AIO operation
- * to make forward progress.  The kiocb argument describes the operation
- * that is to be performed.  As the operation proceeds, perhaps partially,
- * ki_retry is expected to update the kiocb with progress made.  Typically
- * ki_retry is set in the AIO core and it itself calls file_operations
- * helpers.
- *
- * ki_retry's return value determines when the AIO operation is completed
- * and an event is generated in the AIO event ring.  Except the special
- * return values described below, the value that is returned from ki_retry
- * is transferred directly into the completion ring as the operation's
- * resulting status.  Once this has happened ki_retry *MUST NOT* reference
- * the kiocb pointer again.
- *
- * If ki_retry returns -EIOCBQUEUED it has made a promise that aio_complete()
- * will be called on the kiocb pointer in the future.  The AIO core will
- * not ask the method again -- ki_retry must ensure forward progress.
- * aio_complete() must be called once and only once in the future, multiple
- * calls may result in undefined behaviour.
- */
 struct kiocb {
        atomic_t                ki_users;
 
        struct file             *ki_filp;
        struct kioctx           *ki_ctx;        /* NULL for sync ops */
        kiocb_cancel_fn         *ki_cancel;
-       ssize_t                 (*ki_retry)(struct kiocb *);
        void                    (*ki_dtor)(struct kiocb *);
 
        union {