OSDN Git Service

exofs: avoid VLA in structures
authorKees Cook <keescook@chromium.org>
Thu, 14 Jun 2018 22:27:27 +0000 (15:27 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 14 Jun 2018 22:55:24 +0000 (07:55 +0900)
On the quest to remove all VLAs from the kernel[1] this adjusts several
cases where allocation is made after an array of structures that points
back into the allocation.  The allocations are changed to perform
explicit calculations instead of using a Variable Length Array in a
structure.

Additionally, this lets Clang compile this code now, since Clang does
not support VLAIS[2].

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com
[2] https://lkml.kernel.org/r/CA+55aFy6h1c3_rP_bXFedsTXzwW+9Q9MfJaW7GUmMBrAp-fJ9A@mail.gmail.com

[keescook@chromium.org: v2]
Link: http://lkml.kernel.org/r/20180418163546.GA45794@beast
Link: http://lkml.kernel.org/r/20180327203904.GA1151@beast
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Cc: Boaz Harrosh <ooo@electrozaur.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/exofs/ore.c
fs/exofs/ore_raid.c
fs/exofs/super.c

index ddbf872..1b8b446 100644 (file)
@@ -146,68 +146,82 @@ int  _ore_get_io_state(struct ore_layout *layout,
                        struct ore_io_state **pios)
 {
        struct ore_io_state *ios;
-       struct page **pages;
-       struct osd_sg_entry *sgilist;
+       size_t size_ios, size_extra, size_total;
+       void *ios_extra;
+
+       /*
+        * The desired layout looks like this, with the extra_allocation
+        * items pointed at from fields within ios or per_dev:
+
        struct __alloc_all_io_state {
                struct ore_io_state ios;
                struct ore_per_dev_state per_dev[numdevs];
                union {
                        struct osd_sg_entry sglist[sgs_per_dev * numdevs];
                        struct page *pages[num_par_pages];
-               };
-       } *_aios;
-
-       if (likely(sizeof(*_aios) <= PAGE_SIZE)) {
-               _aios = kzalloc(sizeof(*_aios), GFP_KERNEL);
-               if (unlikely(!_aios)) {
-                       ORE_DBGMSG("Failed kzalloc bytes=%zd\n",
-                                  sizeof(*_aios));
+               } extra_allocation;
+       } whole_allocation;
+
+       */
+
+       /* This should never happen, so abort early if it ever does. */
+       if (sgs_per_dev && num_par_pages) {
+               ORE_DBGMSG("Tried to use both pages and sglist\n");
+               *pios = NULL;
+               return -EINVAL;
+       }
+
+       if (numdevs > (INT_MAX - sizeof(*ios)) /
+                      sizeof(struct ore_per_dev_state))
+               return -ENOMEM;
+       size_ios = sizeof(*ios) + sizeof(struct ore_per_dev_state) * numdevs;
+
+       if (sgs_per_dev * numdevs > INT_MAX / sizeof(struct osd_sg_entry))
+               return -ENOMEM;
+       if (num_par_pages > INT_MAX / sizeof(struct page *))
+               return -ENOMEM;
+       size_extra = max(sizeof(struct osd_sg_entry) * (sgs_per_dev * numdevs),
+                        sizeof(struct page *) * num_par_pages);
+
+       size_total = size_ios + size_extra;
+
+       if (likely(size_total <= PAGE_SIZE)) {
+               ios = kzalloc(size_total, GFP_KERNEL);
+               if (unlikely(!ios)) {
+                       ORE_DBGMSG("Failed kzalloc bytes=%zd\n", size_total);
                        *pios = NULL;
                        return -ENOMEM;
                }
-               pages = num_par_pages ? _aios->pages : NULL;
-               sgilist = sgs_per_dev ? _aios->sglist : NULL;
-               ios = &_aios->ios;
+               ios_extra = (char *)ios + size_ios;
        } else {
-               struct __alloc_small_io_state {
-                       struct ore_io_state ios;
-                       struct ore_per_dev_state per_dev[numdevs];
-               } *_aio_small;
-               union __extra_part {
-                       struct osd_sg_entry sglist[sgs_per_dev * numdevs];
-                       struct page *pages[num_par_pages];
-               } *extra_part;
-
-               _aio_small = kzalloc(sizeof(*_aio_small), GFP_KERNEL);
-               if (unlikely(!_aio_small)) {
+               ios = kzalloc(size_ios, GFP_KERNEL);
+               if (unlikely(!ios)) {
                        ORE_DBGMSG("Failed alloc first part bytes=%zd\n",
-                                  sizeof(*_aio_small));
+                                  size_ios);
                        *pios = NULL;
                        return -ENOMEM;
                }
-               extra_part = kzalloc(sizeof(*extra_part), GFP_KERNEL);
-               if (unlikely(!extra_part)) {
+               ios_extra = kzalloc(size_extra, GFP_KERNEL);
+               if (unlikely(!ios_extra)) {
                        ORE_DBGMSG("Failed alloc second part bytes=%zd\n",
-                                  sizeof(*extra_part));
-                       kfree(_aio_small);
+                                  size_extra);
+                       kfree(ios);
                        *pios = NULL;
                        return -ENOMEM;
                }
 
-               pages = num_par_pages ? extra_part->pages : NULL;
-               sgilist = sgs_per_dev ? extra_part->sglist : NULL;
                /* In this case the per_dev[0].sgilist holds the pointer to
                 * be freed
                 */
-               ios = &_aio_small->ios;
                ios->extra_part_alloc = true;
        }
 
-       if (pages) {
-               ios->parity_pages = pages;
+       if (num_par_pages) {
+               ios->parity_pages = ios_extra;
                ios->max_par_pages = num_par_pages;
        }
-       if (sgilist) {
+       if (sgs_per_dev) {
+               struct osd_sg_entry *sgilist = ios_extra;
                unsigned d;
 
                for (d = 0; d < numdevs; ++d) {
index 27cbdb6..199590f 100644 (file)
@@ -71,6 +71,11 @@ static int _sp2d_alloc(unsigned pages_in_unit, unsigned group_width,
 {
        struct __stripe_pages_2d *sp2d;
        unsigned data_devs = group_width - parity;
+
+       /*
+        * Desired allocation layout is, though when larger than PAGE_SIZE,
+        * each struct __alloc_1p_arrays is separately allocated:
+
        struct _alloc_all_bytes {
                struct __alloc_stripe_pages_2d {
                        struct __stripe_pages_2d sp2d;
@@ -82,55 +87,85 @@ static int _sp2d_alloc(unsigned pages_in_unit, unsigned group_width,
                        char page_is_read[data_devs];
                } __a1pa[pages_in_unit];
        } *_aab;
+
        struct __alloc_1p_arrays *__a1pa;
        struct __alloc_1p_arrays *__a1pa_end;
-       const unsigned sizeof__a1pa = sizeof(_aab->__a1pa[0]);
+
+       */
+
+       char *__a1pa;
+       char *__a1pa_end;
+
+       const size_t sizeof_stripe_pages_2d =
+               sizeof(struct __stripe_pages_2d) +
+               sizeof(struct __1_page_stripe) * pages_in_unit;
+       const size_t sizeof__a1pa =
+               ALIGN(sizeof(struct page *) * (2 * group_width) + data_devs,
+                     sizeof(void *));
+       const size_t sizeof__a1pa_arrays = sizeof__a1pa * pages_in_unit;
+       const size_t alloc_total = sizeof_stripe_pages_2d +
+                                  sizeof__a1pa_arrays;
+
        unsigned num_a1pa, alloc_size, i;
 
        /* FIXME: check these numbers in ore_verify_layout */
-       BUG_ON(sizeof(_aab->__asp2d) > PAGE_SIZE);
+       BUG_ON(sizeof_stripe_pages_2d > PAGE_SIZE);
        BUG_ON(sizeof__a1pa > PAGE_SIZE);
 
-       if (sizeof(*_aab) > PAGE_SIZE) {
-               num_a1pa = (PAGE_SIZE - sizeof(_aab->__asp2d)) / sizeof__a1pa;
-               alloc_size = sizeof(_aab->__asp2d) + sizeof__a1pa * num_a1pa;
+       /*
+        * If alloc_total would be larger than PAGE_SIZE, only allocate
+        * as many a1pa items as would fill the rest of the page, instead
+        * of the full pages_in_unit count.
+        */
+       if (alloc_total > PAGE_SIZE) {
+               num_a1pa = (PAGE_SIZE - sizeof_stripe_pages_2d) / sizeof__a1pa;
+               alloc_size = sizeof_stripe_pages_2d + sizeof__a1pa * num_a1pa;
        } else {
                num_a1pa = pages_in_unit;
-               alloc_size = sizeof(*_aab);
+               alloc_size = alloc_total;
        }
 
-       _aab = kzalloc(alloc_size, GFP_KERNEL);
-       if (unlikely(!_aab)) {
+       *psp2d = sp2d = kzalloc(alloc_size, GFP_KERNEL);
+       if (unlikely(!sp2d)) {
                ORE_DBGMSG("!! Failed to alloc sp2d size=%d\n", alloc_size);
                return -ENOMEM;
        }
+       /* From here Just call _sp2d_free */
 
-       sp2d = &_aab->__asp2d.sp2d;
-       *psp2d = sp2d; /* From here Just call _sp2d_free */
-
-       __a1pa = _aab->__a1pa;
-       __a1pa_end = __a1pa + num_a1pa;
+       /* Find start of a1pa area. */
+       __a1pa = (char *)sp2d + sizeof_stripe_pages_2d;
+       /* Find end of the _allocated_ a1pa area. */
+       __a1pa_end = __a1pa + alloc_size;
 
+       /* Allocate additionally needed a1pa items in PAGE_SIZE chunks. */
        for (i = 0; i < pages_in_unit; ++i) {
+               struct __1_page_stripe *stripe = &sp2d->_1p_stripes[i];
+
                if (unlikely(__a1pa >= __a1pa_end)) {
                        num_a1pa = min_t(unsigned, PAGE_SIZE / sizeof__a1pa,
                                                        pages_in_unit - i);
+                       alloc_size = sizeof__a1pa * num_a1pa;
 
-                       __a1pa = kcalloc(num_a1pa, sizeof__a1pa, GFP_KERNEL);
+                       __a1pa = kzalloc(alloc_size, GFP_KERNEL);
                        if (unlikely(!__a1pa)) {
                                ORE_DBGMSG("!! Failed to _alloc_1p_arrays=%d\n",
                                           num_a1pa);
                                return -ENOMEM;
                        }
-                       __a1pa_end = __a1pa + num_a1pa;
+                       __a1pa_end = __a1pa + alloc_size;
                        /* First *pages is marked for kfree of the buffer */
-                       sp2d->_1p_stripes[i].alloc = true;
+                       stripe->alloc = true;
                }
 
-               sp2d->_1p_stripes[i].pages = __a1pa->pages;
-               sp2d->_1p_stripes[i].scribble = __a1pa->scribble ;
-               sp2d->_1p_stripes[i].page_is_read = __a1pa->page_is_read;
-               ++__a1pa;
+               /*
+                * Attach all _lp_stripes pointers to the allocation for
+                * it which was either part of the original PAGE_SIZE
+                * allocation or the subsequent allocation in this loop.
+                */
+               stripe->pages = (void *)__a1pa;
+               stripe->scribble = stripe->pages + group_width;
+               stripe->page_is_read = (char *)stripe->scribble + group_width;
+               __a1pa += sizeof__a1pa;
        }
 
        sp2d->parity = parity;
index 719a315..41cf2fb 100644 (file)
@@ -549,27 +549,26 @@ static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
 static int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
                      struct exofs_dev **peds)
 {
-       struct __alloc_ore_devs_and_exofs_devs {
-               /* Twice bigger table: See exofs_init_comps() and comment at
-                * exofs_read_lookup_dev_table()
-                */
-               struct ore_dev *oreds[numdevs * 2 - 1];
-               struct exofs_dev eds[numdevs];
-       } *aoded;
+       /* Twice bigger table: See exofs_init_comps() and comment at
+        * exofs_read_lookup_dev_table()
+        */
+       const size_t numores = numdevs * 2 - 1;
        struct exofs_dev *eds;
        unsigned i;
 
-       aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
-       if (unlikely(!aoded)) {
+       sbi->oc.ods = kzalloc(numores * sizeof(struct ore_dev *) +
+                             numdevs * sizeof(struct exofs_dev), GFP_KERNEL);
+       if (unlikely(!sbi->oc.ods)) {
                EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
                          numdevs);
                return -ENOMEM;
        }
 
-       sbi->oc.ods = aoded->oreds;
-       *peds = eds = aoded->eds;
+       /* Start of allocated struct exofs_dev entries */
+       *peds = eds = (void *)sbi->oc.ods[numores];
+       /* Initialize pointers into struct exofs_dev */
        for (i = 0; i < numdevs; ++i)
-               aoded->oreds[i] = &eds[i].ored;
+               sbi->oc.ods[i] = &eds[i].ored;
        return 0;
 }