sg: convert the indirect IO path to use the block layer
authorFUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Fri, 29 Aug 2008 10:32:18 +0000 (12:32 +0200)
committerJens Axboe <jens.axboe@oracle.com>
Thu, 9 Oct 2008 06:56:10 +0000 (08:56 +0200)
This patch converts the indirect IO path (including mmap IO and old
struct sg_header) to use the block layer functions (blk_get_request,
blk_execute_rq_nowait, blk_rq_map_user, etc) instead of
scsi_execute_async().

[Jens: fixed compile error with SCSI logging enabled]

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Douglas Gilbert <dougg@torque.net>
Cc: Mike Christie <michaelc@cs.wisc.edu>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
drivers/scsi/sg.c

index cb6de0752ee1318ba9e4b708f2e39256aa9e1008..d6391666502c7cb0dc5d62c9cf045cbb9197bd33 100644 (file)
@@ -47,7 +47,6 @@ static int sg_version_num = 30534;    /* 2 digits for each component */
 #include <linux/seq_file.h>
 #include <linux/blkdev.h>
 #include <linux/delay.h>
-#include <linux/scatterlist.h>
 #include <linux/blktrace_api.h>
 #include <linux/smp_lock.h>
 
@@ -119,7 +118,8 @@ typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
        unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
        unsigned bufflen;       /* Size of (aggregate) data buffer */
        unsigned b_malloc_len;  /* actual len malloc'ed in buffer */
-       struct scatterlist *buffer;/* scatter list */
+       struct page **pages;
+       int page_order;
        char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
        unsigned char cmd_opcode; /* first byte of command */
 } Sg_scatter_hold;
@@ -190,8 +190,6 @@ static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
                        int read_only, Sg_request **o_srp);
 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
                           unsigned char *cmnd, int timeout, int blocking);
-static int sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
-                     int wr_xf, int *countp, unsigned char __user **up);
 static int sg_write_xfer(Sg_request * srp);
 static int sg_read_xfer(Sg_request * srp);
 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
@@ -199,8 +197,6 @@ static void sg_remove_scat(Sg_scatter_hold * schp);
 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
-static struct page *sg_page_malloc(int rqSz, int lowDma, int *retSzp);
-static void sg_page_free(struct page *page, int size);
 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
 static int sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
 static void __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
@@ -771,26 +767,11 @@ sg_common_write(Sg_fd * sfp, Sg_request * srp,
                break;
        }
        hp->duration = jiffies_to_msecs(jiffies);
-/* Now send everything of to mid-level. The next time we hear about this
-   packet is when sg_cmd_done() is called (i.e. a callback). */
-       if (srp->rq) {
-               srp->rq->timeout = timeout;
-               blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
-                                     srp->rq, 1, sg_rq_end_io);
-               return 0;
-       }
-       if (scsi_execute_async(sdp->device, cmnd, hp->cmd_len, data_dir, srp->data.buffer,
-                               hp->dxfer_len, srp->data.k_use_sg, timeout,
-                               SG_DEFAULT_RETRIES, srp, sg_cmd_done,
-                               GFP_ATOMIC)) {
-               SCSI_LOG_TIMEOUT(1, printk("sg_common_write: scsi_execute_async failed\n"));
-               /*
-                * most likely out of mem, but could also be a bad map
-                */
-               sg_finish_rem_req(srp);
-               return -ENOMEM;
-       } else
-               return 0;
+
+       srp->rq->timeout = timeout;
+       blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
+                             srp->rq, 1, sg_rq_end_io);
+       return 0;
 }
 
 static int
@@ -1206,8 +1187,7 @@ sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
        Sg_fd *sfp;
        unsigned long offset, len, sa;
        Sg_scatter_hold *rsv_schp;
-       struct scatterlist *sg;
-       int k;
+       int k, length;
 
        if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
                return VM_FAULT_SIGBUS;
@@ -1217,15 +1197,14 @@ sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
                return VM_FAULT_SIGBUS;
        SCSI_LOG_TIMEOUT(3, printk("sg_vma_fault: offset=%lu, scatg=%d\n",
                                   offset, rsv_schp->k_use_sg));
-       sg = rsv_schp->buffer;
        sa = vma->vm_start;
-       for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
-            ++k, sg = sg_next(sg)) {
+       length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
+       for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
                len = vma->vm_end - sa;
-               len = (len < sg->length) ? len : sg->length;
+               len = (len < length) ? len : length;
                if (offset < len) {
-                       struct page *page;
-                       page = virt_to_page(page_address(sg_page(sg)) + offset);
+                       struct page *page = nth_page(rsv_schp->pages[k],
+                                                    offset >> PAGE_SHIFT);
                        get_page(page); /* increment page count */
                        vmf->page = page;
                        return 0; /* success */
@@ -1247,8 +1226,7 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma)
        Sg_fd *sfp;
        unsigned long req_sz, len, sa;
        Sg_scatter_hold *rsv_schp;
-       int k;
-       struct scatterlist *sg;
+       int k, length;
 
        if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
                return -ENXIO;
@@ -1262,11 +1240,10 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma)
                return -ENOMEM; /* cannot map more than reserved buffer */
 
        sa = vma->vm_start;
-       sg = rsv_schp->buffer;
-       for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
-            ++k, sg = sg_next(sg)) {
+       length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
+       for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
                len = vma->vm_end - sa;
-               len = (len < sg->length) ? len : sg->length;
+               len = (len < length) ? len : length;
                sa += len;
        }
 
@@ -1310,7 +1287,6 @@ sg_cmd_done(void *data, char *sense, int result, int resid)
        if (0 != result) {
                struct scsi_sense_hdr sshdr;
 
-               memcpy(srp->sense_b, sense, sizeof (srp->sense_b));
                srp->header.status = 0xff & result;
                srp->header.masked_status = status_byte(result);
                srp->header.msg_status = msg_byte(result);
@@ -1685,34 +1661,51 @@ static int sg_start_req(Sg_request *srp, unsigned char *cmd)
        Sg_scatter_hold *rsv_schp = &sfp->reserve;
        struct request_queue *q = sfp->parentdp->device->request_queue;
        unsigned long alignment = queue_dma_alignment(q) | q->dma_pad_mask;
+       struct rq_map_data map_data;
 
        SCSI_LOG_TIMEOUT(4, printk("sg_start_req: dxfer_len=%d\n", dxfer_len));
 
+       res = __sg_start_req(srp, hp, cmd);
+       if (res)
+               return res;
+
        if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
-               return __sg_start_req(srp, hp, cmd);
+               return 0;
 
 #ifdef SG_ALLOW_DIO_CODE
        if (sg_allow_dio && (hp->flags & SG_FLAG_DIRECT_IO) &&
            (dxfer_dir != SG_DXFER_UNKNOWN) && (0 == hp->iovec_count) &&
            (!sfp->parentdp->device->host->unchecked_isa_dma) &&
-           !(uaddr & alignment) && !(dxfer_len & alignment)) {
-               res = __sg_start_req(srp, hp, cmd);
-               if (!res)
-                       res = sg_build_direct(srp, sfp, dxfer_len);
-
-               return res;
-       }
+           !(uaddr & alignment) && !(dxfer_len & alignment))
+               return sg_build_direct(srp, sfp, dxfer_len);
 #endif
        if ((!sg_res_in_use(sfp)) && (dxfer_len <= rsv_schp->bufflen))
                sg_link_reserve(sfp, srp, dxfer_len);
-       else {
+       else
                res = sg_build_indirect(req_schp, sfp, dxfer_len);
-               if (res) {
-                       sg_remove_scat(req_schp);
-                       return res;
-               }
+
+       if (!res) {
+               struct request *rq = srp->rq;
+               Sg_scatter_hold *schp = &srp->data;
+               int iovec_count = (int) hp->iovec_count;
+
+               map_data.pages = schp->pages;
+               map_data.page_order = schp->page_order;
+               map_data.nr_entries = schp->k_use_sg;
+
+               if (iovec_count)
+                       res = blk_rq_map_user_iov(q, rq, &map_data, hp->dxferp,
+                                                 iovec_count,
+                                                 hp->dxfer_len, GFP_ATOMIC);
+               else
+                       res = blk_rq_map_user(q, rq, &map_data, hp->dxferp,
+                                             hp->dxfer_len, GFP_ATOMIC);
+
+               if (!res)
+                       srp->bio = rq->bio;
        }
-       return 0;
+
+       return res;
 }
 
 static void
@@ -1730,6 +1723,7 @@ sg_finish_rem_req(Sg_request * srp)
        if (srp->rq) {
                if (srp->bio)
                        blk_rq_unmap_user(srp->bio);
+
                blk_put_request(srp->rq);
        }
 
@@ -1739,21 +1733,12 @@ sg_finish_rem_req(Sg_request * srp)
 static int
 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
 {
-       int sg_bufflen = tablesize * sizeof(struct scatterlist);
+       int sg_bufflen = tablesize * sizeof(struct page *);
        gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
 
-       /*
-        * TODO: test without low_dma, we should not need it since
-        * the block layer will bounce the buffer for us
-        *
-        * XXX(hch): we shouldn't need GFP_DMA for the actual S/G list.
-        */
-       if (sfp->low_dma)
-                gfp_flags |= GFP_DMA;
-       schp->buffer = kzalloc(sg_bufflen, gfp_flags);
-       if (!schp->buffer)
+       schp->pages = kzalloc(sg_bufflen, gfp_flags);
+       if (!schp->pages)
                return -ENOMEM;
-       sg_init_table(schp->buffer, tablesize);
        schp->sglist_len = sg_bufflen;
        return tablesize;       /* number of scat_gath elements allocated */
 }
@@ -1780,11 +1765,10 @@ sg_build_direct(Sg_request * srp, Sg_fd * sfp, int dxfer_len)
 static int
 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
 {
-       struct scatterlist *sg;
-       int ret_sz = 0, k, rem_sz, num, mx_sc_elems;
+       int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
        int sg_tablesize = sfp->parentdp->sg_tablesize;
-       int blk_size = buff_size;
-       struct page *p = NULL;
+       int blk_size = buff_size, order;
+       gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
 
        if (blk_size < 0)
                return -EFAULT;
@@ -1808,15 +1792,26 @@ sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
                } else
                        scatter_elem_sz_prev = num;
        }
-       for (k = 0, sg = schp->buffer, rem_sz = blk_size;
-            (rem_sz > 0) && (k < mx_sc_elems);
-            ++k, rem_sz -= ret_sz, sg = sg_next(sg)) {
-               
+
+       if (sfp->low_dma)
+               gfp_mask |= GFP_DMA;
+
+       if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
+               gfp_mask |= __GFP_ZERO;
+
+       order = get_order(num);
+retry:
+       ret_sz = 1 << (PAGE_SHIFT + order);
+
+       for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
+            k++, rem_sz -= ret_sz) {
+
                num = (rem_sz > scatter_elem_sz_prev) ?
-                     scatter_elem_sz_prev : rem_sz;
-               p = sg_page_malloc(num, sfp->low_dma, &ret_sz);
-               if (!p)
-                       return -ENOMEM;
+                       scatter_elem_sz_prev : rem_sz;
+
+               schp->pages[k] = alloc_pages(gfp_mask, order);
+               if (!schp->pages[k])
+                       goto out;
 
                if (num == scatter_elem_sz_prev) {
                        if (unlikely(ret_sz > scatter_elem_sz_prev)) {
@@ -1824,12 +1819,12 @@ sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
                                scatter_elem_sz_prev = ret_sz;
                        }
                }
-               sg_set_page(sg, p, (ret_sz > num) ? num : ret_sz, 0);
 
                SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k=%d, num=%d, "
                                 "ret_sz=%d\n", k, num, ret_sz));
        }               /* end of for loop */
 
+       schp->page_order = order;
        schp->k_use_sg = k;
        SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k_use_sg=%d, "
                         "rem_sz=%d\n", k, rem_sz));
@@ -1837,8 +1832,15 @@ sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
        schp->bufflen = blk_size;
        if (rem_sz > 0) /* must have failed */
                return -ENOMEM;
-
        return 0;
+out:
+       for (i = 0; i < k; i++)
+               __free_pages(schp->pages[k], order);
+
+       if (--order >= 0)
+               goto retry;
+
+       return -ENOMEM;
 }
 
 static int
@@ -1846,13 +1848,8 @@ sg_write_xfer(Sg_request * srp)
 {
        sg_io_hdr_t *hp = &srp->header;
        Sg_scatter_hold *schp = &srp->data;
-       struct scatterlist *sg = schp->buffer;
        int num_xfer = 0;
-       int j, k, onum, usglen, ksglen, res;
-       int iovec_count = (int) hp->iovec_count;
        int dxfer_dir = hp->dxfer_direction;
-       unsigned char *p;
-       unsigned char __user *up;
        int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
 
        if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_TO_DEV == dxfer_dir) ||
@@ -1866,83 +1863,9 @@ sg_write_xfer(Sg_request * srp)
             && ((SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO) & hp->flags)))
                return 0;
 
-       SCSI_LOG_TIMEOUT(4, printk("sg_write_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
-                         num_xfer, iovec_count, schp->k_use_sg));
-       if (iovec_count) {
-               onum = iovec_count;
-               if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
-                       return -EFAULT;
-       } else
-               onum = 1;
-
-       ksglen = sg->length;
-       p = page_address(sg_page(sg));
-       for (j = 0, k = 0; j < onum; ++j) {
-               res = sg_u_iovec(hp, iovec_count, j, 1, &usglen, &up);
-               if (res)
-                       return res;
-
-               for (; p; sg = sg_next(sg), ksglen = sg->length,
-                    p = page_address(sg_page(sg))) {
-                       if (usglen <= 0)
-                               break;
-                       if (ksglen > usglen) {
-                               if (usglen >= num_xfer) {
-                                       if (__copy_from_user(p, up, num_xfer))
-                                               return -EFAULT;
-                                       return 0;
-                               }
-                               if (__copy_from_user(p, up, usglen))
-                                       return -EFAULT;
-                               p += usglen;
-                               ksglen -= usglen;
-                               break;
-                       } else {
-                               if (ksglen >= num_xfer) {
-                                       if (__copy_from_user(p, up, num_xfer))
-                                               return -EFAULT;
-                                       return 0;
-                               }
-                               if (__copy_from_user(p, up, ksglen))
-                                       return -EFAULT;
-                               up += ksglen;
-                               usglen -= ksglen;
-                       }
-                       ++k;
-                       if (k >= schp->k_use_sg)
-                               return 0;
-               }
-       }
-
-       return 0;
-}
+       SCSI_LOG_TIMEOUT(4, printk("sg_write_xfer: num_xfer=%d, k_use_sg=%d\n",
+                         num_xfer, schp->k_use_sg));
 
-static int
-sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
-          int wr_xf, int *countp, unsigned char __user **up)
-{
-       int num_xfer = (int) hp->dxfer_len;
-       unsigned char __user *p = hp->dxferp;
-       int count;
-
-       if (0 == sg_num) {
-               if (wr_xf && ('\0' == hp->interface_id))
-                       count = (int) hp->flags;        /* holds "old" input_size */
-               else
-                       count = num_xfer;
-       } else {
-               sg_iovec_t iovec;
-               if (__copy_from_user(&iovec, p + ind*SZ_SG_IOVEC, SZ_SG_IOVEC))
-                       return -EFAULT;
-               p = iovec.iov_base;
-               count = (int) iovec.iov_len;
-       }
-       if (!access_ok(wr_xf ? VERIFY_READ : VERIFY_WRITE, p, count))
-               return -EFAULT;
-       if (up)
-               *up = p;
-       if (countp)
-               *countp = count;
        return 0;
 }
 
@@ -1950,21 +1873,18 @@ static void
 sg_remove_scat(Sg_scatter_hold * schp)
 {
        SCSI_LOG_TIMEOUT(4, printk("sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
-       if (schp->buffer && (schp->sglist_len > 0)) {
-               struct scatterlist *sg = schp->buffer;
-
+       if (schp->pages && schp->sglist_len > 0) {
                if (!schp->dio_in_use) {
                        int k;
 
-                       for (k = 0; (k < schp->k_use_sg) && sg_page(sg);
-                            ++k, sg = sg_next(sg)) {
+                       for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
                                SCSI_LOG_TIMEOUT(5, printk(
-                                   "sg_remove_scat: k=%d, pg=0x%p, len=%d\n",
-                                   k, sg_page(sg), sg->length));
-                               sg_page_free(sg_page(sg), sg->length);
+                                   "sg_remove_scat: k=%d, pg=0x%p\n",
+                                   k, schp->pages[k]));
+                               __free_pages(schp->pages[k], schp->page_order);
                        }
 
-                       kfree(schp->buffer);
+                       kfree(schp->pages);
                }
        }
        memset(schp, 0, sizeof (*schp));
@@ -1975,13 +1895,8 @@ sg_read_xfer(Sg_request * srp)
 {
        sg_io_hdr_t *hp = &srp->header;
        Sg_scatter_hold *schp = &srp->data;
-       struct scatterlist *sg = schp->buffer;
        int num_xfer = 0;
-       int j, k, onum, usglen, ksglen, res;
-       int iovec_count = (int) hp->iovec_count;
        int dxfer_dir = hp->dxfer_direction;
-       unsigned char *p;
-       unsigned char __user *up;
        int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
 
        if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_FROM_DEV == dxfer_dir)
@@ -1996,53 +1911,7 @@ sg_read_xfer(Sg_request * srp)
                return 0;
 
        SCSI_LOG_TIMEOUT(4, printk("sg_read_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
-                         num_xfer, iovec_count, schp->k_use_sg));
-       if (iovec_count) {
-               onum = iovec_count;
-               if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
-                       return -EFAULT;
-       } else
-               onum = 1;
-
-       p = page_address(sg_page(sg));
-       ksglen = sg->length;
-       for (j = 0, k = 0; j < onum; ++j) {
-               res = sg_u_iovec(hp, iovec_count, j, 0, &usglen, &up);
-               if (res)
-                       return res;
-
-               for (; p; sg = sg_next(sg), ksglen = sg->length,
-                    p = page_address(sg_page(sg))) {
-                       if (usglen <= 0)
-                               break;
-                       if (ksglen > usglen) {
-                               if (usglen >= num_xfer) {
-                                       if (__copy_to_user(up, p, num_xfer))
-                                               return -EFAULT;
-                                       return 0;
-                               }
-                               if (__copy_to_user(up, p, usglen))
-                                       return -EFAULT;
-                               p += usglen;
-                               ksglen -= usglen;
-                               break;
-                       } else {
-                               if (ksglen >= num_xfer) {
-                                       if (__copy_to_user(up, p, num_xfer))
-                                               return -EFAULT;
-                                       return 0;
-                               }
-                               if (__copy_to_user(up, p, ksglen))
-                                       return -EFAULT;
-                               up += ksglen;
-                               usglen -= ksglen;
-                       }
-                       ++k;
-                       if (k >= schp->k_use_sg)
-                               return 0;
-               }
-       }
-
+                         num_xfer, (int)hp->iovec_count, schp->k_use_sg));
        return 0;
 }
 
@@ -2050,7 +1919,6 @@ static int
 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
 {
        Sg_scatter_hold *schp = &srp->data;
-       struct scatterlist *sg = schp->buffer;
        int k, num;
 
        SCSI_LOG_TIMEOUT(4, printk("sg_read_oxfer: num_read_xfer=%d\n",
@@ -2058,15 +1926,18 @@ sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
        if ((!outp) || (num_read_xfer <= 0))
                return 0;
 
-       for (k = 0; (k < schp->k_use_sg) && sg_page(sg); ++k, sg = sg_next(sg)) {
-               num = sg->length;
+       blk_rq_unmap_user(srp->bio);
+       srp->bio = NULL;
+
+       num = 1 << (PAGE_SHIFT + schp->page_order);
+       for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
                if (num > num_read_xfer) {
-                       if (__copy_to_user(outp, page_address(sg_page(sg)),
+                       if (__copy_to_user(outp, page_address(schp->pages[k]),
                                           num_read_xfer))
                                return -EFAULT;
                        break;
                } else {
-                       if (__copy_to_user(outp, page_address(sg_page(sg)),
+                       if (__copy_to_user(outp, page_address(schp->pages[k]),
                                           num))
                                return -EFAULT;
                        num_read_xfer -= num;
@@ -2101,24 +1972,22 @@ sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
 {
        Sg_scatter_hold *req_schp = &srp->data;
        Sg_scatter_hold *rsv_schp = &sfp->reserve;
-       struct scatterlist *sg = rsv_schp->buffer;
        int k, num, rem;
 
        srp->res_used = 1;
        SCSI_LOG_TIMEOUT(4, printk("sg_link_reserve: size=%d\n", size));
        rem = size;
 
-       for (k = 0; k < rsv_schp->k_use_sg; ++k, sg = sg_next(sg)) {
-               num = sg->length;
+       num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
+       for (k = 0; k < rsv_schp->k_use_sg; k++) {
                if (rem <= num) {
-                       sfp->save_scat_len = num;
-                       sg->length = rem;
                        req_schp->k_use_sg = k + 1;
                        req_schp->sglist_len = rsv_schp->sglist_len;
-                       req_schp->buffer = rsv_schp->buffer;
+                       req_schp->pages = rsv_schp->pages;
 
                        req_schp->bufflen = size;
                        req_schp->b_malloc_len = rsv_schp->b_malloc_len;
+                       req_schp->page_order = rsv_schp->page_order;
                        break;
                } else
                        rem -= num;
@@ -2132,22 +2001,13 @@ static void
 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
 {
        Sg_scatter_hold *req_schp = &srp->data;
-       Sg_scatter_hold *rsv_schp = &sfp->reserve;
 
        SCSI_LOG_TIMEOUT(4, printk("sg_unlink_reserve: req->k_use_sg=%d\n",
                                   (int) req_schp->k_use_sg));
-       if ((rsv_schp->k_use_sg > 0) && (req_schp->k_use_sg > 0)) {
-               struct scatterlist *sg = rsv_schp->buffer;
-
-               if (sfp->save_scat_len > 0)
-                       (sg + (req_schp->k_use_sg - 1))->length =
-                           (unsigned) sfp->save_scat_len;
-               else
-                       SCSI_LOG_TIMEOUT(1, printk ("sg_unlink_reserve: BAD save_scat_len\n"));
-       }
        req_schp->k_use_sg = 0;
        req_schp->bufflen = 0;
-       req_schp->buffer = NULL;
+       req_schp->pages = NULL;
+       req_schp->page_order = 0;
        req_schp->sglist_len = 0;
        sfp->save_scat_len = 0;
        srp->res_used = 0;
@@ -2405,53 +2265,6 @@ sg_res_in_use(Sg_fd * sfp)
        return srp ? 1 : 0;
 }
 
-/* The size fetched (value output via retSzp) set when non-NULL return */
-static struct page *
-sg_page_malloc(int rqSz, int lowDma, int *retSzp)
-{
-       struct page *resp = NULL;
-       gfp_t page_mask;
-       int order, a_size;
-       int resSz;
-
-       if ((rqSz <= 0) || (NULL == retSzp))
-               return resp;
-
-       if (lowDma)
-               page_mask = GFP_ATOMIC | GFP_DMA | __GFP_COMP | __GFP_NOWARN;
-       else
-               page_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
-
-       for (order = 0, a_size = PAGE_SIZE; a_size < rqSz;
-            order++, a_size <<= 1) ;
-       resSz = a_size;         /* rounded up if necessary */
-       resp = alloc_pages(page_mask, order);
-       while ((!resp) && order) {
-               --order;
-               a_size >>= 1;   /* divide by 2, until PAGE_SIZE */
-               resp =  alloc_pages(page_mask, order);  /* try half */
-               resSz = a_size;
-       }
-       if (resp) {
-               if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
-                       memset(page_address(resp), 0, resSz);
-               *retSzp = resSz;
-       }
-       return resp;
-}
-
-static void
-sg_page_free(struct page *page, int size)
-{
-       int order, a_size;
-
-       if (!page)
-               return;
-       for (order = 0, a_size = PAGE_SIZE; a_size < size;
-            order++, a_size <<= 1) ;
-       __free_pages(page, order);
-}
-
 #ifdef CONFIG_SCSI_PROC_FS
 static int
 sg_idr_max_id(int id, void *p, void *data)